New AI Model DIRCR Improves Abstract Visual Reasoning on RAVEN Datasets

A novel AI model named Dual-Inference Rule-Contrastive Reasoning (DIRCR) has been introduced to overcome obstacles in abstract visual reasoning. This model addresses the shortcomings of current approaches that typically emphasize either global context or local row-wise relationships, leading to inadequate rule representation and intertwined outputs. Central to DIRCR is the Dual-Inference Reasoning Module, which integrates a local pathway for row-wise analogical reasoning with a global pathway for comprehensive inference via a gated attention mechanism. Furthermore, it features a Rule-Contrastive Learning Module that generates positive and negative rule samples using pseudo-labels, enhancing feature separability through contrastive learning for improved abstract rule acquisition. Testing on three RAVEN datasets indicates that DIRCR markedly boosts reasoning robustness and generalization. The research detailing this model is published on arXiv under the identifier arXiv:2604.17584v1, and the related codes are publicly available.