National University of Singapore

Solving Raven's Progressive Matrices with Neural Networks

Neural networks demonstrate a significant capacity for solving Raven’s Progressive Matrices, a standard benchmark for measuring abstract reasoning and fluid intelligence.

• Deep learning models using ImageNet pre-training achieve 86.26% accuracy on the RAVEN dataset, exceeding human-level performance in structured abstract reasoning tests.
• Neural networks excel at pattern recognition in matrix puzzles but prioritize statistical correlation over the genuine conceptual understanding used in human cognition.
• The Multilabel Classification with Pseudo Target method enables unsupervised reasoning by converting matrix problems into supervised tasks, doubling the accuracy of random guessing.