Face Feature Recognition System with Deep Belief Networks, for Korean/KIISE T...Mad Scientists
I submitted KIISE Thesis that <face>, 2014.
In this presentation, I present why I use deep learning to find facial features and what is limitation of before method.
Variational이라는 단어로는 아무것도 안떠오릅니다.
그래서, '꿩 대신 닭'이라고 표현해 봤습니다.
초반 독자적인 그림을 통해 개념잡기가 쉬워요.
설명부분은 초록색으로 표시했습니다.
확률변수(random variable)부터 막히면, 아래 블로그 글을 읽어 보세요.
https://blog.naver.com/nonezerok/221428251262
This document describes an implementation of backpropagation algorithm in C code to understand the backpropagation algorithm through programming. It also explains how it can be applied to a real-world case of handwritten digit recognition using TensorFlow code. The neural network code can be used to gain experience by adjusting values and observing their effects on the network.
To install Python and TensorFlow, download Python version 3.5.2 from the Python website and TensorFlow from the TensorFlow website. Python 3.5.2 must be installed before installing TensorFlow. Once Python is installed, add it to the system PATH and install pip. Then use pip to install various Python packages like iPython, PyQt5, Matplotlib, OpenCV, and Spyder for data science and machine learning.
Face Feature Recognition System with Deep Belief Networks, for Korean/KIISE T...Mad Scientists
I submitted KIISE Thesis that <face>, 2014.
In this presentation, I present why I use deep learning to find facial features and what is limitation of before method.
Variational이라는 단어로는 아무것도 안떠오릅니다.
그래서, '꿩 대신 닭'이라고 표현해 봤습니다.
초반 독자적인 그림을 통해 개념잡기가 쉬워요.
설명부분은 초록색으로 표시했습니다.
확률변수(random variable)부터 막히면, 아래 블로그 글을 읽어 보세요.
https://blog.naver.com/nonezerok/221428251262
This document describes an implementation of backpropagation algorithm in C code to understand the backpropagation algorithm through programming. It also explains how it can be applied to a real-world case of handwritten digit recognition using TensorFlow code. The neural network code can be used to gain experience by adjusting values and observing their effects on the network.
To install Python and TensorFlow, download Python version 3.5.2 from the Python website and TensorFlow from the TensorFlow website. Python 3.5.2 must be installed before installing TensorFlow. Once Python is installed, add it to the system PATH and install pip. Then use pip to install various Python packages like iPython, PyQt5, Matplotlib, OpenCV, and Spyder for data science and machine learning.
This document discusses convolutional neural networks (CNNs) for image recognition. It explains key CNN components like convolutional layers, pooling layers, hyperparameters like kernel size and stride. It provides code in TensorFlow to recognize handwritten digits from the MNIST dataset using a CNN model with convolutional and pooling layers. The code trains the model on MNIST data and evaluates test accuracy.
The document discusses interfaces in COM. It explains that interfaces allow invoking functions of COM objects indirectly using virtual function techniques. It shows how to define a pure abstract interface class and have implementation classes inherit the interface. Clients can call interface methods without knowing the underlying implementation class. Later sections discuss adding multiple interfaces to a class, querying for interfaces, and how a client can load and use a COM server from a DLL.
This document discusses the COM (Component Object Model) architecture. It covers topics such as how COM uses the Windows Registry to associate components with their class identifiers (CLSIDs) and locations, how clients can load and interact with servers through functions like CoGetClassObject and CoCreateInstance, how servers implement functions like DllRegisterServer to register themselves, and how interfaces are defined using Interface Definition Language (IDL) and compiled using MIDL. It also compares the COM architecture to web services, noting differences in technologies used like IDL vs WSDL and the Windows Registry vs UDDI for discovery.
This document discusses convolutional neural networks (CNNs) for image recognition. It explains key CNN components like convolutional layers, pooling layers, hyperparameters like kernel size and stride. It provides code in TensorFlow to recognize handwritten digits from the MNIST dataset using a CNN model with convolutional and pooling layers. The code trains the model on MNIST data and evaluates test accuracy.
The document discusses interfaces in COM. It explains that interfaces allow invoking functions of COM objects indirectly using virtual function techniques. It shows how to define a pure abstract interface class and have implementation classes inherit the interface. Clients can call interface methods without knowing the underlying implementation class. Later sections discuss adding multiple interfaces to a class, querying for interfaces, and how a client can load and use a COM server from a DLL.
This document discusses the COM (Component Object Model) architecture. It covers topics such as how COM uses the Windows Registry to associate components with their class identifiers (CLSIDs) and locations, how clients can load and interact with servers through functions like CoGetClassObject and CoCreateInstance, how servers implement functions like DllRegisterServer to register themselves, and how interfaces are defined using Interface Definition Language (IDL) and compiled using MIDL. It also compares the COM architecture to web services, noting differences in technologies used like IDL vs WSDL and the Windows Registry vs UDDI for discovery.
2. 2
인공 신경망
입력층 히든층 출력층
hidden
딥러닝은 인공신경망을 사용하는 기계학습의 한 분야
멀티레이어퍼셉트론이라 불리움
3. 3
심층 신경망; deep neural networks
2개 층 이상
딥러닝은 심층 신경망을 사용하는 기계학습의 한 분야
히든 층이 2개 이상인 인공 신경망, 다층퍼셉트론
4. 4
신경망의 역사
• Progression (1943-1960)
• First Mathematical model of neurons, Pitts & McCulloch (1943)
• Beginning of artificial neural networks–Perceptron, Rosenblatt (1958)
• Degression (1960-1980)
• Perceptron can’t even learn the XOR function
• We don’t know how to train MLP
• 1963 Backpropagation (Bryson et al.)
• Progression (1980-)
• 1986 Backpropagation reinvented
• Degression (1993-)
• SVM: Support Vector Machine is developed by Vapnik et al.[1995]
• Graphical models are becoming more and more popular
• Training deeper networks consistently yields poor results.
• However, Yann LeCun (1998) developed deep convolutional neural networks
• Progression (2006-)
• Deep Belief Networks (DBN) by Hinton et al. (2006)
• Deep Autoencoder based networks by Greedy Layer-Wise Training of Deep Networks. Bengio et al.
• Convolutional neural networks running on GPUs
• AlexNet (2012). Krizhevsky et al.
source: http://www.cs.cmu.edu/~10701/slides/Perceptron_Reading_Material.pdf
7. Neural Networks
Multi-Layer Perceptron
DBN
CNN
RNN
RBM AE
2-Layer Perceptron ~ Regression
Linear
Logistic
Softmax
Deep
Neural
Networks
GAN
Reinforcement Learning
Supervised Learning
Unsupervised Learning
7
심층학습; Deep Learning
Discriminative Model
Generative Model
심층신경망을 사용하는 기계학습 분야
신경망 자체
심층신경망을 이용한 기존/신규 기계학습 방법
8. 8
Machine Learning Data Mining
Decision Support System
Big Data
Cloud ~ Web
Artificial Intelligence
Image Processing
Computer Vision
Machine Vision
Neural Networks
Pattern Recognition
관련 연구분야
Data Science