Advanced Topics in Deep Learning

Catalog description

Prepares graduate students with the knowledge they need to apply Deep Learning techniques for their own research. There has been tremendous success in developing unified neural architectures that achieve state-of-the-art performance on language understanding (GPT-3), visual perception (ViT), and even protein structure prediction (AlphaFold). We plan to understand how they work, and how the success of such unified models can give rise to further developments on self-supervised learning, a technique that trains machine learning models without requiring labeled data; and multimodal learning, a technique that utilizes multiple input sources, such as vision, audio, and text. We will study recent attempts to interpret these models, thus revealing potential risks on model bias. Paper reading, student presentations, and invited guest lectures. Students required to work on a final project that explores a novel direction along the line of the papers we cover.