NSA: Neuro-symbolic ARC Challenge

Paweł Batorski, Jannik Brinkmann, Paul Swoboda
1/8/2025
cs.AIcs.CL

Official code: Batorskq/NSA (9 stars, last commit 2025-01-09).

Abstract

The Abstraction and Reasoning Corpus (ARC) evaluates general reasoning capabilities that are difficult for both machine learning models and combinatorial search methods. We propose a neuro-symbolic approach that combines a transformer for proposal generation with combinatorial search using a domain-specific language. The transformer narrows the search space by proposing promising search directions, which allows the combinatorial search to find the actual solution in short time. We pre-train the trainsformer with synthetically generated data. During test-time we generate additional task-specific training tasks and fine-tune our model. Our results surpass comparable state of the art on the ARC evaluation set by 27% and compare favourably on the ARC train set. We make our code and dataset publicly available at https://github.com/Batorskq/NSA.

View on arXivView PDF

Code Implementations(1)

Batorskq/NSAOfficial100%
93PythonDec 5, 20241 years ago

Similar Papers

Vector Symbolic Algebras for the Abstraction and Reasoning Corpus

87% match1 repo92025

Neural networks for abstraction and reasoning: Towards broad generalization in machines

86% match2 repos2242024

Enhancing SQL Query Generation with Neurosymbolic Reasoning

82% match2024

Multi-step Inference over Unstructured Data

82% match2024

Neuro-Symbolic Artificial Intelligence: A Task-Directed Survey in the Black-Box Models Era

82% match2026

Neuro-Symbolic Artificial Intelligence: Towards Improving the Reasoning Abilities of Large Language Models

82% match2025

Cite this paper

@article{batorski2025nsa,
  title  = {NSA: Neuro-symbolic ARC Challenge},
  author = {Paweł Batorski and Jannik Brinkmann and Paul Swoboda},
  year   = {2025},
  eprint = {2501.04424},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2501.04424v1}
}

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/2501.04424.svg)](https://codeofpaper.com/paper/2501.04424)

Discussion