Up to three PhD student positions in Computer Science formally based at the Department of Computer and Information Science (IDA), Linköping University.
Your work assignments
The student(s) will carry out research directed by Hector Geffner, a Wallenberg Guest Professor in Artificial Intelligence at LiU, and recipient of an Advanced ERC Grant on representation learning for acting and planning (RLeap, 2020-2025).
The research is focused on a problem that is at the heart of the current split in AI between data-based learners and model-based reasoners: the problem of learning models from data, and more generally, the problem of learning representations from data over known formal languages with a known structure and semantics. While these representations are to be learned using deep learning methods, the use of domain-independent target languages for learning is aimed at obtaining representations that generalize and can be reused, and which are more transparent and require less data. In the setting of action and planning, language-based representations have been used to represent general dynamics, general control strategies, and general subgoal structures (intrinsic rewards). See the paper “Target Languages (vs. Inductive Biases) for Learning to Act and Plan” (AAAI 2022) for details (https://arxiv.org/abs/2109.07195).
You will belong to RLPLAB (Representation, Learning and Planning Lab) within the AIICS division (Artificial Intelligence and Integrated Computer Systems).