< Back to previous page

Project

Neural Probabilistic Logic Programming

A key open question of artificial intelligence is how to combine low-level perception and high-level reasoning. While deep learning excels at the former, probability theory and logic are the main frameworks for high-level reasoning. The goal of this project is to reconcile the low- and high-levels in artificial intelligence by tightly integrating deep learning with probabilistic and logical reasoning to yield a powerful framework for neuro-symbolic computation. This project builds upon DeepProbLog, an initial framework that combines the probabilistic logic programming language ProbLog with neural networks. This framework will be used to study various challenges such as: scaling up inference and learning through approximate inference, optimization and loss functions for neuro-symbolic computation, program induction and structure learning, as well as relational reinforcement learning. The resulting techniques will be applied to datasets in language and knowledge graphs, program synthesis, and game playing.

Date:13 Oct 2021 →  Today
Keywords:neurosymbolic computation
Disciplines:Machine learning and decision making, Knowledge representation and reasoning, Neural, evolutionary and fuzzy computation
Project type:PhD project