ECE 457DDeep Learning

Recent advances in neural network architectures and training algorithms have catalyzed significant breakthroughs in image classification, machine translation, protein folding, and beyond. This course follows the evolution of neural networks and their training algorithms, from the introduction of the perceptron in the 1950s and 1960s to the advent of ChatGPT in the 2020s. Topics covered include key training approaches such as maximum likelihood, contrastive learning, and diffusion modelling, as well as significant architectures such as convolutional nets, graph nets, and transformers. The course will also discuss how neural nets can learn useful “representations” of data, and explore recent trends in training models on web-scale datasets.