Temporal and Object Quantification Networks

TL;DR

TOQ-Nets are neuro-symbolic networks designed to recognize complex relational-temporal events, capable of generalizing across varying object counts and sequence lengths through reasoning layers that implement finite-domain quantification.

Contribution

This paper introduces TOQ-Nets, a novel neuro-symbolic architecture with reasoning layers for finite-domain quantification, enabling robust generalization in temporal relational event recognition.

Findings

TOQ-Nets generalize to larger object sets than training data.

They handle temporal warpings effectively.

They require minimal training data for complex scenarios.

Abstract

We present Temporal and Object Quantification Networks (TOQ-Nets), a new class of neuro-symbolic networks with a structural bias that enables them to learn to recognize complex relational-temporal events. This is done by including reasoning layers that implement finite-domain quantification over objects and time. The structure allows them to generalize directly to input instances with varying numbers of objects in temporal sequences of varying lengths. We evaluate TOQ-Nets on input domains that require recognizing event-types in terms of complex temporal relational patterns. We demonstrate that TOQ-Nets can generalize from small amounts of data to scenarios containing more objects than were present during training and to temporal warpings of input sequences.