François Pitié

Deep Learning

Department of Electronic & Electrical Engineering • Trinity College Dublin

Deep Learning Coursework and Visualisations

Module Overview

An intensive graduate-level introduction to deep learning architectures and their practical applications in signal processing, audio, computer vision, and natural language processing. The curriculum emphasizes first-principles mathematical derivations alongside high-performance PyTorch implementations and distributed GPU training.

Core Topics Covered

Foundations & Optimisation Dynamics

Computational graphs, backpropagation mechanics, loss surfaces, gradient descent dynamics, and regularisation strategies (BatchNorm, LayerNorm, Dropout).

Convolutional Architectures & Computer Vision

Spatial convolutions, receptive fields, feature hierarchies, residual connections (ResNet, ConvNeXt), and dense spatial representations.

Sequence Modelling & Attention Mechanics

Recurrent architectures (RNNs, LSTMs), sequence-to-sequence translation, and scaled dot-product self-attention.

Transformers & Generative Modelling

Transformer encoder-decoder architectures, Vision Transformers (ViT), generative adversarial networks (GANs), and diffusion probabilistic models.

Pedagogy & Lab Infrastructure

Following a mastery-learning framework, the course pairs theoretical lectures with a continuous autograded laboratory system. Students build deep learning models from scratch in Colab notebooks and receive rapid, automated performance verification upon submitting code to GitHub Classroom.

The entire curriculum is accompanied by comprehensive open courseware and interactive lecture notes authored specifically for the course, available publicly to students and researchers worldwide.

The design and delivery of the 4C16 automated lab platform and open curriculum was recognised with the University Excellence in Teaching Award and the Inspiring Educator Award at Trinity College Dublin (2026).