ANC Workshop - Antonio Vergari

Tuesday, 18th October 2022

Semantic Probabilistic Layers for Neuro-Symbolic Learning - Antonio Vergari

Abstract: Many structured-output prediction (SOP) tasks in machine learning sport both soft (probabilistic) and hard (symbolic) constraints. Extending current deep learning architectures to correctly capture these constraints and guaranteeing their satisfaction at deployment time is a big open problem. In this work, we design a predictive layer for SOP that can be plugged into any neural network while guaranteeing its predictions to be consistent with a set of predefined symbolic constraints. Our Semantic Probabilistic Layer (SPL) can model intricate correlations, and hard constraints, over a structured output space all while being amenable to end-to-end learning via exact maximum likelihood. SPLs do so by combining exact probabilistic inference with logical reasoning in a clean and modular way, learning complex distributions and restricting their support to solutions of the constraint. As such, they can faithfully, and efficiently, model complex SOP tasks beyond the reach of alternative neuro-symbolic approaches. We empirically demonstrate that SPLs outperform these competitors in terms of accuracy on challenging SOP tasks including hierarchical multi-label classification, pathfinding and preference learning, while retaining perfect constraint satisfaction.

Event type: Workshop

Date: Tuesday, 18th October 2022

Time: 11:00

Location: G.03

Speaker(s): Antonio Vergari

Chair/Host: Sohan Seth