This talk attemps to motivate the use of Graph Signal Processing to analyse neuroimaging data. After introducing recent paradigm shifts in neuroimaging research (network neuroscience and principal gradients of connectivity), we present our recent work in combining GSP and machine learning, which show substantial improvements in inference based approach using simple machine learning techniques. We finally open new perspectives regarding the potential of using GSP for interpretable neuroscientific research.
Deep Learning: Recurrent Neural Network (Chapter 10) Larry Guo
This Material is an in_depth study report of Recurrent Neural Network (RNN)
Material mainly from Deep Learning Book Bible, http://www.deeplearningbook.org/
Topics: Briefing, Theory Proof, Variation, Gated RNNN Intuition. Real World Application
Application (CNN+RNN on SVHN)
Also a video (In Chinese)
https://www.youtube.com/watch?v=p6xzPqRd46w
Deep Learning: Recurrent Neural Network (Chapter 10) Larry Guo
This Material is an in_depth study report of Recurrent Neural Network (RNN)
Material mainly from Deep Learning Book Bible, http://www.deeplearningbook.org/
Topics: Briefing, Theory Proof, Variation, Gated RNNN Intuition. Real World Application
Application (CNN+RNN on SVHN)
Also a video (In Chinese)
https://www.youtube.com/watch?v=p6xzPqRd46w
Introduction to Graph Neural Networks: Basics and Applications - Katsuhiko Is...Preferred Networks
This presentation explains basic ideas of graph neural networks (GNNs) and their common applications. Primary target audiences are students, engineers and researchers who are new to GNNs but interested in using GNNs for their projects. This is a modified version of the course material for a special lecture on Data Science at Nara Institute of Science and Technology (NAIST), given by Preferred Networks researcher Katsuhiko Ishiguro, PhD.
Recurrent Neural Networks have shown to be very powerful models as they can propagate context over several time steps. Due to this they can be applied effectively for addressing several problems in Natural Language Processing, such as Language Modelling, Tagging problems, Speech Recognition etc. In this presentation we introduce the basic RNN model and discuss the vanishing gradient problem. We describe LSTM (Long Short Term Memory) and Gated Recurrent Units (GRU). We also discuss Bidirectional RNN with an example. RNN architectures can be considered as deep learning systems where the number of time steps can be considered as the depth of the network. It is also possible to build the RNN with multiple hidden layers, each having recurrent connections from the previous time steps that represent the abstraction both in time and space.
This Edureka Recurrent Neural Networks tutorial will help you in understanding why we need Recurrent Neural Networks (RNN) and what exactly it is. It also explains few issues with training a Recurrent Neural Network and how to overcome those challenges using LSTMs. The last section includes a use-case of LSTM to predict the next word using a sample short story
Below are the topics covered in this tutorial:
1. Why Not Feedforward Networks?
2. What Are Recurrent Neural Networks?
3. Training A Recurrent Neural Network
4. Issues With Recurrent Neural Networks - Vanishing And Exploding Gradient
5. Long Short-Term Memory Networks (LSTMs)
6. LSTM Use-Case
Evaluating Graph Signal Processing for Neuroimaging Through Classification an...Nicolas Farrugia
Graph Signal Processing (GSP) is a promising framework to analyze multi-dimensional neuroimaging datasets, while taking into account both the spatial and functional dependencies between brain signals. In the present work, we apply dimensionality reduction techniques based on graph representations of the brain to decode brain activity from real and simulated fMRI datasets. We introduce seven graphs obtained from a) geometric structure and/or b) functional connectivity between brain areas at rest, and compare them when performing dimension reduction for classification. We show that mixed graphs using both a) and b) offer the best performance. We also show that graph sampling methods perform better than classical dimension reduction including Principal Component Analysis (PCA) and Independent Component Analysis (ICA).
Introduction to Graph Neural Networks: Basics and Applications - Katsuhiko Is...Preferred Networks
This presentation explains basic ideas of graph neural networks (GNNs) and their common applications. Primary target audiences are students, engineers and researchers who are new to GNNs but interested in using GNNs for their projects. This is a modified version of the course material for a special lecture on Data Science at Nara Institute of Science and Technology (NAIST), given by Preferred Networks researcher Katsuhiko Ishiguro, PhD.
Recurrent Neural Networks have shown to be very powerful models as they can propagate context over several time steps. Due to this they can be applied effectively for addressing several problems in Natural Language Processing, such as Language Modelling, Tagging problems, Speech Recognition etc. In this presentation we introduce the basic RNN model and discuss the vanishing gradient problem. We describe LSTM (Long Short Term Memory) and Gated Recurrent Units (GRU). We also discuss Bidirectional RNN with an example. RNN architectures can be considered as deep learning systems where the number of time steps can be considered as the depth of the network. It is also possible to build the RNN with multiple hidden layers, each having recurrent connections from the previous time steps that represent the abstraction both in time and space.
This Edureka Recurrent Neural Networks tutorial will help you in understanding why we need Recurrent Neural Networks (RNN) and what exactly it is. It also explains few issues with training a Recurrent Neural Network and how to overcome those challenges using LSTMs. The last section includes a use-case of LSTM to predict the next word using a sample short story
Below are the topics covered in this tutorial:
1. Why Not Feedforward Networks?
2. What Are Recurrent Neural Networks?
3. Training A Recurrent Neural Network
4. Issues With Recurrent Neural Networks - Vanishing And Exploding Gradient
5. Long Short-Term Memory Networks (LSTMs)
6. LSTM Use-Case
Evaluating Graph Signal Processing for Neuroimaging Through Classification an...Nicolas Farrugia
Graph Signal Processing (GSP) is a promising framework to analyze multi-dimensional neuroimaging datasets, while taking into account both the spatial and functional dependencies between brain signals. In the present work, we apply dimensionality reduction techniques based on graph representations of the brain to decode brain activity from real and simulated fMRI datasets. We introduce seven graphs obtained from a) geometric structure and/or b) functional connectivity between brain areas at rest, and compare them when performing dimension reduction for classification. We show that mixed graphs using both a) and b) offer the best performance. We also show that graph sampling methods perform better than classical dimension reduction including Principal Component Analysis (PCA) and Independent Component Analysis (ICA).
Human activity recognition with self-attentionIJECEIAES
In this paper, a self-attention based neural network architecture to address human activity recognition is proposed. The dataset used was collected using smartphone. The contribution of this paper is using a multi-layer multi-head self-attention neural network architecture for human activity recognition and compared to two strong baseline architectures, which are convolutional neural network (CNN) and long-short term network (LSTM). The dropout rate, positional encoding and scaling factor are also been investigated to find the best model. The results show that proposed model achieves a test accuracy of 91.75%, which is a comparable result when compared to both the baseline models.
Enhanced signal detection slgorithm using trained neural network for cognitiv...IJECEIAES
Over the past few years, Cognitive Radio has become an important research area in the field of wireless communications. It can play an important role in dynamic spectrum management and interference identification. There are many spectrum sensing techniques proposed in literature for cognitive radio, but all those techniques detect only presence or absence of the primary user in the designated band and do not give any information about the used modulation scheme. In certain applications, in cognitive radio receiver, it is necessary to identify the modulation type of the signal so that the receiver parameters can be adjusted accordingly. Most of the modulated signals exhibit the property of Cyclostationarity that can be used for the purpose of correct detection of primary user and the modulation type. In this paper, we have proposed an enhanced signal detection algorithm for cognitive radio receiver which makes use of cyclostationarity property of the modulated signal to exactly detect, the modulation type of the received signal using a trained neural network. The algorithm gives better accuracy of signal detection even in low SNR conditions. The use of a trained neural network makes it more flexible and extendible for future applications
Master Thesis Presentation (Subselection of Topics)Alina Leidinger
This presentation shows some of my work carried out as part of my master thesis on "Mathematical Analysis of Neural Networks" at TUM Chair of Applied Numerical Analysis under Prof. Dr. Massimo Fornasier. The thesis constitutes a literature review with the aim of analysing and contrasting some of the approaches in the mathematical analysis of neural networks. The thesis focuses on 3 key aspects: Modern and classical approximation theory, robustness and stability of neural networks and unique identification of network weights. While the three themes carry approximately equal weight in the thesis, this presentation gives only a very short overview over the first and third chapter of my thesis and focuses on the robustness chapter. See also the full text version available on SlideShare/LinkedIn.
Photoplethysmogram signal reconstruction through integrated compression sensi...IAESIJAI
The transmission of photoplethysmogram (PPG) signals in real-time is extremely challenging and facilitates the use of an internet of things (IoT) environment for healthcare- monitoring. This paper proposes an approach for PPG signal reconstruction through integrated compression sensing and basis function aware shallow learning (CSBSL). Integrated-CSBSL approach for combined compression of PPG signals via multiple channels thereby improving the reconstruction accuracy for the PPG signals essential in healthcare monitoring. An optimal basis function aware shallow learning procedure is employed on PPG signals with prior initialization; this is further fine-tuned by utilizing the knowledge of various other channels, which exploit the further sparsity of the PPG signals. The proposed method for learning combined with PPG signals retains the knowledge of spatial and temporal correlation. The proposed Integrated-CSBSL approach consists of two steps, in the first step the shallow learning based on basis function is carried out through training the PPG signals. The proposed method is evaluated using multichannel PPG signal reconstruction, which potentially benefits clinical applications through PPG monitoring and diagnosis.
Feature Extraction and Classification of NIRS DataPritam Mondal
A thesis paper submitted to the department of Electronics and Communication
Engineering, Khulna University of Engineering & Technology, Khulna, Bangladesh, in
partial fulfillment of the requirement for the degree of “Bachelor of Science” in Electronics
and Communication Engineering
Neural networks have gained a great deal of importance in the area of soft computing and are widely used in making predictions. The work presented in this paper is about the development of Artificial Neural Network (ANN) based models for the prediction of sugarcane yield in India. The ANN models have been experimented using different partitions of training patterns and different combinations of ANN parameters.
Experiments have also been conducted for different number of neurons in hidden layer and the algorithms for ANN training. For this work, data has been obtained from the website of Directorate of Economics and Statistics, Ministry of Agriculture, Government of India. In this work, the experiments have been conducted for 2160 different ANN models. The least Root Mean Square Error (RMSE) value that could be achieved on
test data was 4.03%. This has been achieved when the data was partitioned in such a way that there were 10% records in the test data, 10 neurons in hidden layer, learning rate was 0.001, the error goal was set to 0.01 and traincgb algorithm in MATLAB was used for ANN training.
Artificial neural networks (ANN) consider classification as one of the most dynamic research and
application areas. ANN is the branch of Artificial Intelligence (AI). The neural network was trained by
back propagation algorithm. The different combinations of functions and its effect while using ANN as a
classifier is studied and the correctness of these functions are analyzed for various kinds of datasets. The
back propagation neural network (BPNN) can be used as a highly successful tool for dataset classification
with suitable combination of training, learning and transfer functions. When the maximum likelihood
method was compared with backpropagation neural network method, the BPNN was more accurate than
maximum likelihood method. A high predictive ability with stable and well functioning BPNN is possible.
Multilayer feed-forward neural network algorithm is also used for classification. However BPNN proves to
be more effective than other classification algorithms.
Model of Differential Equation for Genetic Algorithm with Neural Network (GAN...Sarvesh Kumar
The work is carried on the application of differential equation (DE) and its computational technique of genetic algorithm and neural (GANN) in C#, which is frequently used in globalised world by human wings. Diagrammatical and flow chart presentation is the major concerned for easy undertaking of these two concepts with indication of its present and future application is the new initiative taken in this paper along with computational approaches in C#. Little observation has been also pointed during working, functioning and development process of above algorithm in C# under given boundary value condition of DE for genetic and neural. Operations of fitness function and Genetic operations were completed for behavioural transmission of chromosome.
SCCAI- A Student Career Counselling Artificial Intelligencevivatechijri
As education is growing day by day, the competition has prompted a need for the student to
understand more about the educational field. Many times the counselor isn’t available all the time and
sometimes due to the lack of proper knowledge about some educational field. Due to this, it creates an issue of
misconception of that field. This creates a problem for the student to decide a proper educational trajectory and
guidance is not always useful. The proposed paper will overcome all these problem using machine learning
algorithm. Various algorithms are being considered and amongst them the best suitable for our project are used
here. There are 3 major problems that come across our path and they are solved using Random forest, Linear
regression and Searching algorithm using Google API. At first Searching algorithm solves the problem of
location by segregating the college’s location vice, then Random Forest provides the list of colleges by using
stream and range of percentage and finally Linear Regression predicts the current cutoff using previous years’
data. Rather than this, the proposed system also provides information regarding all fields of education helping
students to understand and know about their field of interest better. The following idea is a total fresh idea with
no existing projects of similar kind. This project will help students guide them throughout.
UNet-VGG16 with transfer learning for MRI-based brain tumor segmentationTELKOMNIKA JOURNAL
A brain tumor is one of a deadly disease that needs high accuracy in its medical surgery. Brain tumor detection can be done through magnetic resonance imaging (MRI). Image segmentation for the MRI brain tumor aims to separate the tumor area (as the region of interest or ROI) with a healthy brain and provide a clear boundary of the tumor. This study classifies the ROI and non-ROI using fully convolutional network with new architecture, namely UNet-VGG16. This model or architecture is a hybrid of U-Net and VGG16 with transfer Learning to simplify the U-Net architecture. This method has a high accuracy of about 96.1% in the learning dataset. The validation is done by calculating the correct classification ratio (CCR) to comparing the segmentation result with the ground truth. The CCR value shows that this UNet-VGG16 could recognize the brain tumor area with a mean of CCR value is about 95.69%.
An Artificial Intelligence Approach to Ultra High Frequency Path Loss Modelli...ijtsrd
This study proposes Artificial Intelligence AI based path loss prediction models for the suburban areas of Abuja, Nigeria. The AI based models were created on the bases of two deep learning networks, namely the Adaptive Neuro Fuzzy Inference System ANFIS and the Generalized Radial Basis Function Neural network RBF NN . These prediction models were created, trained, validated and tested for path loss prediction using path loss data recorded at 1800MHz from multiple Base Transceiver Stations BTSs distributed across the areas under investigation. Results indicate that the ANFIS and RBF NN based models with Root Mean Squared Error RMSE values of 5.30dB and 5.31dB respectively, offer greater prediction accuracy over the widely used empirical COST 231 Hata, which has an RMSE of 8.18dB. Deme C. Abraham ""An Artificial Intelligence Approach to Ultra-High Frequency Path Loss Modelling of the Suburban Areas of Abuja, Nigeria"" 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/ijtsrd30227.pdf
Paper Url : https://www.ijtsrd.com/computer-science/artificial-intelligence/30227/an-artificial-intelligence-approach-to-ultra-high-frequency-path-loss-modelling-of-the-suburban-areas-of-abuja-nigeria/deme-c-abraham
International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
Nutraceutical market, scope and growth: Herbal drug technologyLokesh Patil
As consumer awareness of health and wellness rises, the nutraceutical market—which includes goods like functional meals, drinks, and dietary supplements that provide health advantages beyond basic nutrition—is growing significantly. As healthcare expenses rise, the population ages, and people want natural and preventative health solutions more and more, this industry is increasing quickly. Further driving market expansion are product formulation innovations and the use of cutting-edge technology for customized nutrition. With its worldwide reach, the nutraceutical industry is expected to keep growing and provide significant chances for research and investment in a number of categories, including vitamins, minerals, probiotics, and herbal supplements.
THE IMPORTANCE OF MARTIAN ATMOSPHERE SAMPLE RETURN.Sérgio Sacani
The return of a sample of near-surface atmosphere from Mars would facilitate answers to several first-order science questions surrounding the formation and evolution of the planet. One of the important aspects of terrestrial planet formation in general is the role that primary atmospheres played in influencing the chemistry and structure of the planets and their antecedents. Studies of the martian atmosphere can be used to investigate the role of a primary atmosphere in its history. Atmosphere samples would also inform our understanding of the near-surface chemistry of the planet, and ultimately the prospects for life. High-precision isotopic analyses of constituent gases are needed to address these questions, requiring that the analyses are made on returned samples rather than in situ.
Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...Sérgio Sacani
Since volcanic activity was first discovered on Io from Voyager images in 1979, changes
on Io’s surface have been monitored from both spacecraft and ground-based telescopes.
Here, we present the highest spatial resolution images of Io ever obtained from a groundbased telescope. These images, acquired by the SHARK-VIS instrument on the Large
Binocular Telescope, show evidence of a major resurfacing event on Io’s trailing hemisphere. When compared to the most recent spacecraft images, the SHARK-VIS images
show that a plume deposit from a powerful eruption at Pillan Patera has covered part
of the long-lived Pele plume deposit. Although this type of resurfacing event may be common on Io, few have been detected due to the rarity of spacecraft visits and the previously low spatial resolution available from Earth-based telescopes. The SHARK-VIS instrument ushers in a new era of high resolution imaging of Io’s surface using adaptive
optics at visible wavelengths.
Cancer cell metabolism: special Reference to Lactate PathwayAADYARAJPANDEY1
Normal Cell Metabolism:
Cellular respiration describes the series of steps that cells use to break down sugar and other chemicals to get the energy we need to function.
Energy is stored in the bonds of glucose and when glucose is broken down, much of that energy is released.
Cell utilize energy in the form of ATP.
The first step of respiration is called glycolysis. In a series of steps, glycolysis breaks glucose into two smaller molecules - a chemical called pyruvate. A small amount of ATP is formed during this process.
Most healthy cells continue the breakdown in a second process, called the Kreb's cycle. The Kreb's cycle allows cells to “burn” the pyruvates made in glycolysis to get more ATP.
The last step in the breakdown of glucose is called oxidative phosphorylation (Ox-Phos).
It takes place in specialized cell structures called mitochondria. This process produces a large amount of ATP. Importantly, cells need oxygen to complete oxidative phosphorylation.
If a cell completes only glycolysis, only 2 molecules of ATP are made per glucose. However, if the cell completes the entire respiration process (glycolysis - Kreb's - oxidative phosphorylation), about 36 molecules of ATP are created, giving it much more energy to use.
IN CANCER CELL:
Unlike healthy cells that "burn" the entire molecule of sugar to capture a large amount of energy as ATP, cancer cells are wasteful.
Cancer cells only partially break down sugar molecules. They overuse the first step of respiration, glycolysis. They frequently do not complete the second step, oxidative phosphorylation.
This results in only 2 molecules of ATP per each glucose molecule instead of the 36 or so ATPs healthy cells gain. As a result, cancer cells need to use a lot more sugar molecules to get enough energy to survive.
Unlike healthy cells that "burn" the entire molecule of sugar to capture a large amount of energy as ATP, cancer cells are wasteful.
Cancer cells only partially break down sugar molecules. They overuse the first step of respiration, glycolysis. They frequently do not complete the second step, oxidative phosphorylation.
This results in only 2 molecules of ATP per each glucose molecule instead of the 36 or so ATPs healthy cells gain. As a result, cancer cells need to use a lot more sugar molecules to get enough energy to survive.
introduction to WARBERG PHENOMENA:
WARBURG EFFECT Usually, cancer cells are highly glycolytic (glucose addiction) and take up more glucose than do normal cells from outside.
Otto Heinrich Warburg (; 8 October 1883 – 1 August 1970) In 1931 was awarded the Nobel Prize in Physiology for his "discovery of the nature and mode of action of the respiratory enzyme.
WARNBURG EFFECT : cancer cells under aerobic (well-oxygenated) conditions to metabolize glucose to lactate (aerobic glycolysis) is known as the Warburg effect. Warburg made the observation that tumor slices consume glucose and secrete lactate at a higher rate than normal tissues.
Seminar of U.V. Spectroscopy by SAMIR PANDASAMIR PANDA
Spectroscopy is a branch of science dealing the study of interaction of electromagnetic radiation with matter.
Ultraviolet-visible spectroscopy refers to absorption spectroscopy or reflect spectroscopy in the UV-VIS spectral region.
Ultraviolet-visible spectroscopy is an analytical method that can measure the amount of light received by the analyte.
Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...Ana Luísa Pinho
Functional Magnetic Resonance Imaging (fMRI) provides means to characterize brain activations in response to behavior. However, cognitive neuroscience has been limited to group-level effects referring to the performance of specific tasks. To obtain the functional profile of elementary cognitive mechanisms, the combination of brain responses to many tasks is required. Yet, to date, both structural atlases and parcellation-based activations do not fully account for cognitive function and still present several limitations. Further, they do not adapt overall to individual characteristics. In this talk, I will give an account of deep-behavioral phenotyping strategies, namely data-driven methods in large task-fMRI datasets, to optimize functional brain-data collection and improve inference of effects-of-interest related to mental processes. Key to this approach is the employment of fast multi-functional paradigms rich on features that can be well parametrized and, consequently, facilitate the creation of psycho-physiological constructs to be modelled with imaging data. Particular emphasis will be given to music stimuli when studying high-order cognitive mechanisms, due to their ecological nature and quality to enable complex behavior compounded by discrete entities. I will also discuss how deep-behavioral phenotyping and individualized models applied to neuroimaging data can better account for the subject-specific organization of domain-general cognitive systems in the human brain. Finally, the accumulation of functional brain signatures brings the possibility to clarify relationships among tasks and create a univocal link between brain systems and mental functions through: (1) the development of ontologies proposing an organization of cognitive processes; and (2) brain-network taxonomies describing functional specialization. To this end, tools to improve commensurability in cognitive science are necessary, such as public repositories, ontology-based platforms and automated meta-analysis tools. I will thus discuss some brain-atlasing resources currently under development, and their applicability in cognitive as well as clinical neuroscience.
Graph Signal Processing: an interpretable framework to link neurocognitive architectures with brain activity
1. Graph Signal Processing: an interpretable
framework to link neurocognitive architectures
with brain activity
Nicolas Farrugia
June 4th, 2019
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 1 / 26
2. Prelude
Signal processing in Neuroscience
Interpreting (oscillatory) brain activity using Fourier or Wavelet
transforms.
Network Neuroscience
Applying graph theory metrics to study functional and structural
connectivity.
Graph Signal Processing for Cognitive Neurosciences
The best of both worlds ?
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 2 / 26
3. Prelude
Signal processing in Neuroscience
Interpreting (oscillatory) brain activity using Fourier or Wavelet
transforms.
Network Neuroscience
Applying graph theory metrics to study functional and structural
connectivity.
Graph Signal Processing for Cognitive Neurosciences
The best of both worlds ?
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 2 / 26
4. Prelude
Signal processing in Neuroscience
Interpreting (oscillatory) brain activity using Fourier or Wavelet
transforms.
Network Neuroscience
Applying graph theory metrics to study functional and structural
connectivity.
Graph Signal Processing for Cognitive Neurosciences
The best of both worlds ?
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 2 / 26
5. 1 Motivations
Functional role of brain oscillations
From brain regions to network neuroscience
Networks and connectome harmonics / gradients
2 Graph Signal Processing for Neuroimaging
Introduction to Graph Signal Processing
Inference using GSP derived measures
Predictive approaches using GSP derived feature vectors
GSP and interpretability
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 3 / 26
6. Spectral (time-frequency) analysis in cognitive
neuroscience
Hans Berger and early measurements of electro-encephalography
-> Observations on alpha (7.5 to 12.5 Hz) and beta (12.5 to 25 Hz)
frequency bands (and later, gamma frequency band from 25 to 40 Hz)
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 4 / 26
7. Spectral (time-frequency) analysis in cognitive
neuroscience
Fast-forward to 21st century...
Functional role of delta-theta phase in sensory prediction.
Arnal and Giraud, 2012, Trends in Cognitive Sciences
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 4 / 26
8. Spectral (time-frequency) analysis in cognitive
neuroscience
Fast-forward to 21st century...
Phase and cross-frequency coupling:
Sauseng et al. 2008, Neuroscience and Biobehavioral reviews
See also : non-sinusoidal oscillations (Cole and Woytek, 2017, TICS)
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 4 / 26
9. 1 Motivations
Functional role of brain oscillations
From brain regions to network neuroscience
Networks and connectome harmonics / gradients
2 Graph Signal Processing for Neuroimaging
Introduction to Graph Signal Processing
Inference using GSP derived measures
Predictive approaches using GSP derived feature vectors
GSP and interpretability
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 5 / 26
10. From regions to networks...
Primary visual cortex
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 6 / 26
11. From regions to networks...
Coactivation of brain areas - networks
Resting-state networks - spontaneous and task activity.
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 6 / 26
12. The network neuroscience revolution
Fornito et al. 2015, Nat. Neuro.
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 7 / 26
13. From networks to spectral graph analysis
Connectome harmonics - spectral analysis of structural connectivity.
Atasoy et al. 2016, Nat. Comms
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 8 / 26
14. From networks to spectral graph analysis
Connectivity gradients - Spectral analysis of functional connectivity.
Margulies et al. 2016, PNAS
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 8 / 26
15. From networks to spectral graph analysis
Functional interpretation of connectivity gradients.
Margulies et al. 2016, PNAS
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 8 / 26
16. 1 Motivations
Functional role of brain oscillations
From brain regions to network neuroscience
Networks and connectome harmonics / gradients
2 Graph Signal Processing for Neuroimaging
Introduction to Graph Signal Processing
Inference using GSP derived measures
Predictive approaches using GSP derived feature vectors
GSP and interpretability
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 9 / 26
17. What is Graph Signal Processing ?
GSP in four steps
Define a graph with N nodes,
Compute Graph Laplacian L (e.g. L = D − A) and find its
eigenvectors U
Define a signal x with N values, one per node.
Graph Fourier transform is defined by ˆx = U x
Analogy with (Discrete) Fourier Transform
Shuman et al. 2013, IEEE Signal Processing Magazine
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 10 / 26
18. Why using GSP for Neuroimaging ?
A straightforward use case
Brain signals on graphs built on structural (white matter) connectivity.
Huang et al. 2018
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 11 / 26
19. Why using GSP for Neuroimaging ?
A straightforward use case
Brain signals on graphs built on structural (white matter) connectivity.
Elegance
Network-centric tools adapted to the problem
Applicable to both inference-based and predictive approaches
Interpretations in term of graph frequencies
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 11 / 26
20. Why using GSP for Neuroimaging ?
A straightforward use case
Brain signals on graphs built on structural (white matter) connectivity.
Elegance
Network-centric tools adapted to the problem
Applicable to both inference-based and predictive approaches
Interpretations in term of graph frequencies
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 11 / 26
21. 1 Motivations
Functional role of brain oscillations
From brain regions to network neuroscience
Networks and connectome harmonics / gradients
2 Graph Signal Processing for Neuroimaging
Introduction to Graph Signal Processing
Inference using GSP derived measures
Predictive approaches using GSP derived feature vectors
GSP and interpretability
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 12 / 26
22. Inference using GSP metrics
What measures are authors currently using ?
Useful measures on graph signals
Smoothness (x Lx), also called Dirichlet energy,
Mutual information / F measure of GFT decompositions,
Concentration in certain frequency bands (Alignement / Liberality)
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 13 / 26
23. Inference using GSP metrics
Huang et al. 2018
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 13 / 26
24. Inference using GSP metrics
Statistical methods : group-wise inference
Dependent variables derived from GSP statistically related to
behavior.
Smoothness / Dirichlet energy (Smith et al. 2017)
Alignement and Liberality related to attention switching (Huang et
al., 2018)
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 13 / 26
25. Inference using GSP metrics
Resting state networks predicted by connectome harmonics.
Atasoy et al. 2016, Nat. Comms
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 13 / 26
26. 1 Motivations
Functional role of brain oscillations
From brain regions to network neuroscience
Networks and connectome harmonics / gradients
2 Graph Signal Processing for Neuroimaging
Introduction to Graph Signal Processing
Inference using GSP derived measures
Predictive approaches using GSP derived feature vectors
GSP and interpretability
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 14 / 26
27. Predictive approach using GSP
Proposed approach
Graph nodes defined by a brain parcellation
Graph edges defined by functional or structural connectivity
Feature Extraction on brain activity using GSP (e.g. GFT or Graph
Wavelets)
Feature selection
Supervised learning (using an interpretable method)
Ménoret, Farrugia, Pasdeloup and Gripon, 2017, GlobalSIP
Brahim and Farrugia, 2019, GRETSI
Pilavci and Farrugia, 2019, ICASSP
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 15 / 26
28. Predictive approach using GSP
Proposed approach
Graph nodes defined by a brain parcellation
Graph edges defined by functional or structural connectivity
Feature Extraction on brain activity using GSP (e.g. GFT or Graph
Wavelets)
Feature selection
Supervised learning (using an interpretable method)
Ménoret, Farrugia, Pasdeloup and Gripon, 2017, GlobalSIP
Brahim and Farrugia, 2019, GRETSI
Pilavci and Farrugia, 2019, ICASSP
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 15 / 26
29. Predictive approach using GSP
Proposed approach
Graph nodes defined by a brain parcellation
Graph edges defined by functional or structural connectivity
Feature Extraction on brain activity using GSP (e.g. GFT or Graph
Wavelets)
Feature selection
Supervised learning (using an interpretable method)
Ménoret, Farrugia, Pasdeloup and Gripon, 2017, GlobalSIP
Brahim and Farrugia, 2019, GRETSI
Pilavci and Farrugia, 2019, ICASSP
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 15 / 26
30. Predictive approach using GSP
Proposed approach
Graph nodes defined by a brain parcellation
Graph edges defined by functional or structural connectivity
Feature Extraction on brain activity using GSP (e.g. GFT or Graph
Wavelets)
Feature selection
Supervised learning (using an interpretable method)
Ménoret, Farrugia, Pasdeloup and Gripon, 2017, GlobalSIP
Brahim and Farrugia, 2019, GRETSI
Pilavci and Farrugia, 2019, ICASSP
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 15 / 26
31. Graph Fourier Transform on resting state data
Brahim and Farrugia, submitted
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 16 / 26
32. Graph Fourier Transform on resting state data
Resuls on ADHD200 dataset.
5 10 15 20 25 30 35 40 45
Features
0.0
0.2
0.4
0.6
0.8
1.0
Accuracy
STD+SG
STD
FC
EC
CC
NS
Mean
Mean+SG
5 10 15 20 25 30 35 40 45
Features
0.0
0.5
1.0
1.5
2.0
2.5
-log(P)
1.0 2.5
Brahim and Farrugia, submitted
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 16 / 26
33. Graph Fourier Transform on resting state data
ADHD-200
Approaches Acc (%) Sen (%) Spe (%)
STD 52.5±0.037 55±0.043 55±0.066
STD+GSP 85±0.019 85±0.036 85±0.036
FC 65±0.032 60±0.059 70±0.063
ABIDE NYU Langone
Approaches Acc (%) Sen (%) Spe (%)
STD 66.65±0.007 63.75±0.015 74.44±0.010
STD+GSP 70.36±0.007 67.68±0.011 75.67±0.012
FC 62.83±0.006 56.61±0.014 70.22±0.015
Table: Maximum classification rates of the different approaches for
ADHD-200 and NYU databases using SVM classifier with linear kernel (max
± standard error of the mean).
Brahim and Farrugia, submitted
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 16 / 26
34. Graph Fourier Transform on resting state data
Feature Selection (results from ABIDE NYU Langone)
Top: Ratios of selected ROIs across 10 folds, using STD (Accuracy 66.65±0.007 ).
Bottom : Average of selected GFT modes across 10 folds, using STD + GFT (Accuracy 70.36±0.007)
Brahim and Farrugia, submitted
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 16 / 26
35. Spectral Graph Wavelet Transform (SGWT)
Wavelet atoms are obtained by:
ψs,a = ΣN
n=1g(sλn )ˆδ(n)un (1)
where g(.) is the kernel function, s is the scale, ˆδ is the graph Fourier
transform of dirac delta and un are the columns of U.
SGWT of f is obtained using the inner product Wf (s, a) =< ψs,a , f >.
Optimizing spectral coverage using Warped Translate kernel
Pilavci and Farrugia, 2019, ICASSP
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 17 / 26
36. Spectral Graph Wavelet Transform (SGWT)
We tested this approach using an averaged functional connectivity
graph parcellated on 523 regions of BASC Atlas
Usings signals from :
A synthetic regression problem on a brain graph
fMRI data from Chang et al., predicting affective rating of
emotional pictures
Pilavci and Farrugia, 2019, ICASSPIMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 17 / 26
37. Spectral Graph Wavelet Transform (SGWT)
Pilavci and Farrugia, 2019, ICASSP
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 17 / 26
38. 1 Motivations
Functional role of brain oscillations
From brain regions to network neuroscience
Networks and connectome harmonics / gradients
2 Graph Signal Processing for Neuroimaging
Introduction to Graph Signal Processing
Inference using GSP derived measures
Predictive approaches using GSP derived feature vectors
GSP and interpretability
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 18 / 26
39. GSP and Interpretability
Inference
Generate and test hypothesis on graph frequency bands (similar
to M/EEG oscillations)
Spatial contributions of Graph Fourier modes
Prediction / data-driven
Visualize GSP features or contributions of Grapˆh Fourier modes
Use of interpretable ML methods with GSP features as inputs
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 19 / 26
40. GSP and Interpretability
Inference
Generate and test hypothesis on graph frequency bands (similar
to M/EEG oscillations)
Spatial contributions of Graph Fourier modes
Prediction / data-driven
Visualize GSP features or contributions of Grapˆh Fourier modes
Use of interpretable ML methods with GSP features as inputs
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 19 / 26
41. Generating new hypothesis using GSP (1/2)
Hypothesis using Frequencies
What would such an hypothesis look like ?
In the field of brain oscillations, empirical knowledge has been
built using frequency bands.
Should we start to make hypothesis on graph frequencies ?
Measures of (temporal) excursions from liberality or alignement
(Huang et al. 2018) - linking temporal frequencies and graph
frequencies
Can we directly make hypothesis on Graph fourier modes ?
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 20 / 26
42. Generating new hypothesis using GSP (1/2)
Hypothesis using Frequencies
What would such an hypothesis look like ?
In the field of brain oscillations, empirical knowledge has been
built using frequency bands.
Should we start to make hypothesis on graph frequencies ?
Measures of (temporal) excursions from liberality or alignement
(Huang et al. 2018) - linking temporal frequencies and graph
frequencies
Can we directly make hypothesis on Graph fourier modes ?
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 20 / 26
43. Generating new hypothesis using GSP (1/2)
Hypothesis using Frequencies
What would such an hypothesis look like ?
In the field of brain oscillations, empirical knowledge has been
built using frequency bands.
Should we start to make hypothesis on graph frequencies ?
Measures of (temporal) excursions from liberality or alignement
(Huang et al. 2018) - linking temporal frequencies and graph
frequencies
Can we directly make hypothesis on Graph fourier modes ?
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 20 / 26
44. Generating new hypothesis using GSP (2/2)
Hypothesis using Harmonics
In previous slides we have shown that propagation (diffusion)
patterns:
Were functionally meaningful,
Could predict spontaneous activity.
Analysis of experimental data could exploit decompositions based on
Harmonics
Could we exploit (connectome) harmonics to understand
spatiotemporal dynamics ?
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 21 / 26
45. Generating new hypothesis using GSP (2/2)
Hypothesis using Harmonics
In previous slides we have shown that propagation (diffusion)
patterns:
Were functionally meaningful,
Could predict spontaneous activity.
Analysis of experimental data could exploit decompositions based on
Harmonics
Could we exploit (connectome) harmonics to understand
spatiotemporal dynamics ?
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 21 / 26
46. Generating new hypothesis using GSP (2/2)
Hypothesis using Harmonics
In previous slides we have shown that propagation (diffusion)
patterns:
Were functionally meaningful,
Could predict spontaneous activity.
Analysis of experimental data could exploit decompositions based on
Harmonics
Could we exploit (connectome) harmonics to understand
spatiotemporal dynamics ?
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 21 / 26
47. What is missing: Dynamic (hyper?) graphs
Estimating a graph measure at each time point may not be valid
(Basset and Sizemore 2017.
Ongoing debate on time(?)-varying functional connectivity using sliding
windows(e.g. Hindriks et al. 2016)
Building dynamic graph and associated measures using slepians ?
(Huang et al. 2018), Preti and Van De Ville 2017
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 22 / 26
48. What is missing: Dynamic (hyper?) graphs
Estimating a graph measure at each time point may not be valid
(Basset and Sizemore 2017.
Ongoing debate on time(?)-varying functional connectivity using sliding
windows(e.g. Hindriks et al. 2016)
Building dynamic graph and associated measures using slepians ?
(Huang et al. 2018), Preti and Van De Ville 2017
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 22 / 26
49. What is missing: Dynamic (hyper?) graphs
Estimating a graph measure at each time point may not be valid
(Basset and Sizemore 2017.
Ongoing debate on time(?)-varying functional connectivity using sliding
windows(e.g. Hindriks et al. 2016)
Building dynamic graph and associated measures using slepians ?
(Huang et al. 2018), Preti and Van De Ville 2017
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 22 / 26
50. What is missing: Dynamic (hyper?) graphs
Estimating a graph measure at each time point may not be valid
(Basset and Sizemore 2017.
Ongoing debate on time(?)-varying functional connectivity using sliding
windows(e.g. Hindriks et al. 2016)
Building dynamic graph and associated measures using slepians ?
(Huang et al. 2018), Preti and Van De Ville 2017
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 22 / 26
51. Conclusions and perspectives
The promising part:
Connectome harmonics are relevant,
Certain graph frequency may be useful to generate new
hypothesis,
GSP may help levarage explanatory power in small datasets.
The difficult part...
Brain signals (oscillations) are highly non-linear and
non-stationnary,
Dynamic graphs are needed to integrate both the complex spatial
dynamics and non-stationary aspects,
Unique opportunity of collaborative research between Neuroscience
and AI!
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 23 / 26
52. Conclusions and perspectives
The promising part:
Connectome harmonics are relevant,
Certain graph frequency may be useful to generate new
hypothesis,
GSP may help levarage explanatory power in small datasets.
The difficult part...
Brain signals (oscillations) are highly non-linear and
non-stationnary,
Dynamic graphs are needed to integrate both the complex spatial
dynamics and non-stationary aspects,
Unique opportunity of collaborative research between Neuroscience
and AI!
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 23 / 26
53. Conclusions and perspectives
The promising part:
Connectome harmonics are relevant,
Certain graph frequency may be useful to generate new
hypothesis,
GSP may help levarage explanatory power in small datasets.
The difficult part...
Brain signals (oscillations) are highly non-linear and
non-stationnary,
Dynamic graphs are needed to integrate both the complex spatial
dynamics and non-stationary aspects,
Unique opportunity of collaborative research between Neuroscience
and AI!
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 23 / 26
54. Thanks!
Questions?
Thanks to Vincent Gripon, Abdelbasset Brahim, Yusuf Yigit Pilavci,
Mathilde Ménoret and Bastien Pasdeloup.
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 24 / 26
55. Prediction and Inference
"Inference typically focuses on isolating variables deemed
individually important above some chance level, often based on p
values."
"Prediction commonly aims at identifying variable sets that
together enable accurate guessing of outcomes based on other or
future data."
Bzdok and Ioannidis. 2019, Trends in Neurosciences
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 25 / 26
56. Graph Fourier Transform on task fMRI data
Feature Extraction using GSP
Supervised learning with dimensionality reduction using Graph
Fourier Transform and Graph Sampling (Menoret et al. 2017)
Table: Comparison of Graph Sampling (Semilocal graph), PCA, ICA and
ANOVA. Classification accuracy with 50 components for the Simulated and
Haxby datasets.
Method
Simulation Haxby
Easy Difficult Face-House Cat-Face
PCA 88.8% 65.5% 82.7% 63.6%
ICA 90.2% 65.3% 84.4% 67.0%
ANOVA 92.1% 67.3% 85.5% 65.5%
Graph sampling 90.9% 72.5% 88.2% 69.0%
Ménoret, Farrugia, Pasdeloup and Gripon, 2017, GlobalSIP
IMT-Atlantique GSP and interpretability in Neuroimaging June 4th, 2019 26 / 26