Probabilistic Abduction for Visual Abstract Reasoning via Learning Rules in Vector-symbolic Architectures

Michael Hersche, Francesco di Stefano, Thomas Hofmann, Abu Sebastian, Abbas Rahimi
1/29/2024
cs.LGcs.AI

Official code: IBM/learn-vector-symbolic-architectures-rule-formulations (10 stars, GPL-3.0, last commit 2025-09-18).

Abstract

Abstract reasoning is a cornerstone of human intelligence, and replicating it with artificial intelligence (AI) presents an ongoing challenge. This study focuses on efficiently solving Raven's progressive matrices (RPM), a visual test for assessing abstract reasoning abilities, by using distributed computation and operators provided by vector-symbolic architectures (VSA). Instead of hard-coding the rule formulations associated with RPMs, our approach can learn the VSA rule formulations (hence the name Learn-VRF) with just one pass through the training data. Yet, our approach, with compact parameters, remains transparent and interpretable. Learn-VRF yields accurate predictions on I-RAVEN's in-distribution data, and exhibits strong out-of-distribution capabilities concerning unseen attribute-rule pairs, significantly outperforming pure connectionist baselines including large language models. Our code is available at https://github.com/IBM/learn-vector-symbolic-architectures-rule-formulations.

View on arXivView PDF

Code Implementations(1)

IBM/learn-vector-symbolic-architectures-rule-formulationsOfficial100%
GPL-3.0

PyTorch Implementation of the paper "Probabilistic Abduction for Visual Abstract Reasoning via Learning Rules in Vector-symbolic Architectures" published in the 3rd Workshop on Mathematical Reasoning and AI at NeurIPS 2023.

103Shell, PythonNov 13, 20231 years agoGPL-3.0

Similar Papers

Towards Learning to Reason: Comparing LLMs with Neuro-Symbolic on Arithmetic Relations in Abstract Reasoning

87% match1 repo252024

Towards Learning Abductive Reasoning using VSA Distributed Representations

86% match1 repo252024

A-I-RAVEN and I-RAVEN-Mesh: Two New Benchmarks for Abstract Visual Reasoning

85% match2024

Learning Differentiable Logic Programs for Abstract Visual Reasoning

83% match1 repo152023

A Unified View of Abstract Visual Reasoning Problems

82% match2024

Vector Symbolic Algebras for the Abstraction and Reasoning Corpus

82% match1 repo92025

Cite this paper

@article{hersche2024probabilistic,
  title  = {Probabilistic Abduction for Visual Abstract Reasoning via Learning Rules in Vector-symbolic Architectures},
  author = {Michael Hersche and Francesco di Stefano and Thomas Hofmann and Abu Sebastian and Abbas Rahimi},
  year   = {2024},
  eprint = {2401.16024},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2401.16024v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2401.16024.svg)](https://codeofpaper.com/paper/2401.16024)

Discussion