Master AI courses#
M01 Linear Algebra And Probability#
Course M01-1#
Review on vector and matrix and geometric interpretations
Review on matrix operations
Solving linear equation: Ax = b
Elimination, permutation, inverse matrix
LU decomposition
Course M01-2#
Vector space and fundamental subspaces
Rank, solution with null space component
Independence, basis and dimensions
Course M01-3#
Orthogonality and projection
Least squares
Orthogonal basis
Course M01-4#
Determinant
Eigenvalue decomposition (example of Markov matrices)
Symmetric and positive definite matrices (Covariance matrix)
Course M01-5#
Singular value decomposition and pseudo inverse
Linear transform and change of basis: Fourier and wavelet
Weighted least square and statistics
Course M01-6#
Sets, events, probability theory, axioms and derived laws
Conditional probability related rules and theorems (including Bayes)
Independence
Counting and combinatorics
Course M01-7#
Random variable (RV) discrete and continuous
Function of RV
Expectation, variance
Course M01-8#
Normal distribution
CDF and derived distribution
Joint distributions, multivariate normal and covariance
Course M01-9#
Markov chains, steady state, absorption
Law of large number
Central limit theorem
M02 Data Structure and Algorithms for AI#
Course M02-1#
Introduction
Course M02-2#
Abstract Datastructure
Sorting
Course M02-3#
Datastructure Implementation
Course M02-4#
Graphs
Course M02-5#
Graphs Algorithms
Course M02-6#
Dynamic programming
Randomized Algorithms
Viterbi Algorithm
Course M02-7#
Monte carlo search tree
Course M02-8#
A* (A-Star) algorithm
M03 Signal Processing#
This course is based on this book: Signals and Systems: Theory and Applications, by Fawwaz T. Ulaby and Andrew E. Yagle
Course M03-1#
Introduction to Signal Processing
Course M03-2#
Linear Time Invariant System and Convolutions
Course M03-3#
Fourier transform
Course M03-4#
Applications of the Fourier transform
Course M03-5#
Sampling, discrete signals and discrete convolutions
Course M03-6#
Discrete filtering, the Z-transform and frequency response
Course M03-7#
Discrete filtering applications
Course M03-8#
Multi-rate DSP
Course M03-9#
Introduction to Wavelets
Course M03-10#
Review
M04 Foundations in Statistics#
Course M04-1#
Introduction
Course M04-2-3#
Discrete distributions
Goals:
Master elementary calculations with discrete distributions
Gain understanding about the notions of statistical estimator and confidence intervals
Understand the effect of summation on independent random variables and get an intuition of the central limit theorem
Perform statistical simulation and estimation experiments on a real data set
Master the definition and basic properties of the Poisson distribution
Understand the law of small numbers and related conditions of applicability
Encounter a first example of Maximum Likelihood Estimation
Learn that and how one can simulate from the Poisson
Course M04-4#
Continuous univariate distribution
Goals:
Review and manipulate classical univariate continuous distributions
Calculate moments, probabilities and related quantities based on PDFs
Learn some approaches to simulate univariate continuous distributions
Get introduced to Bayesian statistics and the notion of conjugate prior
Course M04-5#
Parametric statistical estimation
Goals:
Understand and be in position to implement Maximum Likelihood Estimation and other estimation approaches in the context of parametric univariate continuous distributions
Reinforce knowledge and understanding of Bayesian statistics in the context of parametric univariate continuous distributions
Get introduced to selected advanced results on asymptotic distributions of extimators (particularly towards asymptotic normality in the MLE case)
Course M04-6#
Bayesian estimation
Goals:
Understand and be in position to implement basic Bayesian approaches in the context of parametric univariate continuous distributions.
Perform a comparison between selected frequentist and Bayesian inference approaches
Course M04-7#
Basic principles of statistical hypothesis testing
Goals:
Introduction to statistical testing and more specifically deriving and applying t-tests in one- and two-sample cases
Getting and insight on further standard families of tests (such as chi2-tests or the Kolmogorov-Smirnov test)
Course M04-8#
Complements on two-sample tests and overview of speech synthesis testing
Goals:
Deeper understanding of tests.
Acquiring basic knowledge of Bayesian hypothesis testing (if time allows)
Conducting testing in a practical context with data stemming from experiments dedicated to comparing tow speech synthesis methods
M05 Open Science and Ethics#
Course M05-1#
Introduction - AI and the Law
Course M05-2#
AI and Data Protection
Course M05-3#
AI and Ethics
Course M05-4#
Reproducibility, What is it?
Course M05-5#
Data and Workflow Management
Course M05-6#
Version Control with Git
Course M05-7#
Code sharing with GitHub
Course M05-8#
Unit Testing and Continuous Integration
Course M05-9#
Documentation and Reporting
Course M05-10#
Packaging, Deployment and Licensing
M06 Fundamentals in machine learning 1#
Course M06-1#
Introduction to Pattern Recognition
Linear Regression
Course M06-2#
Logistic Regression
Course M06-3#
Decision Trees
Course M06-4#
Boosting Theory
Course M06-5#
Multi-Layer Perceptron
M07 Introduction to image processing and computer vision#
Course M07-1#
Introduction and point operations
Course M07-2#
Histogram Equalization
Spatial Filters
Course M07-3#
Image representation
Edges detection
Course M07-4#
2D Fourier Transform
Course M07-5#
Colors and Compressions
Course M07-6#
Interest Points and Descriptors
Course M07-7#
Calibration
Course M07-8#
Motion Analysis
M08 Fundamentals in machine learning 2#
Course M08-1#
Dimensionality Reduction and Clustering Theory
Course M08-2#
Kernel Methods and Support Vector Machines Theory
Course M08-3#
Graphical Models Theory
Course M08-4#
Exact and Approximate Inference in Bayesian Networks (sampling, variational) Theory
Course M08-5#
Probability Distribution Modelling (EM/GMM) Theory
M09 Introduction to Speech Processing#
This course will introduce the students the fundamentals of speech processing and provide them with the key formalisms, models and algorithms to implement speech processing applications such as, speech recognition, speech synthesis, paralinguistic speech processing, multichannel speech processing.
Course M09-1#
Introduction, fundamentals of speech processing
Why speech processing ?
Speech production, perception
Basic phonetics
Course M09-2#
Speech signal analysis
Sampling, quantization
Time domain processing, frequency domain processing
Linear prediction, cepstrum, speech coding
Introduction to Digital Speech Processing by Lawrence R. Rabiner and Ronald W. Schafer
Course M09-3#
Machine learning for speech processing
Static classification
Sequence classification
Regression
Hidden Markov models
Course M09-4#
Automatic speech recognition 1
Dynamic programming
Instance-based speech recognition
Hidden Markov model-based speech recognition
Evaluation measures
Course M09-5#
Automatic speech recognition 2
The Application of Hidden Markov Models in Speech Recognition
Tutorial on HMM and their applications in speech recognition
Evaluating the performance of speech technology systems
Connectionist speech recognition, a hybrid approach by Hervé Bourlard
Course M09-6#
Text to speech synthesis 1
NLP for Speech Synthesis
Articulatory speech synthesis
Formant speech synthesis
Concatenative speech synthesis
Course M09-7#
Text to speech synthesis 2
Statistical parametric speech synthesis
Hybrid speech synthesis
End-to-end speech synthesis
Evaluation
Course M09-8#
Paralinguistics
M10 Deep Learning#
This course is based on the one from François Fleuret and Andrew Ng. This latter can be found at https://fleuret.org/dlc/ with slides and videos.
A summary of the importants points of the course is available in here: summary.pdf
Course M10-1#
Introduction
From neural networks to deep learning.
Current applications and success.
What is really happening?
Tensor basics and linear regression.
High dimension tensors.
Tensor internals.
Machine learning fundamentals
Loss and risk.
Over and under fitting.
Bias-variance dilemma.
Proper evaluation protocols.
Basic clusterings and embeddings.
Course M10-2#
Multi-layer perceptron and back-propagation
The perceptron.
Probabilistic view of a linear classifier.
Linear separability and feature design.
Multi-Layer Perceptrons.
Gradient descent.
Back-propagation.
Course M10-3#
Graphs of operators, autograd
DAG networks.
Autograd.
PyTorch modules and batch processing.
Convolutions.
Pooling.
Writing a PyTorch module.
Course M10-4#
Convolution neural networks
Convolutions.
Pooling.
Writing a PyTorch module.
Course M10-5#
Initialization and optimization
Cross-entropy loss.
Stochastic gradient descent.
PyTorch optimizers.
L2 and L1 penalties.
Parameter initialization.
Architecture choice and training protocol.
Writing an autograd function.
Course M10-6#
Going deeper
Benefits of depth.
Rectifiers.
Dropout.
Batch normalization.
Residual networks.
Using GPUs.
Course M10-7#
Autoencoders
Transposed convolutions.
Autoencoders.
Denoising autoencoders.
Variational autoencoders.
Course M10-8#
Computer vision
Computer vision tasks.
Networks for image classification.
Networks for object detection.
Networks for semantic segmentation.
DataLoader and neuro-surgery.
Course M10-9#
Under the hood
Looking at parameters.
Looking at activations.
Visualizing the processing in the input.
Optimizing inputs.
Course M10-10#
Generative models
Auto-regression.
Causal convolutions.
Non-volume preserving networks.
Course M10-11#
Generative adversarial models
Generative Adversarial Networks.
Wasserstein GAN.
Conditional GAN and image translation.
Model persistence and checkpoints.
Course M10-12#
Recurrent models and NLP
Recurrent Neural Networks.
LSTM and GRU.
Word embeddings and translation.
Course M10-13#
Attention models
Attention for Memory and Sequence Translation.
Attention Mechanisms.
Transformer Networks.
A01 Biometrics#
This module aims to provide an introduction to Biometrics and an understanding of the main biometric modalities (such as face or fingerprint) and associated recognition algorithms. The module also addresses the security and privacy preservation aspects of biometrics.
Course A01-1#
Introduction to Biometrics
Course A01-2#
Evaluation of Biometric Systems
Course A01-3#
Iris Recognition
Course A01-4#
Face Recognition
Course A01-5#
Vascular Recognition
Course A01-6#
Fingerprint Recognition
Course A01-7#
Multibiometrics
Course A01-8#
Speaker Recognition
Course A01-9#
Presentation Attack Detection
Course A01-10#
Biometric Template Protection
A02 Multimodal Computational Sensing of People#
Multimodal processing is at the core of the perception of our world: we see, we hear, we touch, we smell, we taste, and we move. Being able to analyse and combine multimodal streams of data is therefore an inherent ability that should be addressed by artificial intelligence systems, and comprises several core challenges like how to represent multimodal information? how to deal with modality asynchrony? how to fuse complementary or redundant information? how to do co-training of models?
The course will provide an overview of this topic, with some emphasis on the analysis of people and of their behaviors from multimodal sensor streams (with a bias towards vision and audio). We will rely on Bayesian statistical models and deep learning as main modeling tools.
Course A02-1#
Introduction
Course A02-2#
Audio Processing
Course A02-3#
Visual Processing
Course A02-4#
Visual Tracking
Course A02-5#
Multimodal Processing
Dubbing detection
Attention and gaze modeling
A03 Natural Language Processing#
The Natural Language Processing module introduces the basic NLP tasks and the main deep learning models currently being applied to them. We will cover learning about words, sentences, and translations between languages, using representation learning, recurrent neural networks, and attention-based models.
Course A03-1#
Introduction
Course A03-2#
Representation learning for word embeddings
Course A03-3#
Recurrent neural networks for text
Course A03-4#
Sequence to sequence models for machine translation
Course A03-5#
Attention based models of texts
A04 Robotics#
The Robotics Module aims at discovering the recent trends and challenges of AI in robotics, with a specific focus on robot skill acquisition from demonstration, adaptive control, and human-robot collaboration.
With this Module, the students will acquire an overview of applications in robotics requiring the use of AI and associated adaptive control problems. The Module will exploit a large range of models, techniques and algorithms that have already been demonstrated and practiced in previous Modules. Notably, it will rely on notions of multivariate Gaussian distributions, linear regression, weighted least squares, Fourier transforms, Jacobians, gradient descent and constrained optimization. As outcomes, the students will be familiar with the recent trends and challenges of AI in robotics (including practical challenges), robot skill learning from demonstration and exploration, adaptive control, and human-robot collaboration.
Additional lecture : Calinon_MMchapter2019.pdf
Course A04-1#
Introduction
A brief history of robotics and autonomous machines
Learning from demonstration (observational learning, kinesthetic teaching, correspondence problems)
Tools for AI in robotics (simulators, ROS middleware)
Course A04-2#
Movement primitives
Multivariate Gaussian distributions
Newton’s method
Decomposition as a superposition of basis functions
Bézier curves and Bernstein polynomials
Course A04-3#
Movement primitives 2
Weighted least squares (recap)
Locally weighted regression
Probabilistic movement primitives
Dynamical movement primitives
Course A04-4#
Operational space control
Forward kinematics
Inverse kinematics
Task prioritization and nullspace control
Course A04-5#
Human robot collaboration
Linear dynamical systems
Torque control, dynamic model (inertia, …), gravity compensation
Impedance control
Course A04-6#
Anticipation and planning
Optimal control
Linear quadratic regulator (LQR)
Linear quadratic tracking (LQT)
Course A04-7#
Anticipation and planning 2
Iterative LQR (iLQR)
Differential dynamic programming (DDP)
Model predictive control (MPC)
Course A04-8#
Ergodic control for exploration behaviour
Exploration behaviors
Decomposition as Fourier series
Spatial coverage problems
Course A04-9#
Manifolds in robotics
Representations of orientation data
Quaternions
Riemannian geometry