Belle Lab

The Lab carries out research in artificial intelligence, by unifying ideas from machine learning and logic, with a recent emphasis on explainability and ethics.

We are motivated by the need to augment learning and perception with high-level structured, commonsensical knowledge, to enable systems to learn faster and more accurate models of the world. We are interested in developing computational frameworks that are able to explain their decisions, modular, re-usable, and robust to variations in problem description. A non-exhaustive list of topics include:

• probabilistic and statistical knowledge bases
• ethics and explainability in AI
• exact and approximate probabilistic inference
• statistical relational learning and causality
• unifying deep learning and probabilistic learning methods
• probabilistic programming
• numerical optimization
• automated planning and high-level programming
• reinforcement learning and learning for automated planning
• cognitive robotics
• automated reasoning
• modal logics (knowledge, action, belief)
• multi-agent systems and epistemic planning

For example, our recent work has touched upon:

• morality in machine learning systems
• tractable learning with relational logic
• deep tractable probabilistic generative models
• learning with missing data
• program learning for explainability
• implementing fairness
• model abstraction for explainability
• strategies for interpretable & responsible AI

Faculty: Vaishak Belle

Postdoctoral fellows and PhD students:

• Rafael Karampatsis (postdoc), interested in ML interpretability
• Paulius Dilkas, interested in logical abstractions
• Miguel Mendez Lucero, interested in causality
• Jonathan Feldstein (with James Cheney), interested in probabilistic programming
• Eleanor Platt (with Amos Storkey), interested in interpretable deep learning
• Fazl Barez (with Ekaterina Komendantskaya), interested in explainable AI
• Giannis Papantonis, interested in causality
• Ionela-Georgiana Mocanu, interested in PAC learning
• Gary Smith (with Ron Petrick), interested in epistemic planning
• Andreas Bueff, interested in tractable learning and reinforcement learning
• Sandor Bartha (with James Cheney), interested in program induction

Alumni:

• Samuel Kolb (PhD 2019, KU Leuven, with Luc De Raedt), interested in inference for hybrid domains
• Amélie Levray (Postdoctoral fellow), interested in tractable learning with credal networks
• Davide Nitti (PhD 2016, KU Leuven, with Luc De Raedt), interested in machine learning for hybrid domains

Visitors:

• Esra Erdem, Sabanci University
• Yoram Moses, Technion
• Brendan Juba, Washington University in St. Louis
• Loizos Michael (via the Alan Turing Institute), Open University of Cyprus
• Till Hoffman, RWTH Aachen University