Senior Research Scientist @ DeepMind
Diana Borsa is a senior research scientist at DeepMind (Alphabet) working on fundamental concepts and algorithms in reinforcement learning (RL) and continual learning.
She is a mathematician by training and at heart, with a natural curiosity towards the world, how things work, how things behave and why, how learning and intelligence emerge. She loves research and especially projects at the interplay of many disciplines: computer science, mathematics, neuroscience and cognitive science.
More formally, she holds Bachelor degrees in Mathematics, Electrical Engineering and Computer Science as well as International Business. She came to London, and joined University College London for her graduate studies in 2011, where she completed a MSc. Machine Learning and further on a MRes./Phd in the Computational Statistics and Machine Learning Centre. She spent quite a bit of time between 2013-2016 in Cambridge at the Microsoft Research lab, working on automatic machine learning, generative models and variational methods, as well as reinforcement learning and games. Finally at the beginning of 2017, she joined one of DeepMind’s core research teams working on RL.
She is very passionate about education and promoting diversity and inclusion in Machine Learning and STEM.
Jul 18, 15:55
The potential of Reinforcement Learning
In this talk, we are going to begin with the premise of (general) artificial intelligence – as a means to automate the process of finding solutions to complex problems. In this context, I will introduce reinforcement learning (RL), as a paradigm to reason about learning in sequential decision learning problems, with little supervision (parse feedback) and long-term consequences. So far, RL has had impressive successes in the realm of games (like Atari, Go, Chess, Dota and Starcraft), achieving and discovering superhuman performing solutions. We can think of these as a great proof-of-concept. Yet there are many more problems out there that would be like to tackle. The focus of the talk will be to highlight what I believe to be some of the main challenges RL systems still face and current progress along these dimensions. Progress in any and all of these will have an incredible impact on this field, aiding our quest for a general-purpose learning system, as well as increasing the applicability of RL methods in real-world applications.