Section outline
-
The module presents the fundamental concepts, challenges, architectures and methodologies of deep learning. We introduce the learning of neural representations from vectorial, sequential and image data, covering both supervised and unsupervised learning, and hinting at various forms of weak supervision. Models covered include: deep autoencoders, convolutional neural networks, long-short term memory, gated recurrent units, advanced recurrent architectures, sequence-to-sequence, neural attention, Transformers, neural Turing machines. Methodological lectures will be complemented by introductory seminars to Keras-TF and Pytorch.
Date Topic References (NEW) Additional Material 20 03/04/2025 Deep Autoencoders
Sparse, denoising and contractive AE; deep RBM[SD] Coverage of the Prince book on this lecture is inadequate but you can use the lecture slides and complement with the additional material if necessary. (e.g. chapter 14 of the deep learning book). Additional Readings
[15] DBN: the paper that started deep learning
[16] Deep Boltzmann machines paper
[17] Review paper on deep generative models
[18] Long review paper on autoencoders from the perspective of representation learning
[19] Paper discussing regularized autoencoder as approximations of likelihood gradient21 08/04/2025
(11-13)Convolutional Neural Networks I
Introduction to the deep learning module; introduction to CNN; basic CNN elements[SD] Chapter 10 Additional Readings
[20-24] Original papers for LeNet, AlexNet, VGGNet, GoogLeNet and ResNet.
22 09/04/2025
(16-18)Convolutional Neural Networks II
CNN architectures for image recognition; convolution visualization; advanced topics (deconvolution, dense nets); applications and code
[SD] Chapter 10Additional Readings
[25] Complete summary of convolution arithmetics
[26] Seminal paper on batch normalization
[27] CNN interpretation using deconvolutions
[28] CNN interpretation with GradCAM[29] Seminal paper on dilated convolutions
[30] Object detection by Faster RCNN
23 10/04/2025
(14-16)Gated Recurrent Networks I
Deep learning for sequence processing; gradient issues;Coverage of Prince book on this lecture is inadequate (for reasons I do not understand). You can use the course slides for this topic, and if you like you can integrate those with chapter 10 from the Deep Learning Book. Additional Readings
[31] Paper describing gradient vanish/explosion24 11/04/2025
(14-16)
ROOM D3Gated Recurrent Networks II
long-short term memory; gated recurrent units; generative use of RNN
RECOVERY LECTUREAdditional Readings
[32] Original LSTM paper
[33] An historical view on gated RNN[34] Gated recurren units paper
[35] Seminal paper on dropout regularizationSoftware
- A simple introduction to generative use of LSTM
- The up-to-date implementation of NeuraTalk
25 15/04/2025
(11-13)Attention-based architectures
sequence-to-sequence; attention modules; transformers and vision transformers[SD] Chapter 12 Additional Readings
[36,37] Models of sequence-to-sequence and image-to-sequence transduction with attention
[38] Seminal paper on Transformers
[39] Transformers in vision26 16/04/2025
(16-18)Coding practice I - Guest lecture by Riccardo Massidda
Pytorch
27 17/04/2025
(14-16)Coding practice II - Guest lecture by Riccardo Massidda
Keras/TensorFlow
18/04/2025 - 25/04/2025 Spring Break: No Lectures
28 29/04/2025
(11-13)Memory-based models
multiscale network; hierarchical models; memory networks; neural Turing machines