Séminaire en format hybride au local 4488 du GERAD ou Zoom.

Neural networks that are able to reliably execute algorithmic computation may hold transformative potential to both machine learning and theoretical computer science. On one hand, they could enable the kind of extrapolative generalisation scarcely seen with deep learning models. On another, they may allow for running classical algorithms on inputs previously considered inaccessible to them. Both of these promises are shepherded by the neural algorithmic reasoning blueprint. In this talk, I aim to provide an introduction to how to develop neural networks that execute algorithmic computation, followed by examples on how to deploy such neural networks in real-world problems, such as learning to do implicit planning.

Biographie: I’m a PhD student at Mila/University of Montreal, supervised by Jian Tang. I am broadly interested in how learning can be improved through the use of graph representations, having worked on neural algorithmic reasoning, reinforcement learning and applications to biotechnology, in particular drug discovery. During my PhD, I spent time at DeepMind, working with Doina Precup, George Papamakarios and Theophane Weber, and at MSR Cambridge, working with Marc Brockschmidt. Prior to that, I completed my MEng with distinction under the supervision of Pietro Lio at University of Cambridge. For more information, please visit https://andreeadeac22.github.io.