WASP is very proud to have so many excellent researchers involved in the program. More than 450 researchers, reaching from assistant to senior professors, are affiliated with WASP. Some are international recruitments who have come to Sweden to join the WASP community, others are already well established in the Swedish academic system.
Through a series of portraits, you get the opportunity to get to know them a little bit better.
Meet Hector Geffner
Hector Geffner is a WASP Guest Wallenberg Professor in AI at the Department of Computer and Information Science at Linköping University (LiU), and ICREA Professor at the Universitat Pompeu Fabra, Barcelona. Professor Geffner was recruited by WASP in 2019.
What is your position/role in WASP?
Guest Wallenberg Professor at Linköping University.
Why did you choose to join WASP?
A good opportunity to get to know Sweden better, to interact with my colleagues at LiU, and to start a new research group in AI and ML.
What are the benefits you see in WASP?
Funding, focus on research, a new pool of talent, challenges, collaborations.
Briefly describe your research topic.
The boundary and interaction between data-based AI and model-based AI in the setting of action and planning. Data-based approaches support fast, reactive behavior; model-based approaches support slow reasoning. The two approaches are very much like Daniel Kahneman’s Systems 1 and 2 in his Thinking, Fast and Slow (Farrar, Straus and Giroux, 2011). We need to get data-based AI and model-based AI to talk to each other very much as our Systems 1 and 2 do. This is also the theme of an ERC grant that supports my research in Barcelona.
We have shown that it is possible to learn language-based models for acting and planning directly from unstructured data without assuming any background knowledge. The prior structural knowledge is in the design of the formal languages, which enable the learning of representations that are more general and transparent, have a known semantics, and support reasoning.
In what way can your research be of importance to our society in the future?
Deep learning and deep reinforcement learning have delivered true AI breakthroughs and impressive systems. Yet these systems are black boxes: we do not fully understand them, and they do not understand the physical and social world around them. By learning representations over languages that we design, we hope to contribute with the development of AI systems that are more transparent and that support a more meaningful interaction with humans.
For more information about Professor Geffner, see https://www.dtic.upf.edu/~hgeffner/
Published: June 28th, 2022