Neurocomputing

Author

Affiliation

Abstract

This website contains the materials for the module Neurocomputing, covering basics in machine learning, deep learning and neuro-AI.

Lectures

You will find below the links to the slides for each lecture (html and pdf).

1 - Introduction

	Slides
1.1 - Introduction Introduction to the main concepts of reinforcement learning and showcasing of the current applications.	html, pdf
1.2 - (optional) Basics in math Mathematical background necessary to follow this course.	html, pdf
1.3 - Neurons Quick journey from biological neurons to artificial neurons.	html, pdf

2 - Linear algorithms

	Slides
2.1 - Optimization Overview of gradient descent and regularization.	html, pdf
2.2 - Linear regression Linear regression, multiple linear regression, logistic regression, polynomial regression and how to evaluate them.	html, pdf
2.3 - Linear classification Hard linear classification, Maximum Likelihood Estimation, Soft linear classication, multi-class softmax classification.	html, pdf
2.4 - Learning theory Vapnik-Chervonenkis dimension, Cover’s theorem, feature spaces and the kernel methods..	html, pdf

3 - Deep learning

	Slides
3.1 - Feedforward neural networks Basic neural network aka Multi-layer perceptrons (MLP), and the almighty backpropagation algorithm.	html, pdf
3.2 - Modern neural networks Advanced methods for training neural networks: optimizers, activation functions, normalization, etc.	html, pdf
3.3 - Convolutional neural networks CNNs like AlexNet and its followers (VGG, ResNet, Inception) started the deep learning hype and revolutionized computer vision..	html, pdf
3.4 - Object detection Object detection networks (R-CNN, YOLO, SSD) are able to locate objects in an image.	html, pdf
3.5 - Segmentation network Segmentation networks (U-Net) can tell which pixels belong to an object.	html, pdf
3.6 - Autoencoders Autoencoders and variational autoencoders (VAE) can be used to extract latent representations from raw data.	html, pdf
3.7 - Restricted Boltzmann machines RBMs are generative stochastic neural networks that can learn the distribution of their inputs.	html, pdf
3.8 - Generative Adversarial Networks GANs are generative networks able to generate images from pure noise.	html, pdf
3.9 - Recurrent neural networks RNNs, especially LSTMs, were long the weapon of choice to process temporal sequences (text, video, etc)..	html, pdf

4 - Generative AI

	Slides
4.1 - Transformers The Transformer architecture of (Vaswani, 2017) used self-attention to replace RNNs and start the second wave of AI hype.	html, pdf
4.2 - Contrastive learning Contrastive learning is a form of self-supervised allowing to learn context-relevant representations from raw data..	html, pdf
4.3 - Vision Transformer Vision transformers use the Transformer architecture to be the new state of the art in computer vision.	html, pdf
4.4 - Diffusion models Diffusion models are a novel probabilistic architecture allowing to learn to generate images (Midjourney, Dall-E, etc) through incremental denoising.	html, pdf

5 - Neuro-AI

	Slides
5.1 - Limits of deep learning This lecture (provocatively) explains why deep learning-based approaches will never be able to achieve Artificial General Intelligence and why more brain-inspired approaches (neuro-AI) are the next step for AI..	html, pdf
5.2 - Hopfield networks Hopfield network allow to implement associative memory, a fundamental aspect of cognition..	html, pdf
5.3 - Reservoir Computing Reservoir Computing (RC) is a paradigm allowing to train recurrent neural networks on time series with much less compuations than with deep learning approaches.	html, pdf
5.4 - Spiking networks Spiking networks, in addition to being closer to brain functioning, allow to perform the same computations as deep netowkrs without requiring as much communication, allowing energy-efficient implementations on neuro-morphic hardware.	html, pdf
5.5 - Beyond deep learning To conclude, we will see some of the requirement of genetral intelligence that need to be added to our models.	html, pdf

Exercises

You will find below links to download the notebooks for the exercises (which you have to fill) and their solution (which you can look at after you have finished the exercise). It is recommended not to look at the solution while doing the exercise unless you are lost. Alternatively, you can run the notebooks directly on Colab (https://colab.research.google.com/) if you have a Google account.

For instructions on how to install a Python distribution on your computer, check this page.

	Notebook	Solution
1 - Introduction to Python Introduction to the Python programming language. Optional for students already knowing Python.	ipynb, colab	ipynb, colab
2 - Numpy and Matplotlib Presentation of the numpy library for numerical computations and matplotlib for visualization. Also optional for students already familiar.	ipynb, colab	ipynb, colab
3 - Linear regression Implementation of the basic linear regression algorithm in Python and scikit-learn.	ipynb, colab	ipynb, colab
4 - Multiple Linear regression MLR on the California Housing dataset using scikit-learn.	ipynb, colab	ipynb, colab
5 - Cross-validation Different approaches to cross-validation using scikit-learn.	ipynb, colab	ipynb, colab
6 - Linear classification Hard and soft linear classification.	ipynb, colab	ipynb, colab
7 - Softmax classifier Softmax classifier for multi-class classification.	ipynb, colab	ipynb, colab
8 - Multi-layer perceptron Basic implementation in Python+Numpy of the multi-layer perceptron and the backpropagation algorithm.	ipynb, colab	ipynb, colab
9 - MNIST classification using keras Keras tutorial applied to classifying the MNIST dataset with a MLP.	ipynb, colab	ipynb, colab
10 - Convolutional neural networks Implementation of a CNN in keras for MNIST.	ipynb, colab	ipynb, colab
11 - Transfer learning Leveraging data augmentation and/or pre-trained CNNs (Xception) for learning a small cats vs. dogs dataset.	ipynb, colab	ipynb, colab
12 - Variational autoencoders Implementing a VAE in keras.	ipynb, colab	ipynb, colab
13 - Recurrent neural networks Sentiment analysis and time series prediction using LSTM layers.	ipynb, colab	ipynb, colab