Programme of the Master in Artificial Intelligence

This course covers Linear Algebra from basic matrix/vector operations to singular value decomposition and probabilities from fundamental basics to Markov chains and limit theorems, which are prerequires for most of the AI courses.

The course will be directed by examples and intuition rather than formalism. Python language will be used in examples and exercises. Octave (matlab) equivalent will also be available for the linear algebra part.

Although this course covers most of the basics, it is assumed students have some notion and background in linear algebra, probability and coding.

Labs will be application exercises (numeric or not) and exercises aiming at introducing aspects or notions that are not discussed in the course.

The course gives global knowledge in data structure and algorithms. It is organized in 4 parts:

1. Introduction

2. Data structures and algorithms

3. Advanced algorithms

4. Computing tools

1. Signals and Signal Processing

Classification of Signals

Simple Time-Domain Operations, Filtering, sampling

2. Discrete-Time Signals and Systems

Time-Domain Representation

Sampling Rate Alteration

3. Discrete-Time Fourier Transform

The Continuous-Time Fourier Transform

The Discrete-Time Fourier Transform

Discrete-Time Fourier Transform Theorems

Digital Processing of Continuous-Time Signals

4. Discrete-Time Systems

Discrete Time System Examples

Classification of Discrete Time Systems (FIR/IIR)

Frequency Response

5. Finite-Length Discrete Transforms

Orthogonal Transforms

The Discrete Fourier Transform

Relation Between the Fourier Transform and the DFT and Their Inverses

Circular Convolution

DFT properties and theorems

Computation of the DFT of Real Sequences

Linear Convolution Using the DFT

6. z-Transform

Computation of the Convolution Sum of Finite-Length Sequences

The Transfer Function

Transfer Function Expression, relation to frequency response

7. LTI Discrete-Time Systems in the Transform Domain

characterization of LTI systems, stability

filter design

8. DSP Algorithm Implementation

Computation of the Discrete Fourier Transform

Splines and wavelets • Multirate Filter Banks and Wavelets

The syllabus is divided into five topics each spanning two weeks. The first two topics are introductory, covering basic statistical modelling. The three subsequent ones are more involved, covering the estimation of parameters (the core of machine learning), inference (the basics of AI), and finally testing.

Topic 1: Discrete distributions

Introduction to the course, it’s teaching staff and structure

Origins and characteristics of some important discrete distributions

Introductory exercises and python notebook based labs

Topic 2: Continuous distributions

Derivation of some important continuous distributions

Introduction to Bayesian concepts including conjugacy

Conclusion of the first two “introductory” topics

Topic 3: Estimation

Bayesian estimation

Minimum mean squared error

Maximum likelihood and Maximum a-Posteriori concepts

Topic 4: Inference and priors

Bayesian inference and the predictive distribution

Prior elicitation

The non-informative (Jeffreys) prior

Beta credible interval

Topic 5: Testing

Hypothesis testing

Tests based on a normal assumption

Approximating non-normal cases

AI and the Law

AI and Data Protection

AI and Ethics

Reproducibility, What is it?

Data Organization and Evaluation

Version Control with git

Code Sharing with GitLab

Unit Testing and Continuous Integration

Documentation and Reporting

Packaging and Deployment

Linear regression

Logistic Regression

Decision Trees

Boosting

Multi-layer Perceptron

This class covers basic concepts in image and video processing as well as computer vision. Topics include image formation and sampling, image transforms, image enhancement, and image and video compression. Computer vision topics include points of interest, optical flow, and camera calibration.

• Introduction to Digital Image processing (imaging types and

formats, applications)

• Point operations, image histograms

• Spatial Filtering and convolutions

• Edge detection

• 2D Fourier Transforms and representation of images, sampling, and image resizing (low pass filters, pyramids)

• Color images and color transformations

• Interest points (detection, representation, invariance, matching, RANSAC...)

• Calibration

• Optical Flow

• Dimensionality Reduction and Clustering

• Kernel Methods and Support Vector Machines

• Graphical Models

• Exact and Approximate Inference in Bayesian Networks

• Probability Distribution Modelling

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 content

Introduction

why speech processing? speech production, speech perception, basic

phonetics

Speech signal analysis

Sampling, Quantization, Time domain processing, Frequency domain

processing, linear prediction, cepstrum, speech coding

Practical: Speech signal analysis in Octave and Praat

Machine learning for speech processing

Static classification, Sequence classification, Regression

Practical: Statistical pattern recognition, Hidden Markov models in Octave

Automatic speech recognition

Dynamic programming, Instance-based speech recognition, Hidden

Markov model-based speech recognition, Evaluation measures

Practical: Kaldi tutorial

Text-to-speech synthesis

Concatenative speech synthesis, Statistical parametric speech synthesis, Evaluation measures

Practical: Grapheme-to-phoneme conversion, HMM-based speech synthesis

Paralinguistics speech processing

Emotion, gender, accent, pathological speech assessment, Evaluation

measures

Practical: OpenSMILE tutorial

The “deep learning” course aims at providing an overview of the deep learning area (theory, methods, applications, and tools) to the students and to use the most important tools via practical sessions.

Labs content

Bias-variance dilemma through k-NN and polynomial fitting, mini deeplearning framework in numpy, PyTorch basics, MLP on MNIST and CIFAR, LeNet5 on MNIST and CIFAR, optimisation algorithms, finetuning on a pre-trained network, creation and training of model for person detection from top-view depth images, YOLOv3 for object detection, adversarial examples, distribution of activation maps as embeddings.

The module provides the student with project planning skills. The goal is to define the project(s) that will be developed in module P02. It can be divided in 6 parts:

• Familiarize with and understand your company, its corporate culture and strategy.

• Determine how the artificial intelligence takes part in your com-pany’s strategy

• Analyze state-of-the-art development of AI in the field of inter-est

• Design proof(s) of concept to guarantee the success of your project(s)

• Set a roadmap to plan your project(s) development

• Define the terms of reference of your project(s)