Learning Human-Aligned Representations with Contrastive Learning and Generative Similarity

2024-05-29

AI Summary

Existing few-shot learning methods struggle to model human-like abstract reasoning—such as shape regularity, Euclidean geometric concepts, and semantic hierarchies—limiting cognitive alignment, generalization, and cross-domain robustness. Method: We propose a novel framework that embeds Bayesian generative similarity—defined as the probability of sampling pairs from the same underlying distribution—into a contrastive learning architecture. This enables unsupervised learning of cognitively aligned representations without manual similarity annotations. Our approach integrates probabilistic generative models (including probabilistic programs), contrastive learning, and Bayesian similarity estimation, while explicitly optimizing the geometric properties of the embedding space. Contribution/Results: The method successfully reproduces human similarity judgments across diverse abstract reasoning tasks. It achieves significant improvements in few-shot classification accuracy and demonstrates enhanced robustness to domain shifts, validating its capacity to learn representations grounded in human cognitive structure.

Technology Category

• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior
• Machine Learning: Bayesian Learning
• Computer Vision: Visual Reasoning & Symbolic Representations

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Humans rely on effective representations to learn from few examples and abstract useful information from sensory data. Inducing such representations in machine learning models has been shown to improve their performance on various benchmarks such as few-shot learning and robustness. However, finding effective training procedures to achieve that goal can be challenging as psychologically rich training data such as human similarity judgments are expensive to scale, and Bayesian models of human inductive biases are often intractable for complex, realistic domains. Here, we address this challenge by leveraging a Bayesian notion of generative similarity whereby two data points are considered similar if they are likely to have been sampled from the same distribution. This measure can be applied to complex generative processes, including probabilistic programs. We incorporate generative similarity into a contrastive learning objective to enable learning of embeddings that express human cognitive representations. We demonstrate the utility of our approach by showing that it can be used to capture human-like representations of shape regularity, abstract Euclidean geometric concepts, and semantic hierarchies for natural images.

Problem

• Machine Learning
• Human-like Learning
• Complex Real-world Applications

Innovation

• Pattern Similarity
• Machine Learning
• Few-shot Learning