Capsule Networks (CapsNets) have been proposed as an alternative to Convolutional Neural Networks
(CNNs). This paper showcases how CapsNets are more capable than CNNs for autonomous agent
exploration of realistic scenarios. In real world navigation, rewards external to agents may be rare. In
turn, reinforcement learning algorithms can struggle to form meaningful policy functions. This paper’s
approach Capsules Exploration Module (Caps-EM) pairs a CapsNets architecture with an Advantage
Actor Critic algorithm. Other approaches for navigating sparse environments require intrinsic reward
generators, such as the Intrinsic Curiosity Module (ICM) and Augmented Curiosity Modules (ACM). CapsEM uses a more compact architecture without need for intrinsic rewards. Tested using ViZDoom, the CapsEM uses 44% and 83% fewer trainable network parameters than the ICM and Depth-Augmented Curiosity
Module (D-ACM), respectively, for 1141% and 437% average time improvement over the ICM and DACM, respectively, for converging to a policy function across "My Way Home" scenarios.
CAPSULE NETWORK PERFORMANCE WITH AUTONOMOUS NAVIGATIONgerogepatton
Capsule Networks (CapsNets) have been proposed as an alternative to Convolutional Neural Networks (CNNs). This paper showcases how CapsNets are more capable than CNNs for autonomous agent exploration of realistic scenarios. In real world navigation, rewards external to agents may be rare. In turn, reinforcement learning algorithms can struggle to form meaningful policy functions. This paper’s approach Capsules Exploration Module (Caps-EM) pairs a CapsNets architecture with an Advantage Actor Critic algorithm. Other approaches for navigating sparse environments require intrinsic reward generators, such as the Intrinsic Curiosity Module (ICM) and Augmented Curiosity Modules (ACM). CapsEM uses a more compact architecture without need for intrinsic rewards. Tested using ViZDoom, the CapsEM uses 44% and 83% fewer trainable network parameters than the ICM and Depth-Augmented Curiosity Module (D-ACM), respectively, for 1141% and 437% average time improvement over the ICM and DACM, respectively, for converging to a policy function across "My Way Home" scenarios
Optimization of Number of Neurons in the Hidden Layer in Feed Forward Neural ...IJERA Editor
The architectures of Artificial Neural Networks (ANN) are based on the problem domain and it is applied during
the „training phase‟ of sample data and used to infer results for the remaining data in the testing phase.
Normally, the architecture consist of three layers as input, hidden, output layers with the number of nodes in the
input layer as number of known values on hand and the number of nodes as result to be computed out of the
values of input nodes and hidden nodes as the output layer. The number of nodes in the hidden layer is
heuristically decided so that the optimum value is obtained with reasonable number of iterations with other
parameters with its default values. This study mainly focuses on Cascade-Correlation Neural Networks (CCNN)
using Back-Propagation (BP) algorithm which finds the number of neurons during the training phase itself by
appending one from the previous iteration satisfying the error condition gives a promising result on the optimum
number of neurons in the hidden layer
This document summarizes a study that used artificial neural networks (ANNs) and the Multi-Layer Perceptron model (MLP) to predict the bearing capacities of steel driven piles in sandy soils. The ANN was trained on data from full-scale pile load tests, including pile length, diameter, soil elastic modulus, and soil friction angle as inputs. The output was pile bearing capacity. The study examined factors for effective ANN behavior, trained and tested the network, and analyzed the sensitivity of the inputs on the output capacity prediction.
This document provides an overview of self-organizing maps (SOM) as an unsupervised learning technique. It discusses the principles of self-organization including self-amplification, competition, and cooperation. The Willshaw-von der Malsburg model and Kohonen feature maps are presented as two approaches to building topographic maps through self-organization. The Kohonen SOM learning algorithm is described as involving competition between neurons to determine a winning neuron, cooperation between neighboring neurons, and adaptive changes to synaptic weights based on Hebbian learning principles.
Mobile Network Coverage Determination at 900MHz for Abuja Rural Areas using A...ijtsrd
This study proposes Artificial Neural Network ANN based field strength prediction models for the rural areas of Abuja, the federal capital territory of Nigeria. The ANN based models were created on bases of the Generalized Regression Neural network GRNN and the Multi Layer Perceptron Neural Network MLP NN . These networks were created, trained and tested for field strength prediction using received power data recorded at 900MHz from multiple Base Transceiver Stations BTSs distributed across the rural areas. Results indicate that the GRNN and MLP NN based models with Root Mean Squared Error RMSE values of 4.78dBm and 5.56dBm respectively, offer significant improvement over the empirical Hata Okumura counterpart, which overestimates the signal strength by an RMSE value of 20.17dBm. Deme C. Abraham ""Mobile Network Coverage Determination at 900MHz for Abuja Rural Areas using Artificial Neural Networks"" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-4 | Issue-2 , February 2020,
URL: https://www.ijtsrd.com/papers/ijtsrd30228.pdf
Paper Url : https://www.ijtsrd.com/computer-science/artificial-intelligence/30228/mobile-network-coverage-determination-at-900mhz-for-abuja-rural-areas-using-artificial-neural-networks/deme-c-abraham
Modeling of neural image compression using gradient decent technologytheijes
The International Journal of Engineering & Science is aimed at providing a platform for researchers, engineers, scientists, or educators to publish their original research results, to exchange new ideas, to disseminate information in innovative designs, engineering experiences and technological skills. It is also the Journal's objective to promote engineering and technology education. All papers submitted to the Journal will be blind peer-reviewed. Only original articles will be published.
The papers for publication in The International Journal of Engineering& Science are selected through rigorous peer reviews to ensure originality, timeliness, relevance, and readability.
Theoretical work submitted to the Journal should be original in its motivation or modeling structure. Empirical analysis should be based on a theoretical framework and should be capable of replication. It is expected that all materials required for replication (including computer programs and data sets) should be available upon request to the authors.
The International Journal of Engineering & Science would take much care in making your article published without much delay with your kind cooperation
An efficient technique for color image classification based on lower feature ...Alexander Decker
This document discusses an efficient technique for color image classification using support vector machines with radial basis functions (SVM-RBF). It presents SVM-RBF as an improvement over other classification methods like SVM with ant colony optimization (SVM-ACO) and directed acyclic graph (SVM-DAG). The paper tests the different classifiers on 600 images across 3 classes, finding SVM-RBF achieved the highest precision and recall rates, with precision of 92.3-94% and recall of 84.8-91%. It concludes SVM-RBF more effectively reduces noise and the semantic gap to enhance image classification performance compared to the other methods.
CAPSULE NETWORK PERFORMANCE WITH AUTONOMOUS NAVIGATIONgerogepatton
Capsule Networks (CapsNets) have been proposed as an alternative to Convolutional Neural Networks (CNNs). This paper showcases how CapsNets are more capable than CNNs for autonomous agent exploration of realistic scenarios. In real world navigation, rewards external to agents may be rare. In turn, reinforcement learning algorithms can struggle to form meaningful policy functions. This paper’s approach Capsules Exploration Module (Caps-EM) pairs a CapsNets architecture with an Advantage Actor Critic algorithm. Other approaches for navigating sparse environments require intrinsic reward generators, such as the Intrinsic Curiosity Module (ICM) and Augmented Curiosity Modules (ACM). CapsEM uses a more compact architecture without need for intrinsic rewards. Tested using ViZDoom, the CapsEM uses 44% and 83% fewer trainable network parameters than the ICM and Depth-Augmented Curiosity Module (D-ACM), respectively, for 1141% and 437% average time improvement over the ICM and DACM, respectively, for converging to a policy function across "My Way Home" scenarios
Optimization of Number of Neurons in the Hidden Layer in Feed Forward Neural ...IJERA Editor
The architectures of Artificial Neural Networks (ANN) are based on the problem domain and it is applied during
the „training phase‟ of sample data and used to infer results for the remaining data in the testing phase.
Normally, the architecture consist of three layers as input, hidden, output layers with the number of nodes in the
input layer as number of known values on hand and the number of nodes as result to be computed out of the
values of input nodes and hidden nodes as the output layer. The number of nodes in the hidden layer is
heuristically decided so that the optimum value is obtained with reasonable number of iterations with other
parameters with its default values. This study mainly focuses on Cascade-Correlation Neural Networks (CCNN)
using Back-Propagation (BP) algorithm which finds the number of neurons during the training phase itself by
appending one from the previous iteration satisfying the error condition gives a promising result on the optimum
number of neurons in the hidden layer
This document summarizes a study that used artificial neural networks (ANNs) and the Multi-Layer Perceptron model (MLP) to predict the bearing capacities of steel driven piles in sandy soils. The ANN was trained on data from full-scale pile load tests, including pile length, diameter, soil elastic modulus, and soil friction angle as inputs. The output was pile bearing capacity. The study examined factors for effective ANN behavior, trained and tested the network, and analyzed the sensitivity of the inputs on the output capacity prediction.
This document provides an overview of self-organizing maps (SOM) as an unsupervised learning technique. It discusses the principles of self-organization including self-amplification, competition, and cooperation. The Willshaw-von der Malsburg model and Kohonen feature maps are presented as two approaches to building topographic maps through self-organization. The Kohonen SOM learning algorithm is described as involving competition between neurons to determine a winning neuron, cooperation between neighboring neurons, and adaptive changes to synaptic weights based on Hebbian learning principles.
Mobile Network Coverage Determination at 900MHz for Abuja Rural Areas using A...ijtsrd
This study proposes Artificial Neural Network ANN based field strength prediction models for the rural areas of Abuja, the federal capital territory of Nigeria. The ANN based models were created on bases of the Generalized Regression Neural network GRNN and the Multi Layer Perceptron Neural Network MLP NN . These networks were created, trained and tested for field strength prediction using received power data recorded at 900MHz from multiple Base Transceiver Stations BTSs distributed across the rural areas. Results indicate that the GRNN and MLP NN based models with Root Mean Squared Error RMSE values of 4.78dBm and 5.56dBm respectively, offer significant improvement over the empirical Hata Okumura counterpart, which overestimates the signal strength by an RMSE value of 20.17dBm. Deme C. Abraham ""Mobile Network Coverage Determination at 900MHz for Abuja Rural Areas using Artificial Neural Networks"" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-4 | Issue-2 , February 2020,
URL: https://www.ijtsrd.com/papers/ijtsrd30228.pdf
Paper Url : https://www.ijtsrd.com/computer-science/artificial-intelligence/30228/mobile-network-coverage-determination-at-900mhz-for-abuja-rural-areas-using-artificial-neural-networks/deme-c-abraham
Modeling of neural image compression using gradient decent technologytheijes
The International Journal of Engineering & Science is aimed at providing a platform for researchers, engineers, scientists, or educators to publish their original research results, to exchange new ideas, to disseminate information in innovative designs, engineering experiences and technological skills. It is also the Journal's objective to promote engineering and technology education. All papers submitted to the Journal will be blind peer-reviewed. Only original articles will be published.
The papers for publication in The International Journal of Engineering& Science are selected through rigorous peer reviews to ensure originality, timeliness, relevance, and readability.
Theoretical work submitted to the Journal should be original in its motivation or modeling structure. Empirical analysis should be based on a theoretical framework and should be capable of replication. It is expected that all materials required for replication (including computer programs and data sets) should be available upon request to the authors.
The International Journal of Engineering & Science would take much care in making your article published without much delay with your kind cooperation
An efficient technique for color image classification based on lower feature ...Alexander Decker
This document discusses an efficient technique for color image classification using support vector machines with radial basis functions (SVM-RBF). It presents SVM-RBF as an improvement over other classification methods like SVM with ant colony optimization (SVM-ACO) and directed acyclic graph (SVM-DAG). The paper tests the different classifiers on 600 images across 3 classes, finding SVM-RBF achieved the highest precision and recall rates, with precision of 92.3-94% and recall of 84.8-91%. It concludes SVM-RBF more effectively reduces noise and the semantic gap to enhance image classification performance compared to the other methods.
Black-box modeling of nonlinear system using evolutionary neural NARX modelIJECEIAES
Nonlinear systems with uncertainty and disturbance are very difficult to model using mathematic approach. Therefore, a black-box modeling approach without any prior knowledge is necessary. There are some modeling approaches have been used to develop a black box model such as fuzzy logic, neural network, and evolution algorithms. In this paper, an evolutionary neural network by combining a neural network and a modified differential evolution algorithm is applied to model a nonlinear system. The feasibility and effectiveness of the proposed modeling are tested on a piezoelectric actuator SISO system and an experimental quadruple tank MIMO system.
Machine Learning Algorithms for Image Classification of Hand Digits and Face ...IRJET Journal
This document discusses machine learning algorithms for image classification using five different classification schemes. It summarizes the mathematical models behind each classification algorithm, including Nearest Class Centroid classifier, Nearest Sub-Class Centroid classifier, k-Nearest Neighbor classifier, Perceptron trained using Backpropagation, and Perceptron trained using Mean Squared Error. It also describes two datasets used in the experiments - the MNIST dataset of handwritten digits and the ORL face recognition dataset. The performance of the five classification schemes are compared on these datasets.
NETWORK LEARNING AND TRAINING OF A CASCADED LINK-BASED FEED FORWARD NEURAL NE...ijaia
Presently, considering the technological advancement of our modern world, we are in dire need for a system that can learn new concepts and give decisions on its own. Hence the Artificial Neural Network is all that is required in the contemporary situation. In this paper, CLBFFNN is presented as a special and intelligent form of artificial neural networks that has the capability to adapt to training and learning of new ideas and be able to give decisions in a trimodal biometric system involving fingerprints, face and iris biometric data. It gives an overview of neural networks.
This document summarizes a study on short-term load forecasting using artificial neural networks. The study compares different neural network architectures, including feedforward, Elman recurrent, and Jordan recurrent networks. It also explores using particle swarm optimization to train an Elman recurrent neural network for improved forecasting accuracy. Results show the particle swarm optimized Elman recurrent network achieved the lowest error compared to other models.
A Deep Belief Network Approach to Learning Depth from Optical FlowReuben Feinman
The document describes using a deep belief network to learn depth from optical flow in videos. A biologically inspired model is used to generate motion features from video frames. These motion features are then used as input to a deep neural network that is trained to predict depth maps. The network is initialized using unsupervised pre-training of restricted Boltzmann machines and then fine-tuned with supervised backpropagation. Computer-generated graphics are used to obtain labeled training data of video frames paired with ground truth depth maps. The results show improved depth prediction over standard classifiers, demonstrating the potential of unsupervised feature learning for computer vision tasks.
The document discusses image recognition using convolutional neural networks (CNNs). It explains that CNNs consist of multiple layers of small neuron collections that look at small portions of an input image called receptive fields. The results are tiled to overlap and represent the original image better. CNNs learn filters through training rather than relying on hand-engineered features. Convolution involves calculating the overlap between functions as one is translated, and is used in CNNs to identify patterns across translated versions of inputs like images. Pointwise nonlinearities are applied between CNN layers to introduce nonlinearity.
Incorporating Kalman Filter in the Optimization of Quantum Neural Network Par...Waqas Tariq
Kalman filter have been used for the estimation of instantaneous states of linear dynamic systems. It is a good tool for inferring of missing information from noisy measurement. The quantum neural network is another approach to the merging of fuzzy logic with the neural network and that by the investment of quantum mechanics theory in building the structure of neural network. The gradient descent algorithm has been used, widely, in training the neural network, but the problem of local minima is one of the disadvantages of this algorithm. This paper presents an algorithm to train the quantum neural network by using the extended kalman filter.
Reflectivity Parameter Extraction from RADAR Images Using Back Propagation Al...IJECEIAES
This document discusses using backpropagation algorithms to extract reflectivity parameters from Doppler weather radar images. It begins with an introduction to pattern recognition using neural networks and an overview of artificial neural networks. It then describes different backpropagation algorithms that can be used for training multilayer perceptrons, including Levenberg-Marquardt, conjugate gradient, and resilient backpropagation. The document presents a method to preprocess Doppler radar images and use a neural network trained with backpropagation to identify colors in the image and estimate the corresponding reflectivity values based on a provided color scale. It analyzes using various backpropagation algorithms to identify colors in Doppler radar images and extract reflectivity information without human intervention.
with the help fof alkfjafnalfnlsnclsnclsnvsnvlsnvlds snlksnldsn nlncldnldncldsnclsd anflnfldnfldnfldsc knfldfnlfnlnfldnfldsnfldsnf lkfndslfndslfnldsfnlsdnflsdlflsfnsldnf lsnflsfdnldslds dsnfldsnflsdnflsnldsnf
In this work, the TREPAN algorithm is enhanced and extended for extracting decision trees from neural networks. We empirically evaluated the performance of the algorithm on a set of databases from real world events. This benchmark enhancement was achieved by adapting Single-test TREPAN and C4.5 decision tree induction algorithms to analyze the datasets. The models are then compared with X-TREPAN for
comprehensibility and classification accuracy. Furthermore, we validate the experimentations by applying statistical methods. Finally, the modified algorithm is extended to work with multi-class regression problems and the ability to comprehend generalized feed forward networks is achieved.
Sector based multicast routing algorithm for mobile ad hoc networksijwmn
Multicast routing algorithms for mobile ad-hoc networks have been extensively researched in the recent
past. In this paper, we present two algorithms for dealing with multicast routing problem using the notion
of virtual forces. We look at the effective force exerted on a packet and determine whether a node could be
considered as a Steiner node. The nodes' location information is used to generate virtual circuits
corresponding to the multicast route. QoS parameters are taken into consideration in the form of virtual
dampening force. The first algorithm produces relatively minimal multicast trees under the set of
constraints. We improve upon the first algorithm and present a second algorithm that provides
improvement in average residual energy in the network as well as effective cost per data packet
transmitted. In this paper, the virtual-force technique has been applied for multicast routing for the first
time in mobile ad-hoc networks.
This document discusses using the Levenberg-Marquardt algorithm for forecasting stock exchange share rates on the Karachi Stock Exchange. It provides an overview of artificial neural networks and how they can be used for financial forecasting applications. The Levenberg-Marquardt algorithm is presented as an efficient method for training neural networks to minimize errors through gradient descent. The document applies this method to train a neural network to predict the direction of change in share prices on the Karachi Stock Exchange. The network is trained on historical stock price data and testing shows it can achieve the performance goal of forecasting next day price changes.
A Novel GA-SVM Model For Vehicles And Pedestrial Classification In Videosijtsrd
The paper presents a novel algorithm for object classification in videos based on improved support vector machine (SVM) and genetic algorithm. One of the problems of support vector machine is selection of the appropriate parameters for the kernel. This has affected the accuracy of the SVM over the years. This research aims at optimizing the SVM Radial Basis kernel parameters using the genetic algorithm. Moving object classification is a requirement in smart visual surveillance systems as it allows the system to know the kind of object in the scene and be able to recognize the actions the object can perform. This paper presents an GA-SVM machine learning approach for real time object classification in videos. Radial distance signal features are extracted from the silhouettes of object detected in videos. The radial distance signals features are then normalized and fed into the GA-SVM model. The classification rate of 99.39% is achieved with the genetically trained SVM algorithm while 99.1% classification accuracy is achieved with the normal SVM. A comparison of this classifier with some other classifiers in terms of classification accuracy shows a better performance than other classifiers such as the normal SVM, Artificial neural network (ANN), Genetic Artificial neural network (GANN), K-nearest neighbor (K-NN) and K-Means classifiers. Akintola Kolawole G."A Novel GA-SVM Model For Vehicles And Pedestrial Classification In Videos" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-1 | Issue-4 , June 2017, URL: http://www.ijtsrd.com/papers/ijtsrd109.pdf http://www.ijtsrd.com/computer-science/artificial-intelligence/109/a-novel-ga-svm-model-for-vehicles-and-pedestrial-classification-in-videos/akintola-kolawole-g
Comparison of Neural Network Training Functions for Hematoma Classification i...IOSR Journals
Classification is one of the most important task in application areas of artificial neural networks
(ANN).Training neural networks is a complex task in the supervised learning field of research. The main
difficulty in adopting ANN is to find the most appropriate combination of learning, transfer and training
function for the classification task. We compared the performances of three types of training algorithms in feed
forward neural network for brain hematoma classification. In this work we have selected Gradient Descent
based backpropagation, Gradient Descent with momentum, Resilence backpropogation algorithms. Under
conjugate based algorithms, Scaled Conjugate back propagation, Conjugate Gradient backpropagation with
Polak-Riebreupdates(CGP) and Conjugate Gradient backpropagation with Fletcher-Reeves updates (CGF).The
last category is Quasi Newton based algorithm, under this BFGS, Levenberg-Marquardt algorithms are
selected. Proposed work compared training algorithm on the basis of mean square error, accuracy, rate of
convergence and correctness of the classification. Our conclusion about the training functions is based on the
simulation results
IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com
Imran Sarwar Bajwa, S. Irfan Hyder [2005], "PCA Based Image Classification of Single-Layered Cloud Types", in 1st IEEE International Conference on Emerging Technologies (ICET 2005), Islamabad, Pakistan, Jan 2005, pp:365-369
This document discusses using a learning automata approach to predict target locations in wireless sensor networks to reduce energy consumption and improve tracking accuracy. It proposes a learning automata based method that uses a target's movement history to predict its next location. Related works on target tracking techniques like tree-based, cluster-based, and prediction-based methods are summarized. Learning automata concepts are introduced. Simulation results are said to show the proposed method improves energy efficiency, reduces missed targets, and decreases transmitted packets compared to other methods.
X-TREPAN : A Multi Class Regression and Adapted Extraction of Comprehensible ...csandit
The document describes an algorithm called X-TREPAN that extracts decision trees from trained neural networks. X-TREPAN is an enhancement of the TREPAN algorithm that allows it to handle both multi-class classification and multi-class regression problems. It can also analyze generalized feed forward networks. The algorithm was tested on several real-world datasets and was found to generate decision trees with good classification accuracy while also maintaining comprehensibility.
Cluster head election using imperialist competitive algorithm (chei) for wire...ijmnct
One of the most important challenges of wireless sensor network is how to prolong its life time. The main
obstacle in these networks is the limited energy of nodes. We can overcome this problem by optimizing the
nodes' power consumption. The clustering mechanismis the one of the representative approachesto reduce
energy consumption, but optimum clustering of wireless sensor network is an NP-Hard problem. This
paper proposes a hybrid algorithm based on Imperialist competitive algorithm to overcome this clustering
problem. The proposed method, acts on one of the clusters in the network to choose the best sensor in
the cluster as a cluster head. To perform this action, the cluster is divided into several sub-clusters,
each of which has a cluster head. These cluster heads using Assimilation
policies, try to attract the regular nodes to themselves, and Using Imperialistic competition,
they compete with each other until one of these cluster heads is selected as the final cluster head. After this
stage, the algorithm work ends. This algorithm will balance the energy consumption in the network and
improve the network lifetime. To prove efficiency of proposed algorithm(CHEI), we simulated the proposed
algorithm compared with two clustering algorithms using the matlab
The document discusses using deep learning models to classify different types of eye movements from raw eye tracking data. Specifically, it explores using an attention convolutional neural network (ACNN) to classify samples as fixations, saccades, smooth pursuit, or noise. The ACNN outperforms current state-of-the-art models on labeled eye tracking datasets. Additionally, the document investigates using unsupervised pre-training of an autoencoder on a different eye tracking dataset to improve the generalization of deep learning models to new datasets.
BLOOD TISSUE IMAGE TO IDENTIFY MALARIA DISEASE CLASSIFICATIONIRJET Journal
This document describes using convolutional neural networks to classify malaria in blood tissue images. The researchers collected a dataset of blood tissue images from malaria-positive and negative patients. They preprocessed the data and developed convolutional neural network models like AlexNet and Lenet to classify the images. The models were trained on the dataset and evaluated based on metrics like accuracy, precision and recall. The goal is to create an automated method for malaria diagnosis that can help improve early detection and treatment in areas with limited resources.
Black-box modeling of nonlinear system using evolutionary neural NARX modelIJECEIAES
Nonlinear systems with uncertainty and disturbance are very difficult to model using mathematic approach. Therefore, a black-box modeling approach without any prior knowledge is necessary. There are some modeling approaches have been used to develop a black box model such as fuzzy logic, neural network, and evolution algorithms. In this paper, an evolutionary neural network by combining a neural network and a modified differential evolution algorithm is applied to model a nonlinear system. The feasibility and effectiveness of the proposed modeling are tested on a piezoelectric actuator SISO system and an experimental quadruple tank MIMO system.
Machine Learning Algorithms for Image Classification of Hand Digits and Face ...IRJET Journal
This document discusses machine learning algorithms for image classification using five different classification schemes. It summarizes the mathematical models behind each classification algorithm, including Nearest Class Centroid classifier, Nearest Sub-Class Centroid classifier, k-Nearest Neighbor classifier, Perceptron trained using Backpropagation, and Perceptron trained using Mean Squared Error. It also describes two datasets used in the experiments - the MNIST dataset of handwritten digits and the ORL face recognition dataset. The performance of the five classification schemes are compared on these datasets.
NETWORK LEARNING AND TRAINING OF A CASCADED LINK-BASED FEED FORWARD NEURAL NE...ijaia
Presently, considering the technological advancement of our modern world, we are in dire need for a system that can learn new concepts and give decisions on its own. Hence the Artificial Neural Network is all that is required in the contemporary situation. In this paper, CLBFFNN is presented as a special and intelligent form of artificial neural networks that has the capability to adapt to training and learning of new ideas and be able to give decisions in a trimodal biometric system involving fingerprints, face and iris biometric data. It gives an overview of neural networks.
This document summarizes a study on short-term load forecasting using artificial neural networks. The study compares different neural network architectures, including feedforward, Elman recurrent, and Jordan recurrent networks. It also explores using particle swarm optimization to train an Elman recurrent neural network for improved forecasting accuracy. Results show the particle swarm optimized Elman recurrent network achieved the lowest error compared to other models.
A Deep Belief Network Approach to Learning Depth from Optical FlowReuben Feinman
The document describes using a deep belief network to learn depth from optical flow in videos. A biologically inspired model is used to generate motion features from video frames. These motion features are then used as input to a deep neural network that is trained to predict depth maps. The network is initialized using unsupervised pre-training of restricted Boltzmann machines and then fine-tuned with supervised backpropagation. Computer-generated graphics are used to obtain labeled training data of video frames paired with ground truth depth maps. The results show improved depth prediction over standard classifiers, demonstrating the potential of unsupervised feature learning for computer vision tasks.
The document discusses image recognition using convolutional neural networks (CNNs). It explains that CNNs consist of multiple layers of small neuron collections that look at small portions of an input image called receptive fields. The results are tiled to overlap and represent the original image better. CNNs learn filters through training rather than relying on hand-engineered features. Convolution involves calculating the overlap between functions as one is translated, and is used in CNNs to identify patterns across translated versions of inputs like images. Pointwise nonlinearities are applied between CNN layers to introduce nonlinearity.
Incorporating Kalman Filter in the Optimization of Quantum Neural Network Par...Waqas Tariq
Kalman filter have been used for the estimation of instantaneous states of linear dynamic systems. It is a good tool for inferring of missing information from noisy measurement. The quantum neural network is another approach to the merging of fuzzy logic with the neural network and that by the investment of quantum mechanics theory in building the structure of neural network. The gradient descent algorithm has been used, widely, in training the neural network, but the problem of local minima is one of the disadvantages of this algorithm. This paper presents an algorithm to train the quantum neural network by using the extended kalman filter.
Reflectivity Parameter Extraction from RADAR Images Using Back Propagation Al...IJECEIAES
This document discusses using backpropagation algorithms to extract reflectivity parameters from Doppler weather radar images. It begins with an introduction to pattern recognition using neural networks and an overview of artificial neural networks. It then describes different backpropagation algorithms that can be used for training multilayer perceptrons, including Levenberg-Marquardt, conjugate gradient, and resilient backpropagation. The document presents a method to preprocess Doppler radar images and use a neural network trained with backpropagation to identify colors in the image and estimate the corresponding reflectivity values based on a provided color scale. It analyzes using various backpropagation algorithms to identify colors in Doppler radar images and extract reflectivity information without human intervention.
with the help fof alkfjafnalfnlsnclsnclsnvsnvlsnvlds snlksnldsn nlncldnldncldsnclsd anflnfldnfldnfldsc knfldfnlfnlnfldnfldsnfldsnf lkfndslfndslfnldsfnlsdnflsdlflsfnsldnf lsnflsfdnldslds dsnfldsnflsdnflsnldsnf
In this work, the TREPAN algorithm is enhanced and extended for extracting decision trees from neural networks. We empirically evaluated the performance of the algorithm on a set of databases from real world events. This benchmark enhancement was achieved by adapting Single-test TREPAN and C4.5 decision tree induction algorithms to analyze the datasets. The models are then compared with X-TREPAN for
comprehensibility and classification accuracy. Furthermore, we validate the experimentations by applying statistical methods. Finally, the modified algorithm is extended to work with multi-class regression problems and the ability to comprehend generalized feed forward networks is achieved.
Sector based multicast routing algorithm for mobile ad hoc networksijwmn
Multicast routing algorithms for mobile ad-hoc networks have been extensively researched in the recent
past. In this paper, we present two algorithms for dealing with multicast routing problem using the notion
of virtual forces. We look at the effective force exerted on a packet and determine whether a node could be
considered as a Steiner node. The nodes' location information is used to generate virtual circuits
corresponding to the multicast route. QoS parameters are taken into consideration in the form of virtual
dampening force. The first algorithm produces relatively minimal multicast trees under the set of
constraints. We improve upon the first algorithm and present a second algorithm that provides
improvement in average residual energy in the network as well as effective cost per data packet
transmitted. In this paper, the virtual-force technique has been applied for multicast routing for the first
time in mobile ad-hoc networks.
This document discusses using the Levenberg-Marquardt algorithm for forecasting stock exchange share rates on the Karachi Stock Exchange. It provides an overview of artificial neural networks and how they can be used for financial forecasting applications. The Levenberg-Marquardt algorithm is presented as an efficient method for training neural networks to minimize errors through gradient descent. The document applies this method to train a neural network to predict the direction of change in share prices on the Karachi Stock Exchange. The network is trained on historical stock price data and testing shows it can achieve the performance goal of forecasting next day price changes.
A Novel GA-SVM Model For Vehicles And Pedestrial Classification In Videosijtsrd
The paper presents a novel algorithm for object classification in videos based on improved support vector machine (SVM) and genetic algorithm. One of the problems of support vector machine is selection of the appropriate parameters for the kernel. This has affected the accuracy of the SVM over the years. This research aims at optimizing the SVM Radial Basis kernel parameters using the genetic algorithm. Moving object classification is a requirement in smart visual surveillance systems as it allows the system to know the kind of object in the scene and be able to recognize the actions the object can perform. This paper presents an GA-SVM machine learning approach for real time object classification in videos. Radial distance signal features are extracted from the silhouettes of object detected in videos. The radial distance signals features are then normalized and fed into the GA-SVM model. The classification rate of 99.39% is achieved with the genetically trained SVM algorithm while 99.1% classification accuracy is achieved with the normal SVM. A comparison of this classifier with some other classifiers in terms of classification accuracy shows a better performance than other classifiers such as the normal SVM, Artificial neural network (ANN), Genetic Artificial neural network (GANN), K-nearest neighbor (K-NN) and K-Means classifiers. Akintola Kolawole G."A Novel GA-SVM Model For Vehicles And Pedestrial Classification In Videos" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-1 | Issue-4 , June 2017, URL: http://www.ijtsrd.com/papers/ijtsrd109.pdf http://www.ijtsrd.com/computer-science/artificial-intelligence/109/a-novel-ga-svm-model-for-vehicles-and-pedestrial-classification-in-videos/akintola-kolawole-g
Comparison of Neural Network Training Functions for Hematoma Classification i...IOSR Journals
Classification is one of the most important task in application areas of artificial neural networks
(ANN).Training neural networks is a complex task in the supervised learning field of research. The main
difficulty in adopting ANN is to find the most appropriate combination of learning, transfer and training
function for the classification task. We compared the performances of three types of training algorithms in feed
forward neural network for brain hematoma classification. In this work we have selected Gradient Descent
based backpropagation, Gradient Descent with momentum, Resilence backpropogation algorithms. Under
conjugate based algorithms, Scaled Conjugate back propagation, Conjugate Gradient backpropagation with
Polak-Riebreupdates(CGP) and Conjugate Gradient backpropagation with Fletcher-Reeves updates (CGF).The
last category is Quasi Newton based algorithm, under this BFGS, Levenberg-Marquardt algorithms are
selected. Proposed work compared training algorithm on the basis of mean square error, accuracy, rate of
convergence and correctness of the classification. Our conclusion about the training functions is based on the
simulation results
IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com
Imran Sarwar Bajwa, S. Irfan Hyder [2005], "PCA Based Image Classification of Single-Layered Cloud Types", in 1st IEEE International Conference on Emerging Technologies (ICET 2005), Islamabad, Pakistan, Jan 2005, pp:365-369
This document discusses using a learning automata approach to predict target locations in wireless sensor networks to reduce energy consumption and improve tracking accuracy. It proposes a learning automata based method that uses a target's movement history to predict its next location. Related works on target tracking techniques like tree-based, cluster-based, and prediction-based methods are summarized. Learning automata concepts are introduced. Simulation results are said to show the proposed method improves energy efficiency, reduces missed targets, and decreases transmitted packets compared to other methods.
X-TREPAN : A Multi Class Regression and Adapted Extraction of Comprehensible ...csandit
The document describes an algorithm called X-TREPAN that extracts decision trees from trained neural networks. X-TREPAN is an enhancement of the TREPAN algorithm that allows it to handle both multi-class classification and multi-class regression problems. It can also analyze generalized feed forward networks. The algorithm was tested on several real-world datasets and was found to generate decision trees with good classification accuracy while also maintaining comprehensibility.
Cluster head election using imperialist competitive algorithm (chei) for wire...ijmnct
One of the most important challenges of wireless sensor network is how to prolong its life time. The main
obstacle in these networks is the limited energy of nodes. We can overcome this problem by optimizing the
nodes' power consumption. The clustering mechanismis the one of the representative approachesto reduce
energy consumption, but optimum clustering of wireless sensor network is an NP-Hard problem. This
paper proposes a hybrid algorithm based on Imperialist competitive algorithm to overcome this clustering
problem. The proposed method, acts on one of the clusters in the network to choose the best sensor in
the cluster as a cluster head. To perform this action, the cluster is divided into several sub-clusters,
each of which has a cluster head. These cluster heads using Assimilation
policies, try to attract the regular nodes to themselves, and Using Imperialistic competition,
they compete with each other until one of these cluster heads is selected as the final cluster head. After this
stage, the algorithm work ends. This algorithm will balance the energy consumption in the network and
improve the network lifetime. To prove efficiency of proposed algorithm(CHEI), we simulated the proposed
algorithm compared with two clustering algorithms using the matlab
The document discusses using deep learning models to classify different types of eye movements from raw eye tracking data. Specifically, it explores using an attention convolutional neural network (ACNN) to classify samples as fixations, saccades, smooth pursuit, or noise. The ACNN outperforms current state-of-the-art models on labeled eye tracking datasets. Additionally, the document investigates using unsupervised pre-training of an autoencoder on a different eye tracking dataset to improve the generalization of deep learning models to new datasets.
BLOOD TISSUE IMAGE TO IDENTIFY MALARIA DISEASE CLASSIFICATIONIRJET Journal
This document describes using convolutional neural networks to classify malaria in blood tissue images. The researchers collected a dataset of blood tissue images from malaria-positive and negative patients. They preprocessed the data and developed convolutional neural network models like AlexNet and Lenet to classify the images. The models were trained on the dataset and evaluated based on metrics like accuracy, precision and recall. The goal is to create an automated method for malaria diagnosis that can help improve early detection and treatment in areas with limited resources.
RunPool: A Dynamic Pooling Layer for Convolution Neural NetworkPutra Wanda
Deep learning (DL) has achieved a significant performance in computer vision problems, mainly in automatic feature extraction and representation. However, it is not easy to determine the best pooling method in a different case study. For instance, experts can implement the best types of pooling in image processing cases, which might not be optimal for various tasks. Thus, it is
required to keep in line with the philosophy of DL. In dynamic neural network architecture, it is not practically possible to find
a proper pooling technique for the layers. It is the primary reason why various pooling cannot be applied in the dynamic and multidimensional dataset. To deal with the limitations, it needs to construct an optimal pooling method as a better option than max pooling and average pooling. Therefore, we introduce a dynamic pooling layer called RunPool to train the convolutional
neuralnetwork(CNN)architecture.RunPoolpoolingisproposedtoregularizetheneuralnetworkthatreplacesthedeterministic
pooling functions. In the final section, we test the proposed pooling layer to address classification problems with online social network (OSN) dataset
AN IMPROVED METHOD FOR IDENTIFYING WELL-TEST INTERPRETATION MODEL BASED ON AG...IAEME Publication
This paper presents an approach based on applying an aggregated predictor formed by multiple versions of a multilayer neural network with a back-propagation optimization algorithm for helping the engineer to get a list of the most appropriate well-test interpretation models for a given set of pressure/ production data. The proposed method consists of three stages: (1) data decorrelation through principal component analysis to reduce the covariance between the variables and the dimension of the input layer in the artificial neural network, (2) bootstrap replicates of the learning set where the data is repeatedly sampled with a random split of the data into train sets and using these as new learning sets, and (3) automatic reservoir model identification through aggregated predictor formed by a plurality vote when predicting a new class. This method is described in detail to ensure successful replication of results. The required training and test dataset were generated by using analytical solution models. In our case, there were used 600 samples: 300 for training, 100 for cross-validation, and 200 for testing. Different network structures were tested during this study to arrive at optimum network design. We notice that the single net methodology always brings about confusion in selecting the correct model even though the training results for the constructed networks are close to 1. We notice also that the principal component analysis is an effective strategy in reducing the number of input features, simplifying the network structure, and lowering the training time of the ANN. The results obtained show that the proposed model provides better performance when predicting new data with a coefficient of correlation approximately equal to 95% Compared to a previous approach 80%, the combination of the PCA and ANN is more stable and determine the more accurate results with lesser computational complexity than was feasible previously. Clearly, the aggregated predictor is more stable and shows less bad classes compared to the previous approach.
This document discusses comparing the performance of different convolutional neural networks (CNNs) when trained on large image datasets using Apache Spark. It summarizes the datasets used - CIFAR-10 and ImageNet - and preprocessing done to standardize image sizes. It then provides an overview of CNN architecture, including convolutional layers, pooling layers, and fully connected layers. Finally, it introduces SparkNet, a framework that allows training deep networks using Spark by wrapping Caffe and providing tools for distributed deep learning on Spark. The goal is to see if SparkNet can provide faster training times compared to a single machine by distributing training across a cluster.
Easily Trainable Neural Network Using TransferLearningIRJET Journal
This document discusses using transfer learning to train a neural network to detect a specific window structure quickly using a limited number of training samples. It uses the pre-trained Tiny YOLOv3 model and compares training it from scratch, using ImageNet weights, and using COCO weights. The best results were obtained by pre-training the model on a dataset of real-world window images, then using those weights to train on the target window structure. This approach achieved a 98% mean average precision after only 300 iterations, an 83% decrease in time to convergence compared to training from scratch. Transfer learning was able to improve performance even for detecting a relatively featureless object like a window.
Locate, Size and Count: Accurately Resolving People in Dense Crowds via Detec...IRJET Journal
This document presents a CNN-MRF based system for counting people in dense crowd images. The system divides dense crowd images into overlapping patches. A CNN is used to extract features from each patch and regress the patch count. Since patches overlap, neighboring patch counts are strongly correlated. An MRF smooths the patch counts using this correlation to obtain a more accurate overall count. The system was developed to address challenges in accurately locating, sizing, and counting people in dense crowds via detection.
IMPROVING SUPERVISED CLASSIFICATION OF DAILY ACTIVITIES LIVING USING NEW COST...cscpconf
The growing population of elders in the society calls for a new approach in care giving. By inferring what activities elderly are performing in their houses it is possible to determine their
physical and cognitive capabilities. In this paper we show the potential of important discriminative classifiers namely the Soft-Support Vector Machines (C-SVM), Conditional Random Fields (CRF) and k-Nearest Neighbors (k-NN) for recognizing activities from sensor patterns in a smart home environment. We address also the class imbalance problem in activity recognition field which has been known to hinder the learning performance of classifiers. Cost sensitive learning is attractive under most imbalanced circumstances, but it is difficult to determine the precise misclassification costs in practice. We introduce a new criterion for selecting the suitable cost parameter C of the C-SVM method. Through our evaluation on four real world imbalanced activity datasets, we demonstrate that C-SVM based on our proposed criterion outperforms the state-of-the-art discriminative methods in activity recognition.
In recent machine learning community, there is a trend of constructing a linear logarithm version of
nonlinear version through the ‘kernel method’ for example kernel principal component analysis, kernel
fisher discriminant analysis, support Vector Machines (SVMs), and the current kernel clustering
algorithms. Typically, in unsupervised methods of clustering algorithms utilizing kernel method, a
nonlinear mapping is operated initially in order to map the data into a much higher space feature, and then
clustering is executed. A hitch of these kernel clustering algorithms is that the clustering prototype resides
in increased features specs of dimensions and therefore lack intuitive and clear descriptions without
utilizing added approximation of projection from the specs to the data as executed in the literature
presented. This paper aims to utilize the ‘kernel method’, a novel clustering algorithm, founded on the
conventional fuzzy clustering algorithm (FCM) is anticipated and known as kernel fuzzy c-means algorithm
(KFCM). This method embraces a novel kernel-induced metric in the space of data in order to interchange
the novel Euclidean matric norm in cluster prototype and fuzzy clustering algorithm still reside in the space
of data so that the results of clustering could be interpreted and reformulated in the spaces which are
original. This property is used for clustering incomplete data. Execution on supposed data illustrate that
KFCM has improved performance of clustering and stout as compare to other transformations of FCM for
clustering incomplete data.
Quantum machine learning, an important element of quantum computing, recently has gained research attention around the world. In this paper, we have proposed a quantum machine learning model to classify images using a quantum classifier. We exhibit the results of a comprehensive quantum classifier with transfer learning applied to image datasets in particular. The work uses hybrid transfer learning technique along with the classical pre-trained network and variational quantum circuits as their final layers on a small scale of dataset. The implementation is carried out in a quantum processor of a chosen set of highly informative functions using PennyLane a cross-platform software package for using quantum computers to evaluate the high-resolution image classifier. The performance of the model proved to be more accurate than its counterpart and outperforms all other existing classical models in terms of time and competence.
This document provides an internship report on classifying handwritten digits using a convolutional neural network. It includes an abstract, introduction on CNNs, explanations of CNN layers including convolution, pooling and fully connected layers. It also discusses padding and applications of CNNs such as computer vision, image recognition and natural language processing.
This document is an internship report submitted by Raghunandan J to Eckovation about a project on classifying handwritten digits using a convolutional neural network. It provides an introduction to convolutional neural networks and explains each layer of a CNN including the input, convolutional layer, pooling layer, and fully connected layer. It also gives examples of real-world applications that use artificial neural networks like Google Maps, Google Images, and voice assistants.
TEST-COST-SENSITIVE CONVOLUTIONAL NEURAL NETWORKS WITH EXPERT BRANCHESsipij
It has been proven that deeper convolutional neural networks (CNN) can result in better accuracy in many
problems, but this accuracy comes with a high computational cost. Also, input instances have not the same
difficulty. As a solution for accuracy vs. computational cost dilemma, we introduce a new test-cost-sensitive
method for convolutional neural networks. This method trains a CNN with a set of auxiliary outputs and
expert branches in some middle layers of the network. The expert branches decide to use a shallower part
of the network or going deeper to the end, based on the difficulty of input instance. The expert branches
learn to determine: is the current network prediction is wrong and if the given instance passed to deeper
layers of the network it will generate right output; If not, then the expert branches stop the computation
process. The experimental results on standard dataset CIFAR-10 show that the proposed method can train
models with lower test-cost and competitive accuracy in comparison with basic models.
This document proposes a dynamic clustering algorithm using fuzzy c-means clustering. It begins with an introduction to fuzzy c-means clustering and its limitations when the chosen number of clusters is incorrect. It then proposes a dynamic clustering algorithm that starts with a fixed number of clusters but can automatically increase the number of clusters during iterations based on the data, improving purity. The algorithm is described and examples are provided to illustrate its effectiveness at forming clear clusters after iterations and determining when clustering has terminated.
Image Captioning Generator using Deep Machine Learningijtsrd
Technologys scope has evolved into one of the most powerful tools for human development in a variety of fields.AI and machine learning have become one of the most powerful tools for completing tasks quickly and accurately without the need for human intervention. This project demonstrates how deep machine learning can be used to create a caption or a sentence for a given picture. This can be used for visually impaired persons, as well as automobiles for self identification, and for various applications to verify quickly and easily. The Convolutional Neural Network CNN is used to describe the alphabet, and the Long Short Term Memory LSTM is used to organize the right meaningful sentences in this model. The flicker 8k and flicker 30k datasets were used to train this. Sreejith S P | Vijayakumar A "Image Captioning Generator using Deep Machine Learning" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-5 | Issue-4 , June 2021, URL: https://www.ijtsrd.compapers/ijtsrd42344.pdf Paper URL: https://www.ijtsrd.comcomputer-science/artificial-intelligence/42344/image-captioning-generator-using-deep-machine-learning/sreejith-s-p
The document describes a study that used a convolutional neural network with a ConvNeXtLarge architecture to classify skin cancer images into benign and malignant classes. The CNN model was trained on a dataset of 3,297 skin cancer images from Kaggle. It achieved an AUC of 0.91 for classifying the images, demonstrating the ConvNeXtLarge architecture is effective for this task. The study aims to help early diagnosis and treatment of skin cancers.
This document describes a deep reinforcement learning method called DQN that achieved human-level performance on 49 Atari 2600 games. The DQN uses a convolutional neural network to learn successful policies for playing games directly from raw pixel inputs. It outperformed existing reinforcement learning methods on 43 of the 49 games and achieved over 75% of a human tester's score on 29 games. The DQN was able to stably train large neural networks using reinforcement learning and stochastic gradient descent to learn policies from high-dimensional visual inputs with minimal prior knowledge.
Image classification with Deep Neural NetworksYogendra Tamang
This document discusses image classification using deep neural networks. It provides background on image classification and convolutional neural networks. The document outlines techniques like activation functions, pooling, dropout and data augmentation to prevent overfitting. It summarizes a paper on ImageNet classification using CNNs with multiple convolutional and fully connected layers. The paper achieved state-of-the-art results on ImageNet in 2010 and 2012 by training CNNs on a large dataset using multiple GPUs.
Integrated Hidden Markov Model and Kalman Filter for Online Object Trackingijsrd.com
Visual prior from generic real-world images study to represent that objects in a scene. The existing work presented online tracking algorithm to transfers visual prior learned offline for online object tracking. To learn complete dictionary to represent visual prior with collection of real world images. Prior knowledge of objects is generic and training image set does not contain any observation of target object. Transfer learned visual prior to construct object representation using Sparse coding and Multiscale max pooling. Linear classifier is learned online to distinguish target from background and also to identify target and background appearance variations over time. Tracking is carried out within Bayesian inference framework and learned classifier is used to construct observation model. Particle filter is used to estimate the tracking result sequentially however, unable to work efficiently in noisy scenes. Time sift variance were not appropriated to track target object with observer value to prior information of object structure. Proposal HMM based kalman filter to improve online target tracking in noisy sequential image frames. The covariance vector is measured to identify noisy scenes. Discrete time steps are evaluated for identifying target object with background separation. Experiment conducted on challenging sequences of scene. To evaluate the performance of object tracking algorithm in terms of tracking success rate, Centre location error, Number of scenes, Learning object sizes, and Latency for tracking.
Similar to Capsule Network Performance with Autonomous Navigation (20)
International Journal of Artificial Intelligence & Applications (IJAIA)gerogepatton
The International Journal of Artificial Intelligence & Applications (IJAIA) is a bi monthly open access peer-reviewed journal that publishes articles which contribute new results in all areas of the Artificial Intelligence & Applications (IJAIA). It is an international journal intended for professionals and researchers in all fields of AI for researchers, programmers, and software and hardware manufacturers. The journal also aims to publish new attempts in the form of special issues on emerging areas in Artificial Intelligence and applications.
PhotoQR: A Novel ID Card with an Encoded Viewgerogepatton
There is an increasing interest in developing techniques to identify and assess data to allow an easy and
continuous access to resources, services or places that require thorough ID control. Usually, in order to
give access to these resources, different kinds of documents are mandatory. In order to avoid forgeries
without the need of extra credentials, a new system –named photoQR, is here proposed. This system is
based on a ID card having two objects: one person’s picture (pre-processed via blur and/or swirl
techniques) and one QR code containing embedded data related to the picture. The idea is that the picture
and the QR code can assess each other by a proper hash value in the QR. The QR without the picture
cannot be assessed and vice versa. An open source prototype of the photoQR system has been implemented
in Python and can be used both in offline and real-time environments, which effectively combines security
concepts and image processing algorithms to obtain data assessment.
12th International Conference of Artificial Intelligence and Fuzzy Logic (AI ...gerogepatton
12th International Conference of Artificial Intelligence and Fuzzy Logic (AI & FL 2024) provides a
forum for researchers who address this issue and to present their work in a peer-reviewed forum. Authors
are solicited to contribute to the conference by submitting articles that illustrate research results, projects,
surveying works and industrial experiences that describe significant advances in the following areas, but
are not limited to these topics only.
International Journal of Artificial Intelligence & Applications (IJAIA)gerogepatton
The International Journal of Artificial Intelligence & Applications (IJAIA) is a bi monthly open access peer-reviewed journal that publishes articles which contribute new results in all areas of the Artificial Intelligence & Applications (IJAIA). It is an international journal intended for professionals and researchers in all fields of AI for researchers, programmers, and software and hardware manufacturers. The journal also aims to publish new attempts in the form of special issues on emerging areas in Artificial Intelligence and applications.
International Conference on NLP, Artificial Intelligence, Machine Learning an...gerogepatton
International Conference on NLP, Artificial Intelligence, Machine Learning and Applications (NLAIM 2024) offers a premier global platform for exchanging insights and findings in the theory, methodology, and applications of NLP, Artificial Intelligence, Machine Learning, and their applications. The conference seeks substantial contributions across all key domains of NLP, Artificial Intelligence, Machine Learning, and their practical applications, aiming to foster both theoretical advancements and real-world implementations. With a focus on facilitating collaboration between researchers and practitioners from academia and industry, the conference serves as a nexus for sharing the latest developments in the field.
DEEP LEARNING FOR SMART GRID INTRUSION DETECTION: A HYBRID CNN-LSTM-BASED MODELgerogepatton
As digital technology becomes more deeply embedded in power systems, protecting the communication
networks of Smart Grids (SG) has emerged as a critical concern. Distributed Network Protocol 3 (DNP3)
represents a multi-tiered application layer protocol extensively utilized in Supervisory Control and Data
Acquisition (SCADA)-based smart grids to facilitate real-time data gathering and control functionalities.
Robust Intrusion Detection Systems (IDS) are necessary for early threat detection and mitigation because
of the interconnection of these networks, which makes them vulnerable to a variety of cyberattacks. To
solve this issue, this paper develops a hybrid Deep Learning (DL) model specifically designed for intrusion
detection in smart grids. The proposed approach is a combination of the Convolutional Neural Network
(CNN) and the Long-Short-Term Memory algorithms (LSTM). We employed a recent intrusion detection
dataset (DNP3), which focuses on unauthorized commands and Denial of Service (DoS) cyberattacks, to
train and test our model. The results of our experiments show that our CNN-LSTM method is much better
at finding smart grid intrusions than other deep learning algorithms used for classification. In addition,
our proposed approach improves accuracy, precision, recall, and F1 score, achieving a high detection
accuracy rate of 99.50%.
10th International Conference on Artificial Intelligence and Applications (AI...gerogepatton
10th International Conference on Artificial Intelligence and Applications (AI 2024) will provide an excellent international forum for sharing knowledge and results in theory, methodology and applications of Artificial Intelligence and its applications. The Conference looks for significant contributions to all major fields of the Artificial Intelligence, Soft Computing in theoretical and practical aspects. The aim of the Conference is to provide a platform to the researchers and practitioners from both academia as well as industry to meet and share cutting-edge development in the field.
International Journal of Artificial Intelligence & Applications (IJAIA)gerogepatton
The International Journal of Artificial Intelligence & Applications (IJAIA) is a bi monthly open access peer-reviewed journal that publishes articles which contribute new results in all areas of the Artificial Intelligence & Applications (IJAIA). It is an international journal intended for professionals and researchers in all fields of AI for researchers, programmers, and software and hardware manufacturers. The journal also aims to publish new attempts in the form of special issues on emerging areas in Artificial Intelligence and applications.
Immunizing Image Classifiers Against Localized Adversary Attacksgerogepatton
This paper addresses the vulnerability of deep learning models, particularly convolutional neural networks
(CNN)s, to adversarial attacks and presents a proactive training technique designed to counter them. We
introduce a novel volumization algorithm, which transforms 2D images into 3D volumetric representations.
When combined with 3D convolution and deep curriculum learning optimization (CLO), itsignificantly improves
the immunity of models against localized universal attacks by up to 40%. We evaluate our proposed approach
using contemporary CNN architectures and the modified Canadian Institute for Advanced Research (CIFAR-10
and CIFAR-100) and ImageNet Large Scale Visual Recognition Challenge (ILSVRC12) datasets, showcasing
accuracy improvements over previous techniques. The results indicate that the combination of the volumetric
input and curriculum learning holds significant promise for mitigating adversarial attacks without necessitating
adversary training.
May 2024 - Top 10 Read Articles in Artificial Intelligence and Applications (...gerogepatton
The International Journal of Artificial Intelligence & Applications (IJAIA) is a bi monthly open access peer-reviewed journal that publishes articles which contribute new results in all areas of the Artificial Intelligence & Applications (IJAIA). It is an international journal intended for professionals and researchers in all fields of AI for researchers, programmers, and software and hardware manufacturers. The journal also aims to publish new attempts in the form of special issues on emerging areas in Artificial Intelligence and applications.
3rd International Conference on Artificial Intelligence Advances (AIAD 2024)gerogepatton
3rd International Conference on Artificial Intelligence Advances (AIAD 2024) will act as a major forum for the presentation of innovative ideas, approaches, developments, and research projects in the area advanced Artificial Intelligence. It will also serve to facilitate the exchange of information between researchers and industry professionals to discuss the latest issues and advancement in the research area. Core areas of AI and advanced multi-disciplinary and its applications will be covered during the conferences.
International Journal of Artificial Intelligence & Applications (IJAIA)gerogepatton
The International Journal of Artificial Intelligence & Applications (IJAIA) is a bi monthly open access peer-reviewed journal that publishes articles which contribute new results in all areas of the Artificial Intelligence & Applications (IJAIA). It is an international journal intended for professionals and researchers in all fields of AI for researchers, programmers, and software and hardware manufacturers. The journal also aims to publish new attempts in the form of special issues on emerging areas in Artificial Intelligence and applications.
Information Extraction from Product Labels: A Machine Vision Approachgerogepatton
This research tackles the challenge of manual data extraction from product labels by employing a blend of
computer vision and Natural Language Processing (NLP). We introduce an enhanced model that combines
Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) in a Convolutional
Recurrent Neural Network (CRNN) for reliable text recognition. Our model is further refined by
incorporating the Tesseract OCR engine, enhancing its applicability in Optical Character Recognition
(OCR) tasks. The methodology is augmented by NLP techniques and extended through the Open Food
Facts API (Application Programming Interface) for database population and text-only label prediction.
The CRNN model is trained on encoded labels and evaluated for accuracy on a dedicated test set.
Importantly, our approach enables visually impaired individuals to access essential information on
product labels, such as directions and ingredients. Overall, the study highlights the efficacy of deep
learning and OCR in automating label extraction and recognition.
10th International Conference on Artificial Intelligence and Applications (AI...gerogepatton
10th International Conference on Artificial Intelligence and Applications (AI 2024) will provide an excellent international forum for sharing knowledge and results in theory, methodology and applications of Artificial Intelligence and its applications. The Conference looks for significant contributions to all major fields of the Artificial Intelligence, Soft Computing in theoretical and practical aspects. The aim of the Conference is to provide a platform to the researchers and practitioners from both academia as well as industry to meet and share cutting-edge development in the field.
International Journal of Artificial Intelligence & Applications (IJAIA)gerogepatton
The International Journal of Artificial Intelligence & Applications (IJAIA) is a bi monthly open access peer-reviewed journal that publishes articles which contribute new results in all areas of the Artificial Intelligence & Applications (IJAIA). It is an international journal intended for professionals and researchers in all fields of AI for researchers, programmers, and software and hardware manufacturers. The journal also aims to publish new attempts in the form of special issues on emerging areas in Artificial Intelligence and applications.
Research on Fuzzy C- Clustering Recursive Genetic Algorithm based on Cloud Co...gerogepatton
Aiming at the problems of poor local search ability and precocious convergence of fuzzy C-cluster
recursive genetic algorithm (FOLD++), a new fuzzy C-cluster recursive genetic algorithm based on
Bayesian function adaptation search (TS) was proposed by incorporating the idea of Bayesian function
adaptation search into fuzzy C-cluster recursive genetic algorithm. The new algorithm combines the
advantages of FOLD++ and TS. In the early stage of optimization, fuzzy C-cluster recursive genetic
algorithm is used to get a good initial value, and the individual extreme value pbest is put into Bayesian
function adaptation table. In the late stage of optimization, when the searching ability of fuzzy C-cluster
recursive genetic is weakened, the short term memory function of Bayesian function adaptation table in
Bayesian function adaptation search algorithm is utilized. Make it jump out of the local optimal solution,
and allow bad solutions to be accepted during the search. The improved algorithm is applied to function
optimization, and the simulation results show that the calculation accuracy and stability of the algorithm
are improved, and the effectiveness of the improved algorithm is verified
International Journal of Artificial Intelligence & Applications (IJAIA)gerogepatton
The International Journal of Artificial Intelligence & Applications (IJAIA) is a bi monthly open access peer-reviewed journal that publishes articles which contribute new results in all areas of the Artificial Intelligence & Applications (IJAIA). It is an international journal intended for professionals and researchers in all fields of AI for researchers, programmers, and software and hardware manufacturers. The journal also aims to publish new attempts in the form of special issues on emerging areas in Artificial Intelligence and applications.
10th International Conference on Artificial Intelligence and Soft Computing (...gerogepatton
10th International Conference on Artificial Intelligence and Soft Computing (AIS 2024) will
provide an excellent international forum for sharing knowledge and results in theory, methodology, and
applications of Artificial Intelligence, Soft Computing. The Conference looks for significant
contributions to all major fields of the Artificial Intelligence, Soft Computing in theoretical and practical
aspects. The aim of the Conference is to provide a platform to the researchers and practitioners from
both academia as well as industry to meet and share cutting-edge development in the field.
International Journal of Artificial Intelligence & Applications (IJAIA)gerogepatton
Employee attrition refers to the decrease in staff numbers within an organization due to various reasons.
As it has a negative impact on long-term growth objectives and workplace productivity, firms have
recognized it as a significant concern. To address this issue, organizations are increasingly turning to
machine-learning approaches to forecast employee attrition rates. This topic has gained significant
attention from researchers, especially in recent times. Several studies have applied various machinelearning methods to predict employee attrition, producing different resultsdepending on the employed
methods, factors, and datasets. However, there has been no comprehensive comparative review of multiple
studies applying machine-learning models to predict employee attrition to date. Therefore, this study aims
to fill this gap by providing an overview of research conducted on applying machine learning to predict
employee attrition from 2019 to February 2024. A literature review of relevant studies was conducted,
summarized, and classified. Most studies agree on conducting comparative experiments with multiple
predictive models to determine the most effective one.From this literature survey, the RF algorithm and
XGB ensemble method are repeatedly the best-performing, outperforming many other algorithms.
Additionally, the application of deep learning to employee attrition prediction issues also shows promise.
While there are discrepancies in the datasets used in previous studies, it is notable that the dataset
provided by IBM is the most widely utilized. This study serves as a concise review for new researchers,
facilitating their understanding of the primary techniques employed in predicting employee attrition and
highlighting recent research trends in this field. Furthermore, it provides organizations with insight into
the prominent factors affecting employee attrition, as identified by studies, enabling them to implement
solutions aimed at reducing attrition rates.
10th International Conference on Artificial Intelligence and Applications (AI...gerogepatton
10th International Conference on Artificial Intelligence and Applications (AIFU 2024) is a forum for presenting new advances and research results in the fields of Artificial Intelligence. The conference will bring together leading researchers, engineers and scientists in the domain of interest from around the world. The scope of the conference covers all theoretical and practical aspects of the Artificial Intelligence.
Nunit vs XUnit vs MSTest Differences Between These Unit Testing Frameworks.pdfflufftailshop
When it comes to unit testing in the .NET ecosystem, developers have a wide range of options available. Among the most popular choices are NUnit, XUnit, and MSTest. These unit testing frameworks provide essential tools and features to help ensure the quality and reliability of code. However, understanding the differences between these frameworks is crucial for selecting the most suitable one for your projects.
Digital Marketing Trends in 2024 | Guide for Staying AheadWask
https://www.wask.co/ebooks/digital-marketing-trends-in-2024
Feeling lost in the digital marketing whirlwind of 2024? Technology is changing, consumer habits are evolving, and staying ahead of the curve feels like a never-ending pursuit. This e-book is your compass. Dive into actionable insights to handle the complexities of modern marketing. From hyper-personalization to the power of user-generated content, learn how to build long-term relationships with your audience and unlock the secrets to success in the ever-shifting digital landscape.
Programming Foundation Models with DSPy - Meetup SlidesZilliz
Prompting language models is hard, while programming language models is easy. In this talk, I will discuss the state-of-the-art framework DSPy for programming foundation models with its powerful optimizers and runtime constraint system.
Ocean lotus Threat actors project by John Sitima 2024 (1).pptxSitimaJohn
Ocean Lotus cyber threat actors represent a sophisticated, persistent, and politically motivated group that poses a significant risk to organizations and individuals in the Southeast Asian region. Their continuous evolution and adaptability underscore the need for robust cybersecurity measures and international cooperation to identify and mitigate the threats posed by such advanced persistent threat groups.
Have you ever been confused by the myriad of choices offered by AWS for hosting a website or an API?
Lambda, Elastic Beanstalk, Lightsail, Amplify, S3 (and more!) can each host websites + APIs. But which one should we choose?
Which one is cheapest? Which one is fastest? Which one will scale to meet our needs?
Join me in this session as we dive into each AWS hosting service to determine which one is best for your scenario and explain why!
leewayhertz.com-AI in predictive maintenance Use cases technologies benefits ...alexjohnson7307
Predictive maintenance is a proactive approach that anticipates equipment failures before they happen. At the forefront of this innovative strategy is Artificial Intelligence (AI), which brings unprecedented precision and efficiency. AI in predictive maintenance is transforming industries by reducing downtime, minimizing costs, and enhancing productivity.
Letter and Document Automation for Bonterra Impact Management (fka Social Sol...Jeffrey Haguewood
Sidekick Solutions uses Bonterra Impact Management (fka Social Solutions Apricot) and automation solutions to integrate data for business workflows.
We believe integration and automation are essential to user experience and the promise of efficient work through technology. Automation is the critical ingredient to realizing that full vision. We develop integration products and services for Bonterra Case Management software to support the deployment of automations for a variety of use cases.
This video focuses on automated letter generation for Bonterra Impact Management using Google Workspace or Microsoft 365.
Interested in deploying letter generation automations for Bonterra Impact Management? Contact us at sales@sidekicksolutionsllc.com to discuss next steps.
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAUpanagenda
Webinar Recording: https://www.panagenda.com/webinars/hcl-notes-und-domino-lizenzkostenreduzierung-in-der-welt-von-dlau/
DLAU und die Lizenzen nach dem CCB- und CCX-Modell sind für viele in der HCL-Community seit letztem Jahr ein heißes Thema. Als Notes- oder Domino-Kunde haben Sie vielleicht mit unerwartet hohen Benutzerzahlen und Lizenzgebühren zu kämpfen. Sie fragen sich vielleicht, wie diese neue Art der Lizenzierung funktioniert und welchen Nutzen sie Ihnen bringt. Vor allem wollen Sie sicherlich Ihr Budget einhalten und Kosten sparen, wo immer möglich. Das verstehen wir und wir möchten Ihnen dabei helfen!
Wir erklären Ihnen, wie Sie häufige Konfigurationsprobleme lösen können, die dazu führen können, dass mehr Benutzer gezählt werden als nötig, und wie Sie überflüssige oder ungenutzte Konten identifizieren und entfernen können, um Geld zu sparen. Es gibt auch einige Ansätze, die zu unnötigen Ausgaben führen können, z. B. wenn ein Personendokument anstelle eines Mail-Ins für geteilte Mailboxen verwendet wird. Wir zeigen Ihnen solche Fälle und deren Lösungen. Und natürlich erklären wir Ihnen das neue Lizenzmodell.
Nehmen Sie an diesem Webinar teil, bei dem HCL-Ambassador Marc Thomas und Gastredner Franz Walder Ihnen diese neue Welt näherbringen. Es vermittelt Ihnen die Tools und das Know-how, um den Überblick zu bewahren. Sie werden in der Lage sein, Ihre Kosten durch eine optimierte Domino-Konfiguration zu reduzieren und auch in Zukunft gering zu halten.
Diese Themen werden behandelt
- Reduzierung der Lizenzkosten durch Auffinden und Beheben von Fehlkonfigurationen und überflüssigen Konten
- Wie funktionieren CCB- und CCX-Lizenzen wirklich?
- Verstehen des DLAU-Tools und wie man es am besten nutzt
- Tipps für häufige Problembereiche, wie z. B. Team-Postfächer, Funktions-/Testbenutzer usw.
- Praxisbeispiele und Best Practices zum sofortigen Umsetzen
Fueling AI with Great Data with Airbyte WebinarZilliz
This talk will focus on how to collect data from a variety of sources, leveraging this data for RAG and other GenAI use cases, and finally charting your course to productionalization.
Trusted Execution Environment for Decentralized Process MiningLucaBarbaro3
Presentation of the paper "Trusted Execution Environment for Decentralized Process Mining" given during the CAiSE 2024 Conference in Cyprus on June 7, 2024.
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdfMalak Abu Hammad
Discover how MongoDB Atlas and vector search technology can revolutionize your application's search capabilities. This comprehensive presentation covers:
* What is Vector Search?
* Importance and benefits of vector search
* Practical use cases across various industries
* Step-by-step implementation guide
* Live demos with code snippets
* Enhancing LLM capabilities with vector search
* Best practices and optimization strategies
Perfect for developers, AI enthusiasts, and tech leaders. Learn how to leverage MongoDB Atlas to deliver highly relevant, context-aware search results, transforming your data retrieval process. Stay ahead in tech innovation and maximize the potential of your applications.
#MongoDB #VectorSearch #AI #SemanticSearch #TechInnovation #DataScience #LLM #MachineLearning #SearchTechnology
TrustArc Webinar - 2024 Global Privacy SurveyTrustArc
How does your privacy program stack up against your peers? What challenges are privacy teams tackling and prioritizing in 2024?
In the fifth annual Global Privacy Benchmarks Survey, we asked over 1,800 global privacy professionals and business executives to share their perspectives on the current state of privacy inside and outside of their organizations. This year’s report focused on emerging areas of importance for privacy and compliance professionals, including considerations and implications of Artificial Intelligence (AI) technologies, building brand trust, and different approaches for achieving higher privacy competence scores.
See how organizational priorities and strategic approaches to data security and privacy are evolving around the globe.
This webinar will review:
- The top 10 privacy insights from the fifth annual Global Privacy Benchmarks Survey
- The top challenges for privacy leaders, practitioners, and organizations in 2024
- Key themes to consider in developing and maintaining your privacy program
Generating privacy-protected synthetic data using Secludy and MilvusZilliz
During this demo, the founders of Secludy will demonstrate how their system utilizes Milvus to store and manipulate embeddings for generating privacy-protected synthetic data. Their approach not only maintains the confidentiality of the original data but also enhances the utility and scalability of LLMs under privacy constraints. Attendees, including machine learning engineers, data scientists, and data managers, will witness first-hand how Secludy's integration with Milvus empowers organizations to harness the power of LLMs securely and efficiently.
Generating privacy-protected synthetic data using Secludy and Milvus
Capsule Network Performance with Autonomous Navigation
1. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
DOI : 10.5121/ijaia.2020.11101 1
CAPSULE NETWORK PERFORMANCE WITH
AUTONOMOUS NAVIGATION
Thomas Molnar and Eugenio Culurciello
ECE, Purdue University, 610 Purdue Mall, West Lafayette, 47907, IN, USA
ABSTRACT
Capsule Networks (CapsNets) have been proposed as an alternative to Convolutional Neural Networks
(CNNs). This paper showcases how CapsNets are more capable than CNNs for autonomous agent
exploration of realistic scenarios. In real world navigation, rewards external to agents may be rare. In
turn, reinforcement learning algorithms can struggle to form meaningful policy functions. This paper’s
approach Capsules Exploration Module (Caps-EM) pairs a CapsNets architecture with an Advantage
Actor Critic algorithm. Other approaches for navigating sparse environments require intrinsic reward
generators, such as the Intrinsic Curiosity Module (ICM) and Augmented Curiosity Modules (ACM). Caps-
EM uses a more compact architecture without need for intrinsic rewards. Tested using ViZDoom, the Caps-
EM uses 44% and 83% fewer trainable network parameters than the ICM and Depth-Augmented Curiosity
Module (D-ACM), respectively, for 1141% and 437% average time improvement over the ICM and D-
ACM, respectively, for converging to a policy function across "My Way Home" scenarios.
KEYWORDS
Neural Networks, Autonomous, Navigation, Capsules Networks
1. INTRODUCTION
Capsule Networks(CapsNets) were first presented by [1] to addressshortcomings
ofConvolutionalNeural Networks (CNNs). The max pooling operation used with CNNs reduces
thespatial size ofdata flowing through a network and thuslosesinformation. Thisleadsto the
problem that “[i]nternaldata representation of a convolutional neural network does not take into
account important spatial hierarchies between simple and complex objects" [1]. CapsNets resolve
this by encoding the probability of detection of a feature as the length of their output vector. The
state of the detected feature is encoded as the direction where that vector points. If detected
features move around an image, then the probability, or vector length, remains constant while the
vector orientation changes. The idea of a capsule resembles the design of an artificial neuron but
extends it to the vector form to enable more powerful representational capabilities. A capsule may
receive vectors from lower level capsules as an input and then performs four operations on the
input: matrix multiplication of input vectors, scalar weighting of input vectors, sum of weighted
input vectors and lastly a vector-to-vector nonlinearity. These operations are illustrated in Figure
1.
2. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
2
Figure 1. Illustration of Capsule Operations. A capsule receives input vectors u1 through un. Vector
lengths encode probabilities that lower-level capsules detected an object, while the vectors’
directions encode the state of detected objects. An affine transformation with weight matrices W1i
through Wni is applied to each vector. The weight matrices encode spatial and other relationships
between lower level features and higher ones. After multiplication, the vectors 𝑢′1 through 𝑢′𝑛
represent the predicted position of higher-level features. These vectors are multiplied by scalar
weights c1 to cn, derived using the routing algorithm, to determine which higher-level capsule a
lower level capsule’s output maps to. The k weighted input vectors 𝑢′′1 through 𝑢′′𝑛 that map to
Capsule i are then summed to form one vector. The Squash nonlinear activation function takes the
vector, forces it to length of max one, while not changing its direction, and outputs vector vi [1].
An approach called dynamic routing is the iterative method used to send lower-level capsule
outputs to higher level capsules with similar outputs. The algorithm outlines how to calculate a
network forward pass with capsules, as discussed by [1]. The method determines the vector ci,
which is all the routing weights for a given lower level capsule i. This is done for all lower level
capsules. After this, the routing algorithm looks at each higher-level capsule, such as capsule j, to
check each input and update weights in the formula. A lower level capsule tries to map its output
to the higher-level capsule whose output is most similar. A dot product gauges the similarity
between a capsule input and output. The algorithm repeats the process of matching lower level
capsule outputs to the appropriate higher-level capsule r times, where r is the number of routing
iterations.
Traditionally, reinforcement learning-based approaches for advancing autonomous agent
navigation in realistic environmentsstruggle to learn meaningful behavior. Reinforcement
learning methods, such as Advantage Actor Critic (A2C), strive to maximize the amount of
rewards that it obtains in an environment by learning an effective policy function. The rewards
may vary given the environment and desired goal. With deep reinforcement learning, neural
networks map input states to actions and are used to approximate a policy function. In
environments with plentiful rewards for an actor to interact with, a neural network can readily
update its policy function and converge to an optimal function for governing behavior. However
in instances where rewards may be lacking and sparse, a network is not able to easily update its
policy. In realistic scenarios that often have sparse rewards, a network can struggle to learn
meaningful behavior and desired skills. This paper’s approach called the Capsules Exploration
Module (Caps-EM) is compared to previous research in addressing reinforcement learning
shortcomings with exploring and navigating real world-like environments given sparse external
rewards. [2] use the Intrinsic Curiosity Module (ICM) in union with an Asynchronous Advantage
3. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
3
Actor Critic (A3C) algorithm to provide a reward signal intrinsic to an actor. This intrinsic reward
supplements extrinsic reward signals that an actor encounters within an environment. When there
are sparse external rewards, the intrinsic reward factor from the ICM still provides an agent with
rewards independent of external rewards in an environment to still stimulate learning of new
behavior and policy function updates. [3] similarly leverage intrinsic rewards with prediction of
depth images in their Depth-Augmented Curiosity Module (D-ACM) to advance autonomous
performance in sparse reward environments.
This paper demonstrates how CapsNets perform well for approximating policy functions when
paired with an A2C algorithm to significantly improve autonomous agent navigation performance
in sparse reward environments. Across a variety of test scenarios, the proposed Caps-EM uses a
small network size to improve upon the ICM and D-ACM performances, which are presented as
performance baselines. Critically relevant is the fact that Caps-EM does not incorporate the use of
intrinsic rewards, which the ICM and D-ACM approaches both use to converge to adequate
policy functions. This research highlights how strictly using external reward factors, Caps-EM
achieves a more encompassing comprehension of image inputs and abstract world representation
to achieve more meaningful action in any given scenario, which CNNs fail to replicate. While the
Caps-EM struggles in certain test environments modeling extremely sparse external rewards, the
module generalizes well across various scenarios with use of curriculum training and shows the
capabilities of CapsNets in instances of real-world scenarios. Using a self-supervised framework,
CapsNets advances autonomous system capabilities for navigation and exploration in challenging
environments that can potentially be applied to robotics and Unmanned Aerial Vehicles (UAVs)
for example.
2. RELATED WORK
Given that CapsNets are a recent development, published research on their applications is limited.
[4] integrates CapsNets with Deep-Q Learning algorithms to evaluate performance across several
different environments, including one with a task of exploring a maze. However, discussion of
the architecture used by the author is limited, and the results show that CapsNets underperform
traditional CNNs. Other work by [5] applies CapsNets to recurrent networks, and [6, 7, 8]
successfully use CapsNets for image and object classification tasks. CapsNets have also been
used for problems with autonomous driving in [9] and are effective with predicting depth for
simultaneous localization and mapping (SLAM) implementations in [10]. [11] demonstrate how
CapsNets may result in reduced neural network model training time and offer a lower number of
training parameters relative to similar CNN architectures. [12] additionally highlight how
capsules present more explainable internal operations than CNNs for better understanding of deep
learning models. This paper’s Caps-EM presents novel work on pairing CapsNets with an A2C
algorithm specifically for autonomous exploration of and navigation through environments.
[2] propose the ICM to provide a supplemental intrinsic reward factor to an agent to handle sparse
rewards in an environment. To generate this intrinsic reward, the ICM forms a prediction of the
next state and compares the prediction to the ground state value of the next state. As shown in
Figure 2, the ICM receives an action 𝑎𝑡, the current state 𝑠𝑡 and the next state 𝑠𝑡+1as inputs. 𝑎𝑡 is
the action taken by the agent to transition from 𝑠𝑡 to 𝑠𝑡+1. The current state and next state are RGB
frames of the actor’s view in VizDoom, a Doom-based platform for AI research. The ICM uses a
forward model and an inverse model to generate the intrinsic reward factor. The forward model
4. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
4
receives 𝑎𝑡 and attempts to predict an embedding of the nextstate 𝜑′𝑡+1. The error between this
prediction and the ground truth label 𝜑𝑡+1 obtained from 𝑠𝑡+1 is used as the intrinsic reward factor.
The intrinsic reward is large when an agent explores new, unseen areas, as the predicted
embedding is based on previously seen areas. This in turn rewards an agent to seek out new areas
of environment. To train the forward model, the inverse model learns to predict the action 𝑎′𝑡 that
relates the two state embeddings 𝜑𝑡 and 𝜑𝑡+1. [3] adapt the ICM and present the D-ACM that
predicts depth images instead of an action in the inverse model. They contend that predicting
depth images helps better encode 3D structural information in the embeddings. Their D-ACM
outperforms the ICM with navigation tasks and is subsequently used an additional benchmark for
comparison against the Caps-EM.
Figure 2. The ICM and D-ACM Architectures.(a) The ICM by [2] receives as inputs an action 𝑎𝑡 and
the inputs states 𝑠𝑡 and 𝑠𝑡+1, which are 42x42 RGB images. Embeddings 𝜃𝑡 and 𝜃𝑡+1 are derived
with a Convolutional Layer block. A Convolutional Layer block consists of four sets of 3x3
convolutions followed by batch normalization and an exponential linear unit (ELU) [13]. The
Inverse Model uses the embeddings to predict action 𝑎′𝑡. Through the Forward Model, 𝜃𝑡 and
label 𝑎′𝑡 are used to predict 𝜃′𝑡+1. The intrinsic reward factor r it is calculated as the difference
between 𝜃′𝑡+1. and 𝜃𝑡+1. (b) The D-ACM by [3] receives the same inputs as the ICM but then
predicts the depth images 𝐷′𝑡 and 𝐷′𝑡+1 from each frame instead of an action. The Encoder and
Decoder blocks have four layers of 32 filters with 3x3 kernels then equal numbers of filters and
kernel sized convolutions. Dashed lines represent shared weights between networks.
3. METHODOLOGY
This next section discusses the Caps-EM approach, its network architecture design and how it
operates with the A2C algorithm. Furthermore, the experiments and evaluation scenarios used to
compare the various implementations are explained here.
3.1. Advantage Actor Critic (A2C) Algorithm
In reinforcement learning, a neural network controls an agent and strives to attain a maximal
score by interacting with external reward factors in an environment. While [2] utilize an A3C
algorithm with their ICM, an A2C algorithm is used by [3] and in this paper. The intrinsic reward
5. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
5
factor generated by the ICM supplements the reward factor an agent receives from interactions
with objects in its environment. The A2C paradigm consists of a critic, measuring the quality of
an action, and an actor, which controls the agent’s behavior. As the actor takes actions, the critic
observes these actions and provides feedback. From this feedback, the actor updates its policy to
improve performance.
At each time-step 𝑡, the current state 𝑠𝑡 of the environment is given as an input to the actor and
critic. The actor, governed by the policy function (𝑠𝑡,𝑎𝑡,𝜃), with state 𝑠𝑡, action 𝑎𝑡 and network
parameters 𝜃, receives 𝑠𝑡 and outputs the action a𝑡. The policy then receives the next state 𝑠𝑡+1 as
well as the reward 𝑟𝑡+1 after the action is taken. In the evaluation environments, the actor is
limited to a discrete action space consisting of four possible actions: move forward, move left,
move right and no action. The critic, expressed as the value function 𝑞̂(𝑠𝑡,𝑎𝑡,𝑤) with parameters
𝑤, returns an estimate of the expected final total reward obtainable by the agent from the given
state. The value function 𝑉
𝑣(𝑠𝑡), with network parameters 𝑣, returns the average value of a given
state. A2C methods offer a variant for the value estimate to reduce the problem of high
variability.
The advantage function, asshown in Equation 1,with 𝛾 the discount factor to account for future
rewards losing value, indicates the improvement of a taken action over the average action taken at
the state.
𝐴 (𝑠𝑡,𝑎𝑡) = 𝑟𝑡+1 + 𝛾𝑉𝑣 (𝑠𝑡+1) − 𝑉𝑣 (𝑠𝑡) (1)
Δ𝜃 = 𝛼𝛻𝜃 (𝑙𝑜𝑔𝜋𝜃(𝑠𝑡,𝑎𝑡))𝑞̂𝑤(𝑠𝑡,𝑎𝑡) (2)
Δ𝑤 =𝛽(𝑟(𝑠𝑡,𝑎𝑡)+ 𝛾𝑞̂𝑤(𝑠𝑡+1,𝑎𝑡+1)−𝑞̂𝑤(𝑠𝑡,𝑎𝑡))𝛻𝑤𝑞̂𝑤(𝑠𝑡,𝑎𝑡) (3)
The actor and critic exist as separate models that are trained and optimized individually. The
policy update for the network parameters, or weights 𝜃, uses the q value of the critic as shown in
Equation 2. The critic updates its value parameters using the actor’s output action 𝑎𝑡+1 for the next
state 𝑠𝑡+1 in Equation 3. The hyperparameter 𝛽 controls the strength of entropy regularization.
A2C and A3C algorithms operate with the premise described above, however A3C algorithms
asynchronously execute different agents in parallel on multiple instances of the environment.
These agents update the globally shared network asynchronously [14]. With A2C the update
occurs synchronously when all agents have completed training and uses the workers’ averaged
gradients to modify the global network at one time [15]. An A2C algorithm alone cannot
converge to an optimal policy function in sparse scenarios, like those used in this paper for
evaluation. Existing approaches with A2C require the use of intrinsic reward factors, as discussed
in Figure 2. However, this paper’s approach Caps-EM can use an A2C algorithm without
supplementary intrinsic rewards to effectively explore extremely sparse scenarios.
3.2. Capsules Exploration Module (Caps-EM) Architecture
CapsNets initially appeared advantageous for the task of exploration even prior to testing. They
can discern both the probability of detection of a feature, stored in an output vector’s magnitude,
and the state of the detected feature, stored in a vector’s direction [1]. Conversely, traditional
CNNs are only able to handle the probability of detection of a feature. This difference proves
vital as CapsNets can then maintain spatial relationships of observed items in environment. This
distinction hypotheticallyenables creation of more sophisticated network embeddings of the
6. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
6
environment space. This paper demonstrates experimentation combining CapsNets with A2C
components as not previously explored in other published literature. Importantly, Caps-EM does
not use intrinsic rewards like the approaches discussed in Figure 2. The architecture
implementation of the Caps-EM is illustrated in Figure 3. It is important to note as well that using
an A2C network design exclusively incorporating CNNs, with no use of intrinsic reward signals,
cannot explore effectively. In testing, such an approach failed to learn and converge to an
effective policy function in any of the evaluation scenarios discussed in this paper.
As done by [1], the network proposed in Figure 3 uses a convolutional layer before the capsule
layers to detect basic features in the 2D RGB image inputs. The subsequent capsule layer then
uses the detected basic features to produce combinations of the features. From experimentation
with various architecture designs, using more than one convolution before the capsule layers
masks the benefits of the capsule layers by lowering the data resolution and degrading
performance.
Figure 3. Caps-EM Advantage Actor Critic Network. In the Caps-EM, the input 𝑠𝑡, a 42x42 RGB
image, first passes through a series of a 3x3 convolution, batch normalization function and ELU.
The remaining layer blocks consist of Capsule Network layers. The first lower level Primary
Capsule layer consists of a 9x9 convolution, with stride of two and padding of zero, followed by a
Squash activation function [1]. This layer has 32 input and output channels with capsule
dimension of eight. The outputs are dynamically routed to the second higher level Dense Capsule
layer consisting of 196 input capsules of dimension eight and four output capsules of dimension
16. Three routing iterations are used in the routing algorithm. Outputs of the Dense Capsule layer
are passed to an LSTM and linear layers to provide the policy function 𝜋(𝑠𝑡,𝑎𝑡,𝜃) and state value
function 𝑉𝑣(𝑠𝑡).
Additionally, using a larger network architecture with more trainable network parameters was not
found to increase the module’s performance in converging to an optimal policy function or with
being more generalizable. In fact, using a larger network architecture degrades overall
performance due to the network model needing longer to train to converge to a policy. Three
routing iterations are used between capsule layers as recommended by [1] to help prevent
overfitting the data. However in experiments, the capsule layers still displayed a tendency for
7. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
7
overfitting. In these instances, the early stopping method was used to avoid overfitting when the
network successful achieved an adequate policy function [16]. Neither did incorporating dropout
significantly improve the problem of network overfitting [17]. Dropout with p values of 0.25 and
0.5 were applied to various layers of the Caps-EM module, with the main effect only being a
slowed training rate. The architecture described in Figure 3 for Caps-EM was found to be one that
balanced the desire for a generalizable network across all evaluation scenarios with also a
minimal number of network parameters.
3.3. Evaluation Environments
Similarly as with [3], ViZDoom is used as the evaluation environment to assess the modules'
capabilities [18]. Within ViZDoom several scenarios are utilized to frame environments with
sparse external rewards. Across the various scenarios, an agent is tasked to search for a piece of
armor that serves as the sole positive external reward. A scenario restarts after an agent reaches
its goal, the armor, and receives a reward factor of +1 or after surpassing 2100 time steps. The
standard scenarios used are the "MyWayHome-v0" (MWH) setups from OpenAIGym [19].
MWH has 8 rooms with unique wall textures. Recreating the evaluation process of [3], two
additional scenarios named "My Way Home Mirrored" (MWH-M) and "My Way Home Giant"
(MWH-G) are used for testing the modules. MWH-M helps to judge how well a module's learned
network knowledge transfers between scenarios for its generalizability. The scenario consists of
the same number of rooms as MWH, however the layout is rotated. MWH-G is similar to MWH
but with 19 rooms, thus presenting a much more complex and challenging evaluation
environment. These different scenarios are shown in Figure 4.
Figure 4. ViZDoom Scenarios. Green circles at the right in the images are the target locations. For
dense settings, an agent may start at any purple circle. For sparse settings, an agent starts at the green circle
at the left of each scenario [3].
Variability is further introduced to the evaluation process using variety in reward sparsity as well
as in visual texture features [2]. To incorporate variations in scenario complexity, scenarios
exhibit either a sparse or dense structure. This targets a given scenario's reward sparsity. With the
target location remaining unchanged, the agent's beginning position is allowed to vary in the
dense case; there are 17 uniformly distributed available start positions. However with sparse, the
position is singular and placed far away from the goal location.
Regarding visual texture features of scenarios, the wall textures of an environment may either be
constant and all identical or vary between each room. These two types of setups are noted as
Uniform Texture and Varied Texture, respectively, further on. These variants are demonstrated in
Figure 5. In the context of Uniform Texture, an embedding must maintain an abstract
8. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
8
interpretation of the whole environment to effectively explore and find the target without
receiving textural cues from the environment.
Figure 5. Variability in Visual Texture Features. (a) Shows a frame from the agent's point of view
in a Uniform Texture scenario. (b) Agent's view of a Varied Texture scenario [3].
3.4. Curriculum Learning
To explore how well Caps-EM applies knowledge obtained in a successful policy function from
one scenario to another, curriculum learning is used in junction with the MWH-M scenario [20].
For this procedure, a module is trained until converged to a policy in each of the MWH Dense
Varied Texture, Dense Uniform Texture, Sparse Varied Texture and Sparse Uniform Texture
scenarios. The learned network parameter values associated with these respective scenarios are
then pre-loaded into the same module prior to beginning training again in the respective MWH-M
Dense Varied Texture, Dense Uniform Texture, Sparse Varied Texture and Sparse Uniform
Texture scenarios. In this way, the module begins training in the MWH-M environments with
prior knowledge learned in the MWH scenarios. The various modules are also allowed to train
inMWH-M without use of curriculum learning, and the results of the two different approaches are
compared.
4. RESULTS
For analyzing the Caps-EM, the ICM by [2] and the D-ACM by [3] are used as baselines for
comparison. Table 1 compares these three approaches, where the percent difference rows indicate
a module’s improved or degraded metrics relative to the ICM. The number of trainable
parameters of each module are used as a comparison metric to account for the differences in
module size and scaling. The size, or number of trainable parameters, of each module has a direct
impact on the efficiency and required time to complete the neural network model training
process. The Caps-EM architecture has substantially fewer trainable network parameters, in turn
completing a single training step more efficiently as well. The times to complete one training step
as shown are standardized values obtained from running each module variation with one worker
on a GeForce GTX 1080 GPU with 8114 MiB memory on the MWH Dense Varied Textured
scenario. Results tables displayed further on showing timing analysis are based on these
standardized times to present an equivalent metric of comparison. Plots are presented with the
mean testing score of a module relative to the number of training steps taken.
9. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
9
Table1. Model Size and Timing Comparisons.
Module # of Trainable Parameters Time to Complete 1
Training Step (ms)
ICM 915,945 13.6
D-ACM 944,170 17.6
% Difference +3% +30%
Caps-EM 515,301 6.2
% Diff. -44% -54%
As shown in Figure 6 and Table 2, the Caps-EM performs exceptionally well in the dense setup
scenarios. While the Caps-EM completes the MWH Sparse Uniform scenario in roughly the same
number of training steps as D-ACM, the Caps-EM performance is superior when considering the
actual time to converge to a policy and how Caps-EM does not use intrinsic rewards. Conversely
in the Sparse Varied Texture scenario, the Caps-EM performs worse than both the ICM and D-
ACM.
Figure 6. My Way Home Results.A minimum of five instances of a module’s performance is averaged for
the score trend line. Shaded areas around trend line indicates the one standard error range, and ovals
roughly indicate where a module converges to a policy function.
10. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
10
Table 2. My Way Home Scenario Time Performance Results. Time required to converge to an optimal
network policy is in seconds.
Module Scenario
Dense, Varied Dense,Uniform Sparse, Varied Sparse,Uniform
ICM (s) 1.46E+5 8.15E+5 5.44E+4 6.18E+5
D-ACM (s) 9.69E+4 2.64E+5 4.85E+4 3.52E+5
% Diff. -51% -209% -12% -75%
Caps-EM (s) 1.24E+4 2.48E+4 1.07E+5 1.37E+5
% Diff. -1076% -3182% +49% -353%
In order to assess how integrated intrinsic rewards with Caps-EM affects the module's
performance with exploration, Figure 7 shows a direct comparison of Caps-EM with and without
intrinsic reward in the same scenarios. The Caps-EM with 515,301 trainable parameters is 42%
smaller than Caps-EM with intrinsic rewards, referred to as Caps-EM (IR), which has 733,705
trainable parameters. Caps-EM (IR) incorporates the approach discussed in Figure 2a to generate
an intrinsic reward based on the accuracy of next state predictions. Caps-EM requires 6.21E-3
seconds to complete one training step, whereas Caps-EM (IR) takes 2.09E-2 seconds and is 236%
slower. Table 3 shows a comparison of performance with respect to time across the MWH
scenarios and that Caps-EM (IR) exhibits poorer performance in each setup. In this analysis,
intrinsic reward do not necessarily improve performance in the Sparse Varied Texture scenario.
Figure 7. My Way Home Caps-EM, with and without Intrinsic Rewards (IR), Results.
11. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
11
Table 3. My Way Home Scenario Caps-EM, with and without IR, Time Performance Results.
Module Scenario
Dense, Varied Dense,Uniform Sparse, Varied Sparse,Uniform
Caps-EM (s) 1.24E+4 2.48E+4 1.07E+5 1.37E+5
Caps-EM (IR) (s) 5.74E+4 5.22E+4 1.78E+5 2.09E+5
% Diff. +78% +52% +40% +35%
Figure 8 and Table 4 show how well each module applies learned network parameter weights
from the MWH scenarios to the MWH-M scenarios. The expectation is that the knowledge
should generalize well and enable the modules to converge to a successful policy faster than
without using curriculum learning. The Caps-EM fails to converge to a policy function in the
MWH-M Sparse Varied Texture and Sparse Uniform Texture scenarios within 1.0E+8 training
steps when not using curriculum learning. This may due to the how extremely sparse these
respective scenarios are in design, combined with how the Caps-EM lacking an intrinsic reward
signal to motivate exploration. Experiments showed that Caps-EM (IR) was able to converge to a
policy in roughly 2.5E+7 training steps in MWH-M Sparse Uniform Texture and in 3.0E+7 steps
in MWH-M Sparse Varied Texture with no curriculum training. However when using curriculum
learning in these same scenarios, the Caps-EM without intrinsic rewards performs exceptionally
well. In general, each module leverages curriculum learning to its advantage and can outperform
the no curriculum learning scores.
Figure 8. My Way Home Mirrored with Curriculum Learning Results.
12. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
12
Table 4. My Way Home Mirrored Scenario Time Performance Results, with and without Curriculum
Learning. Positive percent difference values indicate the improved runtime from curriculum training.
Figure 9 and Table 5 demonstrate each module's performance in MWH-G. None of the modules
converge to a successful policy function in the sparse scenarios variants, with or without use of
curriculum learning. The Caps-EM for example reached in excess of 12.6E+7 training steps
without any learning of a useful function. MWH-G results illustrate how the Caps-EM performs
well in dense scenario variants and significantly outperforms the ICM and D-ACM which both
depend on use of intrinsic rewards.
Figure 9. My Way Home Giant Results.
13. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
13
Table 5. My Way Home Giant Scenario Time Performance Results.
5. CONCLUSIONS
The Caps-EM architecture leverages the A2C scheme to perform well with autonomous
navigation and exploration in sparse reward environments. More compact and efficient with 44%
and 83% fewerparameters than the ICM and D-ACM, respectively, the Caps-EM on average
outperforms both the ICM and D-ACM across the MWH, MWH-M and MWH-G scenarios. The
Caps-EM converges to a policy function in MWH, on average across all four scenario variants,
437% and 1141% quicker than the D-ACM and ICM, respectively, without the use of intrinsic
rewards. Similarly in MWH-M scenarios when using curriculum learning, the Caps-EM has a
10,726% and 13,317% time improvement on average over the D-ACM and ICM, respectively.
Lastly with MWH-G variants, the Caps-EM has a 703% and 1226% time improvement on
average over the D-ACM and ICM, respectively.
While the Caps-EM struggles to converge effectively in certain sparse scenarios, such as with
MWH-M Sparse Uniform Texture and Sparse Varied Texture, the module readily applies learned
knowledge using curriculum learning to generalize well across scenarios. The intrinsic reward
factor used by the D-ACM and ICM likely enables these modules to better handle these specific
sparse scenarios. Yet, the Caps-EM (IR) module did not significantly improve performance in
these scenarios. However, these modules that produce the intrinsic signal must be trained in
addition to the A2C algorithm itself. In turn, these approaches have larger architectures with more
network parameters and lower relative performance in other scenarios. Caps-EM offers a more
lightweight yet still capable design.
The results additionally confirm the hypothesis of Caps-EM's ability to maintain better spatial
relationships and hierarchies for improved performance on average.This finding is evident in
dense scenarios where the Caps-EM maintains relationships between the Varied Texture rooms
well. Future work will explore how to improve the Caps-EM performance in the extremely sparse
environments, such as MWH-M Sparse Uniform Texture, to address this weakness.
Experimentation with Caps-EM variants that did incorporate intrinsic reward factors did improve
effectiveness in these edge cases to a degree. However, this module variant does not appear to be
a viable solution as its performance on average in other scenarios is significantly worse. This
finding arises from how the addition of intrinsicrewards requires a larger network and how any
improvements in navigation only prove applicable in limited cases. An additional area of interest
is to study how different inputs than only 2D RGB frames would affect the Caps-EM
14. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
14
performance with exploration tasks. Moreover, the scenarios used for evaluation the module are
static with no moving objects or features, which could also have an impact on performance and
be useful to investigate.
REFERENCES
[1] Sara Sabour, Nicholas Frosst, and Geoffrey H. Hinton, (2017) “Dynamic Routing Between Capsules”,
CoRR, Vol. abs/1710.09829.
[2] Deepak Pathak, Pulkit Agrawal, Alexei A. Efros, and Trevor Darnell, (2017) “Curiosity-driven
Exploration by Self-supervised Prediction”, in ICML.
[3] Juan Carvajal, Thomas Molnar, Lukasz Burzawa, and Eugenio Culurciello, (2019) “Augmented
Curiosity: Depth and Optical Flow Prediction for Efficient Exploration”, in Advances in Visual
Computing, Cham, pp. 260-271, Spring International Publishing.
[4] Per-Arne Andersen, (2018) “Deep Reinforcement Learning using Capsules in Advanced Game
Environments”, CoRR, Vol. abs/1801.09597
[5] Louis Annabi and Michaël Garcia Ortiz, (2018) “State representation learning with recurrent capsule
networks”, CoRR, Vol. abs/1812.11202.
[6] Edgar Xi, Selina Bing, and Yang Jin (2017) “Capsule Network Performance on Complex Data”,
arXiv e-prints.
[7] Fei Deng, Shengliang Pu, Xuehong Chen, Yusheng Shi, ting Yuan, and Shengyan Pu, (2018)
“Hyperspectral Image Classification with Capsule Network Using Limited Training Samples”,
Sensors(Basel, Switzerland) Vol. 18(9), 3153.
[8] Maximilian Pöpperl, Raghavendra Gulagundi, Senthil Kumar Yogamani, and Stefan Milz, (2019)
“Capsule Neural Network based Height Classification using Low-Cost Automotive Ultrasonic
Sensors”, CoRR, Vol. abs/1902.09839.
[9] Amara Dinesh Kumar, (2018) “Novel deep learning model for traffic sign detection using capsule
networks”, CoRR, Vol. abs/1805.04424.
[10] Sunil Prakash and Gaelan Gu, (2018) “Simultaneous Localization and Mapping with Depth Prediction
using Capsule Networks for UAVs”, CoRR, Vol. abs/1808.05336.
[11] Nguyen HuuPhong and Bernardete Ribeiro, (2019) “Advanced capsule networks via context
awareness”, CoRR, Vol. abs/1903.07497.
[12] AtefehShahroudnejad, Parnian Afshar, Konstantinos. N. Plataniotis And ArashMohammadi, (2018)
"Improved Explainability of Capsule Networks: Relevance Path by Agreement", 2018 IEEE Global
Conference on Signal and Information Processing (GlobalSIP), Anaheim, CA, USA, pp. 549-553.
[13] Djork-ArnéClevert, Thomas Unterthiner, and Sepp Hochreiter, (2015) “Fast and Accurate Deep
Network Learning by Exponential Linear Units (ELUs)”, CoRR, Vol. abs/1511.07289.
15. International Journal of Artificial Intelligence and Applications (IJAIA), Vol.11, No.1, January 2020
15
[14] Volodymyr Mnih, AdriáPuigdomènechBadia, Mehdi Mirza, Alex Graves, Timothy P. Lillicrp, Tim
Harley, David Silver, and KorayKavukcuoglu, (2016) “Asynchronous Methods for Deep
Reinforcement Learning”, CoRR, Vol. abs/1602.01783.
[15] Yuhuai Wu, Elman Mansimov, Shun Liao, Roger B. Grosse, and Jimmy Ba, (2017) “Scalable
trustregion method for deep reinforcement learning using Kronecker-factor approximation”, CoRR,
Vol.abs/1708.05144.
[16] YoshuaBengio (2012) “Practical recommendations for gradient-based training of deep architectures”,
CoRR, Vol. abs/1206.5533.
[17] Nitish Srivasttava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov,
(2014) “Dropout: A Simple Way to Prevent Neural Networks from Overfitting”, Journal of Machine
Learning Research, Vol. 15, pp. 1929-1958.
[18] Michal Kempka, Marek Wydmuch, Grzegorz Runc, Jakub Toczek, and WijciechJaskowski, (2016)
“ViZDoom: A Doom-based AI Research Platform for Visual Reinforcement Learning”, CoRR, Vol.
abs/1605.02097.
[19] Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and
Wojciech Zaremba, (2016) “OpenAI Gym”, CoRR, Vol. abs/1606.01540.
[20] YoshuaBengio, JéromeLouradour, Ronan Collobert, and Jason Weston, (2009) “Curriculum
Learning”, in ICML.