Improving Generalization for Abstract Reasoning Tasks Using Disentangled Feature Representations

Abstract

In this work we explore the generalization characteristics of unsupervised representation learning by leveraging disentangled VAE's to learn a useful latent space on a set of relational reasoning problems derived from Raven Progressive Matrices. We show that the latent representations, learned by unsupervised training using the right objective function, significantly outperform the same architectures trained with purely supervised learning, especially when it comes to generalization.

References in corpus (7)

• Auto-Encoding Variational Bayes
• Representation Learning with Contrastive Predictive Coding
• Disentangling by Factorising
• A simple neural network module for relational reasoning
• Early Visual Concept Learning with Unsupervised Deep Learning
• The Consciousness Prior
• CVAE-GAN: Fine-Grained Image Generation through Asymmetric Training

Cited by in corpus (18)

• Abstraction and Analogy-Making in Artificial Intelligence
• Are Disentangled Representations Helpful for Abstract Visual Reasoning?
• Unsupervised Model Selection for Variational Disentangled Representation Learning
• A Review of Emerging Research Directions in Abstract Visual Reasoning
• Deep Learning Methods for Abstract Visual Reasoning: A Survey on Raven's Progressive Matrices
• Dynamic Inference with Neural Interpreters
• On the Fairness of Disentangled Representations
• Attention on Abstract Visual Reasoning
• Symmetry and Complexity in Object-Centric Deep Active Inference Models
• Generating Correct Answers for Progressive Matrices Intelligence Tests
• Abstract Spatial-Temporal Reasoning via Probabilistic Abduction and Execution
• Multi-Granularity Modularized Network for Abstract Visual Reasoning
• Selective Replay Enhances Learning in Online Continual Analogical Reasoning
• Modeling Gestalt Visual Reasoning on the Raven's Progressive Matrices Intelligence Test Using Generative Image Inpainting Techniques
• Learning Algebraic Representation for Systematic Generalization in Abstract Reasoning
• Analogical and Relational Reasoning with Spiking Neural Networks
• A Feature-based Generalizable Prediction Model for Both Perceptual and Abstract Reasoning
• Towards Better Understanding of Disentangled Representations via Mutual Information