LINKÖPING UNIVERSITY advertises up to four (4) positions as PhD student in Computer Science, formally based at the Department of Computer and Information Science (IDA) within WASP.
The student(s) will carry out research directed by Hector Geffner, a Guest WASP Professor at LiU and recipient of an Advanced ERC Grant on Representation Learning for Planning.
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 symbolic representations from raw perceptions. In our case, first order symbolic representations, involving objects and relations, are to be learned from scratch for planning and generalized planning. Other dimensions of representation learning to be pursued include representation grounding, transfer, composition, and scaffolding.
As an illustration, deep reinforcement learning methods became popular in AI when they were used to learn to play the Atari video games directly from images and scores by producing effective but opaque behavioral boxes. The idea of representation learning for planning is to learn meaningful first-order models based on objects and relations, from the same data, and to use them to generate meaningful and explainable behavior.
We are seeking highly motivated doctoral students eager to make a difference in these problems, with some background in areas such as machine learning, planning, logic and knowledge representation, combinatorial optimization and SAT.
Qualified to PhD studentships are only those who are or have been admitted to graduate studies. Good oral and written skills in English are required.
Ideal candidates should be able to do or will learn to do: theoretical and experimental work, logic and algorithms, and programming and “differential programming” (deep learning)