Programme

Shihan Wang: Learnable and Interactive Autonomous Agents through Reinforcement Learning

(Chair: Jieting Luo)

Abstract: With the success of artificial intelligence in recent years, reinforcement learning (RL), a subfield of machine learning, has attracted great attention in recent years. Reinforcement learning also plays an important role in autonomous agents and multi-agent systems (MAS), such as autonomous driving, senor networks and games. In this talk, we will start with general information of reinforcement learning. Then, I will concentrate on the possible uses of RL in the development of interactive and learnable autonomous agents, both in single-agent and multi-agent scenarios. We will also explore the connections between RL and engineering multi-agent systems in future research.

Bio: Dr. Shihan Wang is a tenured assistant professor at Utrecht University, Intelligent systems group. She received her Ph.D. from Tokyo Institute of Technology and worked as a postdoctoral researcher at machine learning lab of University of Amsterdam. She was also a visiting researcher at Computational Social Science Lab of ETH Zurich. With a knowledge portfolio of machine learning, her current research focuses on developing efficient, scalable and reliable reinforcement learning (RL) approaches for intelligent and interactive systems. Her technical interest covers both single-agent and multi-agent RL, and their practical usage in sequential decision-making tasks. Several practical domains are especially interesting for her, including but not limited to dialogue, personalized healthcare, energy optimization and autonomous driving.

Pierre-Antoine Champin: A New Solid Web for Agents that Rock

(Chair: Andrei Ciortea)

Abstract: In 2001, Jim Hendler, Tim Berners-Lee and Ora Lassila published their vision of 'The Semantic Web', a vision in which intelligent agents (autonomous programs collaboratively achieving complex tasks on behalf of humans) played a key role. Yet, several years later, Jim Hendler was asking: 'Where are all the intelligent agents?'... and the question is still valid today.

I propose that the Web growing more and more centralized is the reason why intelligent personal agents, such as the ones envisioned in 2001, have not yet emerged. The tasks that these agents where supposed to take on are performed instead by huge platforms, aggregating data from millions of users. Who needs complex peer-to-peer collaboration processes, when you can tap all the information from a single source, right?

But this may be about to change. The numerous issues raised by that centralization are becoming more obvious everyday, and initiatives to re-decentralize the Web are gaining momentum. I will focus on the Solid project, an effort to allow for a new kind of Web applications, putting users back in control of their data. Solid is being actively developed by a community of developers, companies and public agencies, and in the process of becoming a W3C standard. But most importantly, Solid is building on existing web standards and largely based on the original Semantic Web vision. As such, it has the potential to make the Web a favourable environment for personal intelligent agents to thrive... at last!

Bio: Pierre-Antoine Champin is seconded researcher in the Wimmics group at Inria (the French National Institute for Research in Digital Science and Technology). He has been a W3C fellow since February 2021. In the W3C, he is part of the strategy team, focusing on activities related to data interoperability and knowledge graphs.