Online
Presenter: Simone Parisi, researcher within Abhinav Gupta's lab
Title: Tabula-Rasa & Task-Specific To Transfer & Task-Agnostic: Towards More Efficient and Generalizable RL
Abstract: Over the last decade, reinforcement learning (RL) has been established as an effective framework for solving a large variety of tasks. A lot of effort has been directed towards scaling RL to solve complex problems, such as robotic tasks with many degrees of freedom or videogames. These advances, however, generally depend on hand-crafted perception modules, pre-structured exploration policies, and either require billions of samples or bootstrapping from human data. In these cases, learning is tabula-rasa and task-agnostic: the agent learns from scratch and is given a specific task. This is in stark contrast to how humans explore: we continuously interact with the world --sometimes without an explicit task-- and carry over past knowledge.
With the aim of moving RL towards a more realistic transfer framework, I will present new methods for learning and transferring task-agnostic exploration policies, and investigate the characteristics of visual representations needed for effective transfer to control tasks.
Bio: Simone Parisi is a researcher within Abhinav Gupta's lab. His research interests lie in the area of reinforcement learning, especially in the fields of exploration, representation learning, transfer learning, policy search, and multi-objective optimization.
During his postdoc at Meta AI, Simone worked on exploration and representation learning methods that move away from the classic tabula-rasa paradigm towards a more realistic transfer framework.
Before his postdoc, Simone completed his PhD at the Intelligent Autonomous System lab of Jan Peters. There, he worked on novel methods for scaling reinforcement learning in the field of robotics to guarantee both a high degree of autonomy and the ability to solve complex tasks.
The University of Alberta Artificial Intelligence (AI) Seminar is a weekly meeting where researchers (including students, developers, and professors) interested in AI can share their current research. Presenters include local speakers from the University of Alberta and industry as well as other institutions. The seminars discuss a wide range of topics related in any way to Artificial Intelligence, from foundational theoretical work to innovative applications of AI techniques to new fields and problems of interest. Learn more at the AI Seminar website and by subscribing to the mailing list!
Looking to build AI capacity? Need a speaker at your event?
