Slides on my talk at SfN 2018 about how signal processing and machine learning can help to model and analyse neural time series.
Slides are on purpose not too technical for an audience of neuroscientists
This document discusses periodic non-uniform sampling (PNS) for satellite communications. PNS can be used to overcome the challenges of high-rate analog-to-digital conversion in satellites, which require costly, complex and power-intensive electronics. PNS utilizes unsynchronized time-interleaved ADCs that introduce non-uniform sampling. It presents a PNS sampling scheme and reconstruction formulas that do not require synchronization. The document also describes improved PNS techniques for faster reconstruction, selective reconstruction with interference cancellation, and analytic signal reconstruction by estimating the sampling delays. PNS shifts complexity from analog to digital domains, allowing the use of lower-cost imperfect analog devices through subsequent digital processing.
This document summarizes techniques for carrier phase estimation, decision-directed loops, and timing estimation in digital communication systems. It discusses maximum likelihood estimation of phase and timing parameters based on maximizing the likelihood function. Carrier phase can be estimated using phase-locked loops or by cross-correlating the received signal with in-phase and quadrature reference carriers. Decision-directed loops allow phase estimation when the transmitted symbols are unknown by assuming symbol decisions. Timing can be estimated for PAM signals by correlating the received signal with pulse shapes centered at different time offsets and choosing the offset that maximizes the likelihood function. Diagrams show example receiver structures for carrier phase tracking and timing estimation.
Tutorial on neural vocoders at the 2021 Speech Processing Courses in Crete, "Inclusive Neural Speech Synthesis."
Presenters: Xin Wang and Junichi Yamagishi, National Institute of Informatics, Japan
The document proposes a neural source-filter waveform model (NSF) for speech synthesis. The NSF model consists of three modules: a condition module that upsamples spectral features and F0, a source module that generates a sine excitation signal, and a filter module with dilated convolutional blocks. The model is trained directly on waveforms using a spectral distance criterion in the STFT domain. Experiments show the NSF model generates high quality waveforms comparable to WaveNet, with faster generation speed. Ablation tests analyze the importance of the sine excitation source and different spectral loss terms. The NSF provides a simpler alternative to autoregressive models for neural speech synthesis.
Algorithm Portfolios for Noisy Optimization: Compare Solvers Early (LION8)Jialin LIU
"Algorithm Portfolios for Noisy Optimization: Compare Solvers Early". Marie-Liesse Cauwet, Jialin Liu and Olivier Teytaud. The 8th Learning and Intelligent OptimizatioN Conference (LION8), 2014.
These are slides used for invited tutorial on "end-to-end text-to-speech synthesis", given at IEICE SP workshop held on 27th Jan 2019.
Part 1: Neural waveform modeling
Presenters: Xin Wang, Yusuke Yasuda (National Institute of Informatics, Japan)
The document discusses analyzing algorithms through mathematical analysis and order-of-growth hypotheses. It covers estimating running time through experiments that measure time as input size increases. Doubling the input size provides the exponent in a power law model. Order-of-growth hypotheses assume algorithms follow patterns like constant, logarithmic, linear, quadratic, or cubic time as a function of input size. Precise mathematical models are difficult but approximations suffice by focusing on dominant terms.
This document discusses periodic non-uniform sampling (PNS) for satellite communications. PNS can be used to overcome the challenges of high-rate analog-to-digital conversion in satellites, which require costly, complex and power-intensive electronics. PNS utilizes unsynchronized time-interleaved ADCs that introduce non-uniform sampling. It presents a PNS sampling scheme and reconstruction formulas that do not require synchronization. The document also describes improved PNS techniques for faster reconstruction, selective reconstruction with interference cancellation, and analytic signal reconstruction by estimating the sampling delays. PNS shifts complexity from analog to digital domains, allowing the use of lower-cost imperfect analog devices through subsequent digital processing.
This document summarizes techniques for carrier phase estimation, decision-directed loops, and timing estimation in digital communication systems. It discusses maximum likelihood estimation of phase and timing parameters based on maximizing the likelihood function. Carrier phase can be estimated using phase-locked loops or by cross-correlating the received signal with in-phase and quadrature reference carriers. Decision-directed loops allow phase estimation when the transmitted symbols are unknown by assuming symbol decisions. Timing can be estimated for PAM signals by correlating the received signal with pulse shapes centered at different time offsets and choosing the offset that maximizes the likelihood function. Diagrams show example receiver structures for carrier phase tracking and timing estimation.
Tutorial on neural vocoders at the 2021 Speech Processing Courses in Crete, "Inclusive Neural Speech Synthesis."
Presenters: Xin Wang and Junichi Yamagishi, National Institute of Informatics, Japan
The document proposes a neural source-filter waveform model (NSF) for speech synthesis. The NSF model consists of three modules: a condition module that upsamples spectral features and F0, a source module that generates a sine excitation signal, and a filter module with dilated convolutional blocks. The model is trained directly on waveforms using a spectral distance criterion in the STFT domain. Experiments show the NSF model generates high quality waveforms comparable to WaveNet, with faster generation speed. Ablation tests analyze the importance of the sine excitation source and different spectral loss terms. The NSF provides a simpler alternative to autoregressive models for neural speech synthesis.
Algorithm Portfolios for Noisy Optimization: Compare Solvers Early (LION8)Jialin LIU
"Algorithm Portfolios for Noisy Optimization: Compare Solvers Early". Marie-Liesse Cauwet, Jialin Liu and Olivier Teytaud. The 8th Learning and Intelligent OptimizatioN Conference (LION8), 2014.
These are slides used for invited tutorial on "end-to-end text-to-speech synthesis", given at IEICE SP workshop held on 27th Jan 2019.
Part 1: Neural waveform modeling
Presenters: Xin Wang, Yusuke Yasuda (National Institute of Informatics, Japan)
The document discusses analyzing algorithms through mathematical analysis and order-of-growth hypotheses. It covers estimating running time through experiments that measure time as input size increases. Doubling the input size provides the exponent in a power law model. Order-of-growth hypotheses assume algorithms follow patterns like constant, logarithmic, linear, quadratic, or cubic time as a function of input size. Precise mathematical models are difficult but approximations suffice by focusing on dominant terms.
A Fast Hadamard Transform for Signals with Sub-linear SparsityRobin Scheibler
The Hadamard transform is a popular orthogonal transform with a low-complexity algorithm with O(N log N) complexity. In this presentation, we describe a new sub-linear complexity algorithm to compute the Hadamard transform of signals whose Hadamard transform coefficients are sparse - that is very few are non-zero.
This document discusses analyzing algorithms through scientific methods. It describes estimating running time, performing mathematical analysis, formulating order-of-growth hypotheses about scaling, using input models in experiments, and measuring space usage. The goal is to predict performance, compare algorithms, and avoid performance bugs through understanding theoretical performance characteristics. Key aspects covered include performing reproducible experiments, validating hypotheses by repeated testing, and analyzing results to form power law models relating running time to problem size.
This document provides an overview of signal fundamentals, including definitions, examples, and properties of signals. It discusses topics such as signal energy and power, signal transformations, periodic and exponential signals. Examples are provided to illustrate concepts such as determining if a signal has finite energy/power, applying signal transformations, decomposing signals into even and odd components, and plotting exponential signals. The document is from a university course on signal fundamentals and is intended to introduce basic signal processing concepts.
This document discusses concepts related to signals and systems. It begins by defining a signal as a time-varying quantity of information and a system as an entity that processes input signals to produce output signals. It then covers signal classification including continuous vs discrete time, analog vs digital, periodic vs aperiodic, deterministic vs random, and causal vs non-causal signals. Signal operations like time shifting, scaling, and inversion are described. Key concepts discussed in detail include signal size using energy and power, signal components and orthogonality, correlation as a measure of signal similarity, and trigonometric Fourier series. Worked examples are provided to illustrate various topics.
Fourier analysis of signals and systemsBabul Islam
This document discusses Fourier analysis of signals and linear time-invariant (LTI) systems. It defines LTI systems and explains that they are mathematically easy to analyze due to properties like superposition. Fourier analysis is used to represent signals in the frequency domain using techniques like the Fourier series for periodic signals and the Fourier transform for aperiodic signals. The frequency response of an LTI system is its output when the input is an impulse, and the output of any LTI system is the convolution of the input signal and impulse response.
Weakly-Supervised Sound Event Detection with Self-AttentionNU_I_TODALAB
IEEE ICASSP 2020
Koichi Miyazaki, Tatsuya Komatsu, Tomoki Hayashi, Shinji Watanabe, Tomoki Toda, Kazuya Takeda, Weakly-supervised sound event detection with self-attention, May 2020
Toda Laboratory, Department of Intelligent Systems, Graduate School of Informatics, Nagoya University
Instrumentation Engineering : Signals & systems, THE GATE ACADEMYklirantga
THE GATE ACADEMY's GATE Correspondence Materials consist of complete GATE syllabus in the form of booklets with theory, solved examples, model tests, formulae and questions in various levels of difficulty in all the topics of the syllabus. The material is designed in such a way that it has proven to be an ideal material in-terms of an accurate and efficient preparation for GATE.
Quick Refresher Guide : is especially developed for the students, for their quick revision of concepts preparing for GATE examination. Also get 1 All India Mock Tests with results including Rank,Percentile,detailed performance analysis and with video solutions
GATE QUESTION BANK : is a topic-wise and subject wise collection of previous year GATE questions ( 2001 – 2013). Also get 1 All India Mock Tests with results including Rank,Percentile,detailed performance analysis and with video solutions
Bangalore Head Office:
THE GATE ACADEMY
# 74, Keshava Krupa(Third floor), 30th Cross,
10th Main, Jayanagar 4th block, Bangalore- 560011
E-Mail: info@thegateacademy.com
Ph: 080-61766222
A lecture given for Stats 285 at Stanford on October 30, 2017. I discuss how OSS technology developed at Anaconda, Inc. has helped to scale Python to GPUs and Clusters.
Parameter space noise is a simple method for exploration in reinforcement learning where noise is added to the policy parameters at the start of each episode. It balances exploration and exploitation better than epsilon-greedy or bootstrapped DQN in environments requiring directed exploration like chain environments. It also outperforms action space noise in continuous control tasks with DDPG and is better than alternatives in sparse reward environments. The method is applicable to both on and off-policy algorithms and provides an orthogonal exploration technique to other advances in deep reinforcement learning.
This document provides an introduction to signals and systems. It begins by classifying different types of signals as continuous-time/discrete-time, analog/digital, deterministic/random, periodic/aperiodic, power/energy. It then discusses representations of signals in the time and frequency domains, including the Fourier series representation of periodic signals. Key concepts covered include the unit step, rectangular, triangular and sinc functions, as well as signal operations like time shifting, scaling and inversion. The document concludes by introducing Parseval's theorem relating the power of a signal to the power of its Fourier coefficients.
This document provides an overview of signals and systems. It defines key terms like signal, system, continuous and discrete time signals, analog and digital signals, periodic and aperiodic signals. It also discusses different types of signals like deterministic and probabilistic signals, energy and power signals. The document then classifies systems as linear/nonlinear, time-invariant/variant, causal/non-causal, and with/without memory. It provides examples of different signals and properties of signals like magnitude scaling, time shifting, reflection and scaling. Overall, the document introduces fundamental concepts in signals and systems.
This document chapter discusses the characterization and representation of communication signals and systems. It describes how band-pass signals and systems can be represented by equivalent low-pass signals and systems using analytic signal representations and complex envelopes. It also discusses how the response of a band-pass system to a band-pass input signal can be determined from the equivalent low-pass representations. Key topics covered include the Fourier transform, Hilbert transform, and convolution properties used to relate band-pass and low-pass signal and system representations.
1) A signal is a physical quantity that varies with respect to time, space, or other independent variables. Signals can be classified as discrete or continuous. 2) Unit impulse and unit step signals are defined for both discrete and continuous time. The discrete unit impulse is 1 at n=0 and 0 otherwise. The continuous unit impulse is 1 at t=0 and 0 otherwise. 3) Periodic signals repeat over a time period T, while aperiodic signals do not have this periodicity property. Even and odd signals satisfy certain symmetry properties when their argument is negated.
Profiling PyTorch for Efficiency & Sustainabilitygeetachauhan
From my talk at the Data & AI summit - latest update on the PyTorch Profiler and how you can use it for optimizations for efficiency. Talk also dives into the future and what we need to do together as an industry to move towards Sustainable AI
This document discusses linear time-invariant (LTI) systems and convolution. Convolution is a fundamental concept in signal processing that is used to determine the output of an LTI system given its impulse response and an input signal. The convolution of two signals is obtained by decomposing the input signal into scaled and shifted impulses, taking the scaled and shifted impulse response for each impulse, and summing them to find the overall output. Convolution amplifies or attenuates different frequency components of the input independently. It plays an important role in applications like image processing and edge detection. Examples are provided to demonstrate evaluating convolution of periodic sequences.
Python for Science and Engineering: a presentation to A*STAR and the Singapor...pythoncharmers
An introduction to Python in science and engineering.
The presentation was given by Dr Edward Schofield of Python Charmers (www.pythoncharmers.com) to A*STAR and the Singapore Computational Sciences Club in June 2011.
Missing Component Restoration for Masked Speech Signals based on Time-Domain ...NU_I_TODALAB
IEEE International Workshop on Machine Learning for Signal Processing (MLSP2017)
Nominated For Best Student Paper Award (student: Shogo Seki)
Shogo Seki, Hirokazu Kameoka, Tomoki Toda, Kazuya Takeda: Missing Component Restoration for Masked Speech Signals based on Time-Domain Spectrogram Factorization,Sep. 2017
Toda Laboratory, Department of Intelligent Systems, Graduate School of Informatics, Nagoya University
Dsp 2018 foehu - lec 10 - multi-rate digital signal processingAmr E. Mohamed
This document discusses multi-rate digital signal processing and concepts related to sampling continuous-time signals. It begins by introducing discrete-time processing of continuous signals using an ideal continuous-to-discrete converter. It then covers the Nyquist sampling theorem and relationships between continuous and discrete Fourier transforms. It discusses ideal and practical reconstruction using zero-order hold and anti-imaging filters. Finally, it introduces the concepts of downsampling and upsampling in multi-rate digital signal processing systems.
PROGRAMMA ATTIVITA’ DIDATTICA A.A. 2016/17
DOTTORATO DI RICERCA IN INGEGNERIA STRUTTURALE E GEOTECNICA
____________________________________________________________
STOCHASTIC DYNAMICS AND MONTE CARLO SIMULATION IN EARTHQUAKE ENGINEERING APPLICATIONS
Lecture Series by
Agathoklis Giaralis, Ph.D., M.ASCE., P.E. City, University of London
Visiting Professor Sapienza University of Rome
Sampling and Reconstruction (Online Learning).pptxHamzaJaved306957
1. Sampling and reconstruction of signals was analyzed using the impulse sampling math model.
2. The analysis showed that a bandlimited signal can be perfectly reconstructed from its samples as long as the sampling rate is at least twice the bandwidth of the signal.
3. If the sampling rate is lower than the minimum required rate, aliasing error occurs where frequency components fold back into the baseband.
The document describes an experiment to verify the Nyquist sampling theorem using MATLAB. It discusses sampling a continuous time signal at frequencies below, equal to, and above twice the maximum frequency of the signal. The results show aliasing when sampling below the Nyquist rate, no aliasing when sampling at the Nyquist rate, and perfect reconstruction when sampling above the Nyquist rate. The experiment generates a sinusoidal signal, samples it at different rates, and plots the discrete and reconstructed continuous signals to demonstrate the sampling theorem.
A Fast Hadamard Transform for Signals with Sub-linear SparsityRobin Scheibler
The Hadamard transform is a popular orthogonal transform with a low-complexity algorithm with O(N log N) complexity. In this presentation, we describe a new sub-linear complexity algorithm to compute the Hadamard transform of signals whose Hadamard transform coefficients are sparse - that is very few are non-zero.
This document discusses analyzing algorithms through scientific methods. It describes estimating running time, performing mathematical analysis, formulating order-of-growth hypotheses about scaling, using input models in experiments, and measuring space usage. The goal is to predict performance, compare algorithms, and avoid performance bugs through understanding theoretical performance characteristics. Key aspects covered include performing reproducible experiments, validating hypotheses by repeated testing, and analyzing results to form power law models relating running time to problem size.
This document provides an overview of signal fundamentals, including definitions, examples, and properties of signals. It discusses topics such as signal energy and power, signal transformations, periodic and exponential signals. Examples are provided to illustrate concepts such as determining if a signal has finite energy/power, applying signal transformations, decomposing signals into even and odd components, and plotting exponential signals. The document is from a university course on signal fundamentals and is intended to introduce basic signal processing concepts.
This document discusses concepts related to signals and systems. It begins by defining a signal as a time-varying quantity of information and a system as an entity that processes input signals to produce output signals. It then covers signal classification including continuous vs discrete time, analog vs digital, periodic vs aperiodic, deterministic vs random, and causal vs non-causal signals. Signal operations like time shifting, scaling, and inversion are described. Key concepts discussed in detail include signal size using energy and power, signal components and orthogonality, correlation as a measure of signal similarity, and trigonometric Fourier series. Worked examples are provided to illustrate various topics.
Fourier analysis of signals and systemsBabul Islam
This document discusses Fourier analysis of signals and linear time-invariant (LTI) systems. It defines LTI systems and explains that they are mathematically easy to analyze due to properties like superposition. Fourier analysis is used to represent signals in the frequency domain using techniques like the Fourier series for periodic signals and the Fourier transform for aperiodic signals. The frequency response of an LTI system is its output when the input is an impulse, and the output of any LTI system is the convolution of the input signal and impulse response.
Weakly-Supervised Sound Event Detection with Self-AttentionNU_I_TODALAB
IEEE ICASSP 2020
Koichi Miyazaki, Tatsuya Komatsu, Tomoki Hayashi, Shinji Watanabe, Tomoki Toda, Kazuya Takeda, Weakly-supervised sound event detection with self-attention, May 2020
Toda Laboratory, Department of Intelligent Systems, Graduate School of Informatics, Nagoya University
Instrumentation Engineering : Signals & systems, THE GATE ACADEMYklirantga
THE GATE ACADEMY's GATE Correspondence Materials consist of complete GATE syllabus in the form of booklets with theory, solved examples, model tests, formulae and questions in various levels of difficulty in all the topics of the syllabus. The material is designed in such a way that it has proven to be an ideal material in-terms of an accurate and efficient preparation for GATE.
Quick Refresher Guide : is especially developed for the students, for their quick revision of concepts preparing for GATE examination. Also get 1 All India Mock Tests with results including Rank,Percentile,detailed performance analysis and with video solutions
GATE QUESTION BANK : is a topic-wise and subject wise collection of previous year GATE questions ( 2001 – 2013). Also get 1 All India Mock Tests with results including Rank,Percentile,detailed performance analysis and with video solutions
Bangalore Head Office:
THE GATE ACADEMY
# 74, Keshava Krupa(Third floor), 30th Cross,
10th Main, Jayanagar 4th block, Bangalore- 560011
E-Mail: info@thegateacademy.com
Ph: 080-61766222
A lecture given for Stats 285 at Stanford on October 30, 2017. I discuss how OSS technology developed at Anaconda, Inc. has helped to scale Python to GPUs and Clusters.
Parameter space noise is a simple method for exploration in reinforcement learning where noise is added to the policy parameters at the start of each episode. It balances exploration and exploitation better than epsilon-greedy or bootstrapped DQN in environments requiring directed exploration like chain environments. It also outperforms action space noise in continuous control tasks with DDPG and is better than alternatives in sparse reward environments. The method is applicable to both on and off-policy algorithms and provides an orthogonal exploration technique to other advances in deep reinforcement learning.
This document provides an introduction to signals and systems. It begins by classifying different types of signals as continuous-time/discrete-time, analog/digital, deterministic/random, periodic/aperiodic, power/energy. It then discusses representations of signals in the time and frequency domains, including the Fourier series representation of periodic signals. Key concepts covered include the unit step, rectangular, triangular and sinc functions, as well as signal operations like time shifting, scaling and inversion. The document concludes by introducing Parseval's theorem relating the power of a signal to the power of its Fourier coefficients.
This document provides an overview of signals and systems. It defines key terms like signal, system, continuous and discrete time signals, analog and digital signals, periodic and aperiodic signals. It also discusses different types of signals like deterministic and probabilistic signals, energy and power signals. The document then classifies systems as linear/nonlinear, time-invariant/variant, causal/non-causal, and with/without memory. It provides examples of different signals and properties of signals like magnitude scaling, time shifting, reflection and scaling. Overall, the document introduces fundamental concepts in signals and systems.
This document chapter discusses the characterization and representation of communication signals and systems. It describes how band-pass signals and systems can be represented by equivalent low-pass signals and systems using analytic signal representations and complex envelopes. It also discusses how the response of a band-pass system to a band-pass input signal can be determined from the equivalent low-pass representations. Key topics covered include the Fourier transform, Hilbert transform, and convolution properties used to relate band-pass and low-pass signal and system representations.
1) A signal is a physical quantity that varies with respect to time, space, or other independent variables. Signals can be classified as discrete or continuous. 2) Unit impulse and unit step signals are defined for both discrete and continuous time. The discrete unit impulse is 1 at n=0 and 0 otherwise. The continuous unit impulse is 1 at t=0 and 0 otherwise. 3) Periodic signals repeat over a time period T, while aperiodic signals do not have this periodicity property. Even and odd signals satisfy certain symmetry properties when their argument is negated.
Profiling PyTorch for Efficiency & Sustainabilitygeetachauhan
From my talk at the Data & AI summit - latest update on the PyTorch Profiler and how you can use it for optimizations for efficiency. Talk also dives into the future and what we need to do together as an industry to move towards Sustainable AI
This document discusses linear time-invariant (LTI) systems and convolution. Convolution is a fundamental concept in signal processing that is used to determine the output of an LTI system given its impulse response and an input signal. The convolution of two signals is obtained by decomposing the input signal into scaled and shifted impulses, taking the scaled and shifted impulse response for each impulse, and summing them to find the overall output. Convolution amplifies or attenuates different frequency components of the input independently. It plays an important role in applications like image processing and edge detection. Examples are provided to demonstrate evaluating convolution of periodic sequences.
Python for Science and Engineering: a presentation to A*STAR and the Singapor...pythoncharmers
An introduction to Python in science and engineering.
The presentation was given by Dr Edward Schofield of Python Charmers (www.pythoncharmers.com) to A*STAR and the Singapore Computational Sciences Club in June 2011.
Missing Component Restoration for Masked Speech Signals based on Time-Domain ...NU_I_TODALAB
IEEE International Workshop on Machine Learning for Signal Processing (MLSP2017)
Nominated For Best Student Paper Award (student: Shogo Seki)
Shogo Seki, Hirokazu Kameoka, Tomoki Toda, Kazuya Takeda: Missing Component Restoration for Masked Speech Signals based on Time-Domain Spectrogram Factorization,Sep. 2017
Toda Laboratory, Department of Intelligent Systems, Graduate School of Informatics, Nagoya University
Dsp 2018 foehu - lec 10 - multi-rate digital signal processingAmr E. Mohamed
This document discusses multi-rate digital signal processing and concepts related to sampling continuous-time signals. It begins by introducing discrete-time processing of continuous signals using an ideal continuous-to-discrete converter. It then covers the Nyquist sampling theorem and relationships between continuous and discrete Fourier transforms. It discusses ideal and practical reconstruction using zero-order hold and anti-imaging filters. Finally, it introduces the concepts of downsampling and upsampling in multi-rate digital signal processing systems.
PROGRAMMA ATTIVITA’ DIDATTICA A.A. 2016/17
DOTTORATO DI RICERCA IN INGEGNERIA STRUTTURALE E GEOTECNICA
____________________________________________________________
STOCHASTIC DYNAMICS AND MONTE CARLO SIMULATION IN EARTHQUAKE ENGINEERING APPLICATIONS
Lecture Series by
Agathoklis Giaralis, Ph.D., M.ASCE., P.E. City, University of London
Visiting Professor Sapienza University of Rome
Sampling and Reconstruction (Online Learning).pptxHamzaJaved306957
1. Sampling and reconstruction of signals was analyzed using the impulse sampling math model.
2. The analysis showed that a bandlimited signal can be perfectly reconstructed from its samples as long as the sampling rate is at least twice the bandwidth of the signal.
3. If the sampling rate is lower than the minimum required rate, aliasing error occurs where frequency components fold back into the baseband.
The document describes an experiment to verify the Nyquist sampling theorem using MATLAB. It discusses sampling a continuous time signal at frequencies below, equal to, and above twice the maximum frequency of the signal. The results show aliasing when sampling below the Nyquist rate, no aliasing when sampling at the Nyquist rate, and perfect reconstruction when sampling above the Nyquist rate. The experiment generates a sinusoidal signal, samples it at different rates, and plots the discrete and reconstructed continuous signals to demonstrate the sampling theorem.
This document proposes an emotion recognition system using EEG signals and time domain analysis. Five different machine learning algorithms (RVM, MLP, DT, SVM, Bayesian) are used to classify three emotions (happy, relaxed, sad) felt by subjects viewing different videos. EEG data is collected from subjects using a headset and amplifier. Six features are extracted from the time domain signals of each EEG channel. The document describes the data collection process, feature extraction methodology, and compares the performance of different algorithms at classifying emotions based on the EEG data.
The document describes the implementation of a wideband spectrum sensing algorithm using a software-defined radio. It discusses using an energy detection based approach to sense the local frequency spectrum and determine which portions are unused. The algorithm is first tested via simulations in MATLAB using known signal parameters. It is then tested using real data collected from a Universal Software Radio Peripheral (USRP) to analyze the actual wireless spectrum.
Cooperative Spectrum Sensing Technique Based on Blind Detection MethodINFOGAIN PUBLICATION
1) The document proposes cooperative spectrum sensing techniques based on blind detection methods for cognitive radio networks. It studies eigenvalue-based and covariance-based spectrum sensing algorithms that do not require prior knowledge of the primary signal or noise.
2) The algorithms analyze the sample covariance matrix of the received signal to extract test statistics for detecting primary signal presence. Thresholds for the algorithms are determined using statistical theories to achieve desired probabilities of detection and false alarm.
3) Simulations evaluate the performance of the techniques under different conditions and signal types. Results show the proposed method has higher detection probability at low signal-to-noise ratios than maximum eigenvalue detection.
This document presents an agenda for a talk on Petri nets. It begins with an introduction to Petri nets that defines their structure, including places, transitions, tokens, and firing rules. It then discusses several analysis methods for Petri nets, including reachability trees, incidence matrices, and reduction rules. Next, it covers high-level Petri nets and colored Petri nets. The document concludes by mentioning an application of Petri nets to rumor detection and blocking in online social networks, and introduces orbital Petri nets as a promising approach.
Exploring temporal graph data with Python: a study on tensor decomposition o...André Panisson
Tensor decompositions have gained a steadily increasing popularity in data mining applications. Data sources from sensor networks and Internet-of-Things applications promise a wealth of interaction data that can be naturally represented as multidimensional structures such as tensors. For example, time-varying social networks collected from wearable proximity sensors can be represented as 3-way tensors. By representing this data as tensors, we can use tensor decomposition to extract community structures with their structural and temporal signatures.
The current standard framework for working with tensors, however, is Matlab. We will show how tensor decompositions can be carried out using Python, how to obtain latent components and how they can be interpreted, and what are some applications of this technique in the academy and industry. We will see a use case where a Python implementation of tensor decomposition is applied to a dataset that describes social interactions of people, collected using the SocioPatterns platform. This platform was deployed in different settings such as conferences, schools and hospitals, in order to support mathematical modelling and simulation of airborne infectious diseases. Tensor decomposition has been used in these scenarios to solve different types of problems: it can be used for data cleaning, where time-varying graph anomalies can be identified and removed from data; it can also be used to assess the impact of latent components in the spreading of a disease, and to devise intervention strategies that are able to reduce the number of infection cases in a school or hospital. These are just a few examples that show the potential of this technique in data mining and machine learning applications.
The document describes an experiment on linear time invariant systems. The objectives are to: 1) Convolve a signal with an impulse response, 2) Find step responses using the impulse response for rectangular, exponential and sinusoidal inputs, 3) Show stable and unstable conditions using pole-zero plots, 4) Apply filtering to an image using circular convolution with overlap add and save methods. Background topics discussed are aliasing, impulse/step inputs, and even/odd signals. Questions involve plotting a signal, its Fourier transform, and filtering an image.
ABSTRACT: In this paper, we proposed a new identification algorithm based on Kolmogorov–Zurbenko Periodogram (KZP) to separate motions in spatial motion image data. The concept of directional periodogram is utilized to sample the wave field and collect information of motion scales and directions. KZ Periodogram enables us detecting precise dominate frequency information of spatial waves covered by highly background noises. The computation of directional periodogram filters out most of the noise effects, and the procedure is robust for missing and fraud spikes caused by noise and measurement errors. This design is critical for the closure-based clustering method to find cluster structures of potential parameter solutions in the parameter space. An example based on simulation data is given to demonstrate the four steps in the procedure of this method. Related functions are implemented in our recent published R package {kzfs}.
This document discusses algorithm analysis and complexity. It introduces algorithm analysis as a way to predict and compare algorithm performance. Different algorithms for computing factorials and finding the maximum subsequence sum are presented, along with their time complexities. The importance of efficient algorithms for problems involving large datasets is discussed.
The document is a lab manual for basic simulation experiments. It contains 18 listed experiments related to signals and systems including: basic operations on matrices, generation of periodic and aperiodic signals, arithmetic operations on signals, finding even and odd parts of signals, linear convolution, autocorrelation and cross correlation. The document provides brief descriptions and MATLAB code examples for experiments related to signals and systems analysis.
The document provides an overview of fuzzy logic concepts including types of fuzzy systems, membership functions, fuzzy inference, fuzzification and defuzzification methods. It discusses knowledge-based and rule-based fuzzy systems, types of membership functions like triangular, trapezoidal and Gaussian. Examples of fuzzy logic applications in autonomous driving cars and methods for defuzzification like weighted average, centroid, max-membership and centre of sums are also summarized.
Gracheva Inessa - Fast Global Image Denoising Algorithm on the Basis of Nonst...AIST
The document proposes a fast global image denoising algorithm based on a nonstationary gamma-normal statistical model. The algorithm effectively removes Gaussian and Poisson noise while satisfying constraints on computational cost to process large datasets with minimal user input. It develops a probabilistic data model and defines the joint prior distribution, leading to a Bayesian estimate of the hidden image field. The algorithm uses a Gauss-Seidel procedure on a trellis of neighborhood graphs to iteratively find optimal hidden variable values. Experimental results show the algorithm achieves similar denoising performance to other techniques but with significantly less computation time.
An Improved Empirical Mode Decomposition Based On Particle Swarm OptimizationIJRES Journal
End effect is the main factor affecting the application of Empirical Mode Decomposition (EMD).
This paper presents an improve EMD for decomposing short signal. First, analyzing the frequency components
of signal to be decomposed, and construct the parameter equation with the amplitude and initial phase of signal
as unknowns. Second, employing particle swarm optimization (PSO) to estimate the unknown parameters, and
extending the inspected signal according to the obtained parameters. Thirdly, using EMD to decompose the
extended signal into a series of intrinsic mode functions (IMFs) and a residual. The IMFs of original signal are
extracted from these obtained IMFs. The correlation coefficients between the IMFs and the signal are calculated
to judge the pseudo-IMFs. The simulation result shows that the presented method is effective and extends the
application of EMD.
1) Machine learning techniques can be used to learn priors for solving inverse problems like image reconstruction from limited data.
2) Fully learned reconstruction is infeasible due to the large number of parameters needed. Learned post-processing and learned iterative reconstruction methods provide better results.
3) Learned iterative reconstruction formulates the problem as learning updating operators in an iterative optimization scheme, but is computationally challenging due to the need to differentiate through the whole solver. Future work includes methods to address this issue.
This document summarizes a research paper that proposes a new swarm intelligence algorithm called a Hybrid Bat Algorithm. The Hybrid Bat Algorithm combines the original Bat Algorithm with strategies from Differential Evolution. The Bat Algorithm is based on the echolocation behavior of bats and has been shown to effectively solve lower-dimensional optimization problems. However, it can struggle with higher-dimensional problems due to its tendency to converge quickly. The researchers propose hybridizing it with Differential Evolution strategies to improve its performance on higher-dimensional problems. They test the Hybrid Bat Algorithm on standard benchmark functions and find that it significantly outperforms the original Bat Algorithm.
This document summarizes simulation results for spectrum sensing using compressive sensing in cognitive radio networks. It shows that an infrastructure-less approach using a consensus algorithm can achieve detection performance close to a centralized approach, and discusses the impact of varying parameters like the number of iterations, link quality between nodes, and number of measurements. Key results include infrastructure-less achieving near-centralized detection accuracy with enough iterations or measurements, and better connectivity and higher SNR improving performance.
This document discusses speech signal processing and speech recognition. It begins by defining speech processing and its relationship to digital signal processing. It then outlines several disciplines related to speech processing including signal processing, physics, pattern recognition, and computer science. The document discusses aspects of speech signals including phonemes, the speech waveform, and spectral envelope. It covers various aspects of speech processing including pre-processing, feature extraction, and recognition. It provides details on techniques for pre-processing, feature extraction including filtering, linear predictive coding, and cepstrum. Finally, it summarizes the main steps in a speech recognition procedure including endpoint detection, framing and windowing, feature extraction, and distortion measure calculations for recognition.
Similar to SfN 2018: Machine learning and signal processing for neural oscillations (20)
This document discusses MNE software for processing MEG and EEG data in Python. MNE started as C code developed over 18 years and MNE-Python began in 2010. MNE allows users to process MEG and EEG data, apply source modeling, and visualize results. It is open source, cross-platform, and has a large community for support.
MAIN Conf Talk: Learning representations from neural signalsagramfort
The document discusses automatic sleep stage classification from polysomnography (PSG) data using deep learning methods. PSG data contains multimodal time series signals including electroencephalography (EEG), electrooculography (EOG), and electromyography (EMG). The objective is to learn a function that can classify each time point into a sleep stage (awake, REM, stage 1, stage 2, etc.) using the raw PSG signals as input. Deep neural networks have shown promising results on this task compared to traditional machine learning and signal processing methods. The document reviews recent literature on using convolutional neural networks and other deep learning approaches for sleep stage classification from EEG data.
ICML 2018 Reproducible Machine Learning - A. Gramfortagramfort
Workshop talk from:
https://mltrain.cc/events/enabling-reproducibility-in-machine-learning-mltrainrml-icml-2018/
Thoughts on the challenges of reproducibility in ML and computational sciences, and some engineering solutions based on my experience writing scikit-learn for the last 8 years.
MNE group analysis presentation @ Biomag 2016 conf.agramfort
The document discusses how to perform MEG group analysis using MNE software. It provides an overview of the MNE analysis workflow, including preprocessing raw data, source space analysis, and group-level analysis. Scripts and examples are presented for filtering, epoching, ICA, source localization, and statistical analysis of evoked responses at the sensor and source levels across multiple subjects.
Anomaly/Novelty detection with scikit-learnagramfort
This document discusses anomaly detection techniques in scikit-learn. It begins by defining anomalies and outliers, then describes the main types of anomaly detection as supervised, semi-supervised (novelty detection), and unsupervised. Popular density-based, kernel, nearest neighbors, and tree/partitioning approaches are covered. Examples are given using Gaussian mixture models, one-class SVM, local outlier factor, and isolation forest algorithms. Challenges in anomaly detection like parameter tuning and evaluation are also noted.
Paris machine learning meetup 17 Sept. 2013agramfort
This document discusses using machine learning and functional magnetic resonance imaging (fMRI) data to predict stimuli viewed by patients. Specifically, it summarizes research by Miyawaki et al. (2008) and Nishimoto et al. (2011) that used fMRI data to predict images viewed by patients, such as faces and houses, with over 50% accuracy. It also provides an example classification task using fMRI data to predict whether a patient viewed a face or house. The document states that this example prediction can be implemented in under 250 lines of code using Scikit-Learn machine learning library.
Phenomics assisted breeding in crop improvementIshaGoswami9
As the population is increasing and will reach about 9 billion upto 2050. Also due to climate change, it is difficult to meet the food requirement of such a large population. Facing the challenges presented by resource shortages, climate
change, and increasing global population, crop yield and quality need to be improved in a sustainable way over the coming decades. Genetic improvement by breeding is the best way to increase crop productivity. With the rapid progression of functional
genomics, an increasing number of crop genomes have been sequenced and dozens of genes influencing key agronomic traits have been identified. However, current genome sequence information has not been adequately exploited for understanding
the complex characteristics of multiple gene, owing to a lack of crop phenotypic data. Efficient, automatic, and accurate technologies and platforms that can capture phenotypic data that can
be linked to genomics information for crop improvement at all growth stages have become as important as genotyping. Thus,
high-throughput phenotyping has become the major bottleneck restricting crop breeding. Plant phenomics has been defined as the high-throughput, accurate acquisition and analysis of multi-dimensional phenotypes
during crop growing stages at the organism level, including the cell, tissue, organ, individual plant, plot, and field levels. With the rapid development of novel sensors, imaging technology,
and analysis methods, numerous infrastructure platforms have been developed for phenotyping.
Or: Beyond linear.
Abstract: Equivariant neural networks are neural networks that incorporate symmetries. The nonlinear activation functions in these networks result in interesting nonlinear equivariant maps between simple representations, and motivate the key player of this talk: piecewise linear representation theory.
Disclaimer: No one is perfect, so please mind that there might be mistakes and typos.
dtubbenhauer@gmail.com
Corrected slides: dtubbenhauer.com/talks.html
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptxMAGOTI ERNEST
Although Artemia has been known to man for centuries, its use as a food for the culture of larval organisms apparently began only in the 1930s, when several investigators found that it made an excellent food for newly hatched fish larvae (Litvinenko et al., 2023). As aquaculture developed in the 1960s and ‘70s, the use of Artemia also became more widespread, due both to its convenience and to its nutritional value for larval organisms (Arenas-Pardo et al., 2024). The fact that Artemia dormant cysts can be stored for long periods in cans, and then used as an off-the-shelf food requiring only 24 h of incubation makes them the most convenient, least labor-intensive, live food available for aquaculture (Sorgeloos & Roubach, 2021). The nutritional value of Artemia, especially for marine organisms, is not constant, but varies both geographically and temporally. During the last decade, however, both the causes of Artemia nutritional variability and methods to improve poorquality Artemia have been identified (Loufi et al., 2024).
Brine shrimp (Artemia spp.) are used in marine aquaculture worldwide. Annually, more than 2,000 metric tons of dry cysts are used for cultivation of fish, crustacean, and shellfish larva. Brine shrimp are important to aquaculture because newly hatched brine shrimp nauplii (larvae) provide a food source for many fish fry (Mozanzadeh et al., 2021). Culture and harvesting of brine shrimp eggs represents another aspect of the aquaculture industry. Nauplii and metanauplii of Artemia, commonly known as brine shrimp, play a crucial role in aquaculture due to their nutritional value and suitability as live feed for many aquatic species, particularly in larval stages (Sorgeloos & Roubach, 2021).
PPT on Direct Seeded Rice presented at the three-day 'Training and Validation Workshop on Modules of Climate Smart Agriculture (CSA) Technologies in South Asia' workshop on April 22, 2024.
The technology uses reclaimed CO₂ as the dyeing medium in a closed loop process. When pressurized, CO₂ becomes supercritical (SC-CO₂). In this state CO₂ has a very high solvent power, allowing the dye to dissolve easily.
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...Sérgio Sacani
Context. With a mass exceeding several 104 M⊙ and a rich and dense population of massive stars, supermassive young star clusters
represent the most massive star-forming environment that is dominated by the feedback from massive stars and gravitational interactions
among stars.
Aims. In this paper we present the Extended Westerlund 1 and 2 Open Clusters Survey (EWOCS) project, which aims to investigate
the influence of the starburst environment on the formation of stars and planets, and on the evolution of both low and high mass stars.
The primary targets of this project are Westerlund 1 and 2, the closest supermassive star clusters to the Sun.
Methods. The project is based primarily on recent observations conducted with the Chandra and JWST observatories. Specifically,
the Chandra survey of Westerlund 1 consists of 36 new ACIS-I observations, nearly co-pointed, for a total exposure time of 1 Msec.
Additionally, we included 8 archival Chandra/ACIS-S observations. This paper presents the resulting catalog of X-ray sources within
and around Westerlund 1. Sources were detected by combining various existing methods, and photon extraction and source validation
were carried out using the ACIS-Extract software.
Results. The EWOCS X-ray catalog comprises 5963 validated sources out of the 9420 initially provided to ACIS-Extract, reaching a
photon flux threshold of approximately 2 × 10−8 photons cm−2
s
−1
. The X-ray sources exhibit a highly concentrated spatial distribution,
with 1075 sources located within the central 1 arcmin. We have successfully detected X-ray emissions from 126 out of the 166 known
massive stars of the cluster, and we have collected over 71 000 photons from the magnetar CXO J164710.20-455217.
ESA/ACT Science Coffee: Diego Blas - Gravitational wave detection with orbita...Advanced-Concepts-Team
Presentation in the Science Coffee of the Advanced Concepts Team of the European Space Agency on the 07.06.2024.
Speaker: Diego Blas (IFAE/ICREA)
Title: Gravitational wave detection with orbital motion of Moon and artificial
Abstract:
In this talk I will describe some recent ideas to find gravitational waves from supermassive black holes or of primordial origin by studying their secular effect on the orbital motion of the Moon or satellites that are laser ranged.
ESR spectroscopy in liquid food and beverages.pptxPRIYANKA PATEL
With increasing population, people need to rely on packaged food stuffs. Packaging of food materials requires the preservation of food. There are various methods for the treatment of food to preserve them and irradiation treatment of food is one of them. It is the most common and the most harmless method for the food preservation as it does not alter the necessary micronutrients of food materials. Although irradiated food doesn’t cause any harm to the human health but still the quality assessment of food is required to provide consumers with necessary information about the food. ESR spectroscopy is the most sophisticated way to investigate the quality of the food and the free radicals induced during the processing of the food. ESR spin trapping technique is useful for the detection of highly unstable radicals in the food. The antioxidant capability of liquid food and beverages in mainly performed by spin trapping technique.
Current Ms word generated power point presentation covers major details about the micronuclei test. It's significance and assays to conduct it. It is used to detect the micronuclei formation inside the cells of nearly every multicellular organism. It's formation takes place during chromosomal sepration at metaphase.
SfN 2018: Machine learning and signal processing for neural oscillations
1. Alexandre Gramfort
alexandre.gramfort@inria.fr
Non-linear machine learning and
signal models reveal new insights on
neural oscillations
SfN conf. - Nov. 2018
Joint work withTom Dupré laTour, Mainak Jas,Thomas Moreau, LucilleTallot,
Laetitia Grabot,Valérie Doyère,Virginie vanWassenhove andYves Grenier
2. Non-linear Auto-Regressive Models for
Cross-Frequency Coupling (CFC) in
Neural Time Series
Signal from the striatum of a rodent
1
Code: https://pactools.github.io
T. Dupré laTour, L.Tallot, L. Grabot,V. Doyère,V. vanWassenhove,Y. Grenier,A. Gramfort,
(2017) PLOS Computational biology
3. Non-linear Auto-Regressive Models for
Cross-Frequency Coupling (CFC) in
Neural Time Series
Signal from the striatum of a rodent
1
Code: https://pactools.github.io
T. Dupré laTour, L.Tallot, L. Grabot,V. Doyère,V. vanWassenhove,Y. Grenier,A. Gramfort,
(2017) PLOS Computational biology
CFC: High frequency bursts coupled with slow waves
4. SOTA of Phase Amplitude Coupling estimation
[Dvorak & Fenton (2014).Toward a proper estimation of phase–amplitude coupling in neural oscillations]
2 Bandpass
Filters
2 Hilbert
Transform
Custom step
2 bandpass
filters
2 Hilbert
transforms
Custom
step
Comodu-
logram
5. SOTA of Phase Amplitude Coupling estimation
[Dvorak & Fenton (2014).Toward a proper estimation of phase–amplitude coupling in neural oscillations]
2 Bandpass
Filters
2 Hilbert
Transform
Custom step
2 bandpass
filters
2 Hilbert
transforms
Custom
step
Comodu-
logram
Issues:
• How do you set the filtering parameters?
• Requires Hilbert transform on broad-band signals
6. SOTA of Phase Amplitude Coupling estimation
[Dvorak & Fenton (2014).Toward a proper estimation of phase–amplitude coupling in neural oscillations]
2 Bandpass
Filters
2 Hilbert
Transform
Custom step
2 bandpass
filters
2 Hilbert
transforms
Custom
step
Comodu-
logram
Issues:
• How do you set the filtering parameters?
• Requires Hilbert transform on broad-band signals
Driven Autor-Regressive (DAR) models:
• optimize for explained variance (not CFC strength!)
• allows model selection/comparison with cross-validation
• works with shorter signals (better for time-varying CFC)
• can tell if low freq. (LF) drives high freq. (HF) or vice versa.
7. Alex Gramfort Non-linear ML for neural oscillations
Driven Auto-Regressive model
• Auto-Regressive (AR) model
4
[Parametric estimation of spectrum driven by an exogenous signal,
T. Dupré la Tour,Y. Grenier A. Gramfort, Proc. IEEE ICASSP, 2017]
[Grenier et al. IEEE Trans.Acoustics, Speech and Signal Processing, 1988]
8. Alex Gramfort Non-linear ML for neural oscillations
Driven Auto-Regressive model
• Auto-Regressive (AR) model
4
[Parametric estimation of spectrum driven by an exogenous signal,
T. Dupré la Tour,Y. Grenier A. Gramfort, Proc. IEEE ICASSP, 2017]
[Grenier et al. IEEE Trans.Acoustics, Speech and Signal Processing, 1988]
• Non-linear AR model
9. Alex Gramfort Non-linear ML for neural oscillations
Driven Auto-Regressive model
• Auto-Regressive (AR) model
4
[Parametric estimation of spectrum driven by an exogenous signal,
T. Dupré la Tour,Y. Grenier A. Gramfort, Proc. IEEE ICASSP, 2017]
[Grenier et al. IEEE Trans.Acoustics, Speech and Signal Processing, 1988]
• Non-linear AR model
AR coefficients are functions of
the amplitude of the driving signal
10. Alex Gramfort Non-linear ML for neural oscillations
Maximum Likelihood
5
regressors, it is possible to obtain an analytical expression of the pa
As the innovation "(t) is assumed to be a Gaussian white noise, t
L is obtained via:
L =
TY
t=p+1
1
q
2⇡ (t)
2
exp
"(t)
2
2 (t)
2
!
or 2 log(L) = T log(2⇡) +
TX
t=p+1
"(t)
2
(t)
2 + 2
TX
t=p+1
log(
DAR models are estimated by likelihood maximization. Here, if the
(t)
2
is considered fixed, maximizing this function boils down to min
Objective: Maximize
"(t) ⇠ N(0, (t)
2
)w.r.t. model parameters
11. Alex Gramfort Non-linear ML for neural oscillations
Maximum Likelihood
5
regressors, it is possible to obtain an analytical expression of the pa
As the innovation "(t) is assumed to be a Gaussian white noise, t
L is obtained via:
L =
TY
t=p+1
1
q
2⇡ (t)
2
exp
"(t)
2
2 (t)
2
!
or 2 log(L) = T log(2⇡) +
TX
t=p+1
"(t)
2
(t)
2 + 2
TX
t=p+1
log(
DAR models are estimated by likelihood maximization. Here, if the
(t)
2
is considered fixed, maximizing this function boils down to min
Objective: Maximize
"(t) ⇠ N(0, (t)
2
)w.r.t. model parameters
• From AR coefficients you get the power spectrum (PSD):
12. Alex Gramfort Non-linear ML for neural oscillations
Maximum Likelihood
5
regressors, it is possible to obtain an analytical expression of the pa
As the innovation "(t) is assumed to be a Gaussian white noise, t
L is obtained via:
L =
TY
t=p+1
1
q
2⇡ (t)
2
exp
"(t)
2
2 (t)
2
!
or 2 log(L) = T log(2⇡) +
TX
t=p+1
"(t)
2
(t)
2 + 2
TX
t=p+1
log(
DAR models are estimated by likelihood maximization. Here, if the
(t)
2
is considered fixed, maximizing this function boils down to min
Objective: Maximize
"(t) ⇠ N(0, (t)
2
)w.r.t. model parameters
• From AR coefficients you get the power spectrum (PSD):
PSD depends on
the driving signal
13. Alex Gramfort Non-linear ML for neural oscillations
Maximum Likelihood
5
regressors, it is possible to obtain an analytical expression of the pa
As the innovation "(t) is assumed to be a Gaussian white noise, t
L is obtained via:
L =
TY
t=p+1
1
q
2⇡ (t)
2
exp
"(t)
2
2 (t)
2
!
or 2 log(L) = T log(2⇡) +
TX
t=p+1
"(t)
2
(t)
2 + 2
TX
t=p+1
log(
DAR models are estimated by likelihood maximization. Here, if the
(t)
2
is considered fixed, maximizing this function boils down to min
Objective: Maximize
"(t) ⇠ N(0, (t)
2
)w.r.t. model parameters
• From AR coefficients you get the power spectrum (PSD):
freq
PSD
PSD depends on
the driving signal
14. Alex Gramfort Non-linear ML for neural oscillations
Maximum Likelihood
5
regressors, it is possible to obtain an analytical expression of the pa
As the innovation "(t) is assumed to be a Gaussian white noise, t
L is obtained via:
L =
TY
t=p+1
1
q
2⇡ (t)
2
exp
"(t)
2
2 (t)
2
!
or 2 log(L) = T log(2⇡) +
TX
t=p+1
"(t)
2
(t)
2 + 2
TX
t=p+1
log(
DAR models are estimated by likelihood maximization. Here, if the
(t)
2
is considered fixed, maximizing this function boils down to min
Objective: Maximize
"(t) ⇠ N(0, (t)
2
)w.r.t. model parameters
• From AR coefficients you get the power spectrum (PSD):
freq
PSD
PSD depends on
the driving signal
x1
15. Alex Gramfort Non-linear ML for neural oscillations
Maximum Likelihood
5
regressors, it is possible to obtain an analytical expression of the pa
As the innovation "(t) is assumed to be a Gaussian white noise, t
L is obtained via:
L =
TY
t=p+1
1
q
2⇡ (t)
2
exp
"(t)
2
2 (t)
2
!
or 2 log(L) = T log(2⇡) +
TX
t=p+1
"(t)
2
(t)
2 + 2
TX
t=p+1
log(
DAR models are estimated by likelihood maximization. Here, if the
(t)
2
is considered fixed, maximizing this function boils down to min
Objective: Maximize
"(t) ⇠ N(0, (t)
2
)w.r.t. model parameters
• From AR coefficients you get the power spectrum (PSD):
freq
PSD
PSD depends on
the driving signal
x0
x1
23. Alex Gramfort Non-linear ML for neural oscillations
Results
8
What does it mean for neuroscience?
• Driving signal is not that narrow band
• Driving signal has a non-symmetric spectrum
• Filtering parameters affect the neuroscientific interpretation
24. Alex Gramfort Non-linear ML for neural oscillations
Directionality and delay estimation
9
By shifting in time the driving signal one can test if
high-frequencies are preceding (“causing”) slow
waves or vice versa
25. Alex Gramfort Non-linear ML for neural oscillations
Directionality and delay estimation
9
By shifting in time the driving signal one can test if
high-frequencies are preceding (“causing”) slow
waves or vice versa
DAR models allow to ask new questions
26. Alex Gramfort Non-linear ML for neural oscillations
Robustness to short signals
10
DAR models need
shorter signals to
capture CFC
28. Convolutional Sparse Coding (CSC)
for learning the morphology of
neural signals
2
Code: https://alphacsc.github.io
Multivariate Convolutional Sparse Coding for Electromagnetic Brain Signals, (2018),
T. Dupré laTour, T. Moreau, M. Jas,A. Gramfort, Proc. NIPS Conf.
Learning the Morphology of Brain Signals Using Alpha-Stable Convolutional Sparse Coding,
(2017), M. Jas,T. Dupré laTour, U. Simsekli,A. Gramfort, Proc. NIPS Conf.
29. Alex Gramfort Non-linear ML for neural oscillations
Shape of brain rhythms matter
Advances in automating analysis of neural time series
13
[Cole andVoytek, 2017]
μ rhythm
30. Alex Gramfort Non-linear ML for neural oscillations
Shape of brain rhythms matter
Advances in automating analysis of neural time series
13
[Cole andVoytek, 2017]
μ rhythm
asymmetry
Problem of linear filtering:
Raw signal
Filtered signal
31. Alex Gramfort Non-linear ML for neural oscillations
Shape of brain rhythms matter
Advances in automating analysis of neural time series
13
[Cole andVoytek, 2017]
μ rhythm
asymmetry
Problem of linear filtering:
Raw signal
Filtered signal
After linear filtering everything
looks like a sinusoid!
32. Alex Gramfort Non-linear ML for neural oscillations
From ICA to CSC
14
https://pypi.python.org/pypi/python-picard/0.1
Independent Component Analysis (ICA)
33. Alex Gramfort Non-linear ML for neural oscillations
From ICA to CSC
14
https://pypi.python.org/pypi/python-picard/0.1
Independent Component Analysis (ICA)
X S
34. Alex Gramfort Non-linear ML for neural oscillations
From ICA…
15
https://pypi.python.org/pypi/python-picard/0.1
X
=
A S
= + +…
a1 aks1 sk
+ +…
https://pierreablin.github.io/picard/auto_examples/plot_ica_eeg.html
https://www.martinos.org/mne/stable/auto_tutorials/plot_artifacts_correction_ica.html
35. Alex Gramfort Non-linear ML for neural oscillations
…
… to CSC
16
https://pypi.python.org/pypi/python-picard/0.1
X
=
+
u1 v1
⇤
z1
uk vk
⇤
zk
convolution
36. Alex Gramfort Non-linear ML for neural oscillations
…
⇤
Topography waveform temporal activations
… to CSC
16
https://pypi.python.org/pypi/python-picard/0.1
X
=
+
u1 v1
⇤
z1
uk vk
⇤
zk
convolution
37. Alex Gramfort Non-linear ML for neural oscillations
…
⇤
Topography waveform temporal activations
… to CSC
16
https://pypi.python.org/pypi/python-picard/0.1
X
=
+
u1 v1
⇤
z1
uk vk
⇤
zk
convolution
CSC allows to learn jointly
• topography (ie. localization)
• signal waveform
• when waveform occurs
38. Alex Gramfort Non-linear ML for neural oscillations
Signal from the striatum of a rodent
CSC on LFP
17
https://pypi.python.org/pypi/python-picard/0.1
[Learning the Morphology of Brain Signals Using Alpha-Stable Convolutional Sparse Coding,
(2017), M. Jas,T. Dupré laTour, U. Simsekli,A. Gramfort, Proc. NIPS Conf.]
~80 Hz
39. Alex Gramfort Non-linear ML for neural oscillations
Signal from the striatum of a rodent
CSC on LFP
17
https://pypi.python.org/pypi/python-picard/0.1
[Learning the Morphology of Brain Signals Using Alpha-Stable Convolutional Sparse Coding,
(2017), M. Jas,T. Dupré laTour, U. Simsekli,A. Gramfort, Proc. NIPS Conf.]
~80 Hz
CSC reveals CFC
40. Alex Gramfort Non-linear ML for neural oscillations
CSC on MEG
18
https://pypi.python.org/pypi/python-picard/0.1
[Multivariate Convolutional Sparse Coding for Electromagnetic Brain Signals, (2018),
T. Dupré laTour, T. Moreau, M. Jas,A. Gramfort, Proc. NIPS Conf.]
•MEG vectorview
•Median nerve stim.
41. Alex Gramfort Non-linear ML for neural oscillations
CSC on MEG
18
https://pypi.python.org/pypi/python-picard/0.1
[Multivariate Convolutional Sparse Coding for Electromagnetic Brain Signals, (2018),
T. Dupré laTour, T. Moreau, M. Jas,A. Gramfort, Proc. NIPS Conf.]
•MEG vectorview
•Median nerve stim.
CSC reveals mu-
shaped waveforms
42. Alex Gramfort Non-linear ML for neural oscillations
CSC on MEG
18
https://pypi.python.org/pypi/python-picard/0.1
[Multivariate Convolutional Sparse Coding for Electromagnetic Brain Signals, (2018),
T. Dupré laTour, T. Moreau, M. Jas,A. Gramfort, Proc. NIPS Conf.]
•MEG vectorview
•Median nerve stim.
See the frequency
harmonics
CSC reveals mu-
shaped waveforms
47. Conclusion
• Neuroscience signals are under exploited
• Need for better models and tools
• Open source software to replicate all slides
• Need more interdisciplinary work (CS, ML,
stats, neuro, physics…)
• If you want the maths look at papers…
48. http://www.martinos.org/mne
MNE software for processing MEG and EEG data, A. Gramfort, M. Luessi, E. Larson, D. Engemann, D.
Strohmeier, C. Brodbeck, L. Parkkonen, M. Hämäläinen, Neuroimage 2013
50. Thanks !
GitHub : @agramfort Twitter : @agramfort
Support ERC SLAB,ANR THALAMEEG ANR-14-NEUC-0002-01
NIH R01 MH106174, DFG HA 2899/21-1.
http://alexandre.gramfort.netContact
T. Dupré la Tour, L.Tallot, L. Grabot,V. Doyère,V. van Wassenhove,Y. Grenier,A. Gramfort,
Non-linear Auto-Regressive Models for Cross-Frequency Coupling (CFC) in Neural
Time Series, (2017) PLOS Computational biology
T. Dupré la Tour, T. Moreau, M. Jas,A. Gramfort, Multivariate Convolutional Sparse
Coding for Electromagnetic Brain Signals, (2018), Proc. NIPS Conf.
M. Jas,T. Dupré la Tour, U. Simsekli,A. Gramfort, Learning the Morphology of Brain Signals
Using Alpha-Stable Convolutional Sparse Coding, (2017), Proc. NIPS Conf.
Tom Dupré laTour
Mainak Jas
Thomas Moreau
LucilleTallot
Laetitia Grabot
Valérie Doyère
Virginie vanWassenhove
Yves Grenier
Joint work with: