Feature, Alignment, and Supervision in Category Learning: A Comparative Approach with Children and Neural Networks

2026-02-03

AI Summary

This study investigates how human children and convolutional neural networks (CNNs) learn object categories under sparse labeling conditions, with a focus on the interplay among supervision signals, feature types (size, shape, pattern), and perceptual alignment. Using a species-fair experimental framework, the authors systematically compare learning behaviors through few-shot semi-supervised tasks and controlled variable paradigms. The findings reveal that children rapidly generalize from minimal labels, yet their performance is significantly modulated by feature biases and alignment fidelity. In contrast, CNNs exhibit improved accuracy with increased supervision, but this gain is constrained by the underlying feature structure and alignment quality. This work provides the first evidence of fundamental mechanistic differences in category learning between humans and artificial systems, underscoring the necessity of comparing interactive learning mechanisms rather than relying solely on aggregate accuracy metrics.

Technology Category

• Machine Learning: Semi-Supervised Learning
• Humans and AI: Learning Human Values and Preferences
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Understanding how humans and machines learn from sparse data is central to cognitive science and machine learning. Using a species-fair design, we compare children and convolutional neural networks (CNNs) in a few-shot semi-supervised category learning task. Both learners are exposed to novel object categories under identical conditions. Learners receive mixtures of labeled and unlabeled exemplars while we vary supervision (1/3/6 labels), target feature (size, shape, pattern), and perceptual alignment (high/low). We find that children generalize rapidly from minimal labels but show strong feature-specific biases and sensitivity to alignment. CNNs show a different interaction profile: added supervision improves performance, but both alignment and feature structure moderate the impact additional supervision has on learning. These results show that human-model comparisons must be drawn under the right conditions, emphasizing interactions among supervision, feature structure, and alignment rather than overall accuracy.

Problem

• category learning
• few-shot learning
• semi-supervised learning
• feature bias
• perceptual alignment

Innovation

• few-shot learning
• semi-supervised learning
• species-fair comparison
• feature bias
• perceptual alignment