Loading Events

Imitation Learning from Limited Demonstrations

Overview

Despite reinforcement learning has shown its great success in recent years, designing hand-crafted reward functions can be extremely difficult and even impossible in many real-world tasks. One alternative way is to teach human behaviours to autonomous agents through imitation learning. High-performance of imitation learning usually requires a large number of high-quality demonstrations consisting of expert state-action pairs. However, collecting such demonstrations can be expensive and impractical in real-world scenarios. In contrast, weak demonstrations are cheaper to collect and often occur in practice. Efficient learning an autonomous agent from such weak demonstrations is desirable and challenging.

In this talk, we will discuss the challenges of extremely limited demonstration data in high-dimensional environments. Several new clues, such as generative intrinsic reward, planning-guided surrogate reward, and reward from weak environment information, will be studied in this talk. Empirically, the learnt reward modules enable the imitation agent to do better exploration, which is critical for the agent to outperform the expert demonstrator in Atari games and continuous controlled tasks!

Learning Outcome

Course Outline

Who Should Attend

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.

Speakers

 Prof. Ivor W. Tsang

Prof. Ivor W. Tsang

Prof Ivor W Tsang is Director of A*STAR Centre for Frontier AI Research (CFAR) since January 2022. Previously, he was a Professor of Artificial Intelligence (AI), at University of Technology Sydney (UTS), and Research Director of the Australian Artificial Intelligence Institute (AAII), the largest AI institute in Australia, which is the key player to drive UTS to rank 10th globally and 1st in Australia for AI research, in the latest AI Research Index. Prof Tsang is working at the forefront of big data analytics and AI. His research focuses on transfer learning, deep generative models, learning with weak supervision, big data analytics for data with extremely high dimensions in features, samples and labels. His work is recognised internationally for its outstanding contributions to those fields.

Recently, Prof Tsang was conferred the IEEE Fellow for his outstanding contributions to large-scale machine learning and transfer learning. Besides these, Prof Tsang serves as the Editorial Board for the Journal of Machine Learning Research, Machine Learning, Journal of AI Research, IEEE Transactions on Pattern Analysis and Machine Intelligence, IEEE Transactions on AI, IEEE Transactions on Big Data, and IEEE Transactions on Emerging Topics in Computational Intelligence.

Details

Status


This event is now closed.

Share This Event