Multi-Viewpoint and Multi-Evaluation with Felicitous Inductive Bias Boost Machine Abstract Reasoning Ability

TL;DR

This paper demonstrates that end-to-end neural networks with well-designed inductive biases can effectively solve RPM abstract reasoning problems without extra data, highlighting multi-viewpoint and multi-evaluation as key strategies.

Contribution

It introduces the importance of felicitous inductive bias and multi-viewpoint, multi-evaluation strategies for neural networks to excel in abstract reasoning tasks like RPM.

Findings

Neural networks with appropriate inductive bias can solve RPM problems without extra meta-data.

Multi-viewpoint and multi-evaluation are crucial learning strategies for reasoning.

Analysis of why connectionist models struggle with generalization.

Abstract

Great endeavors have been made to study AI's ability in abstract reasoning, along with which different versions of RAVEN's progressive matrices (RPM) are proposed as benchmarks. Previous works give inkling that without sophisticated design or extra meta-data containing semantic information, neural networks may still be indecisive in making decisions regarding RPM problems, after relentless training. Evidenced by thorough experiments and ablation studies, we showcase that end-to-end neural networks embodied with felicitous inductive bias, intentionally design or serendipitously match, can solve RPM problems elegantly, without the augment of any extra meta-data or preferences of any specific backbone. Our work also reveals that multi-viewpoint with multi-evaluation is a key learning strategy for successful reasoning. Finally, potential explanations for the failure of connectionist models in generalization are provided. We hope that these results will serve as inspections of AI's ability beyond perception and toward abstract reasoning. Source code can be found in https://github.com/QinglaiWeiCASIA/RavenSolver.