Bridging Cognitive Programs and Machine Learning

TL;DR

This paper discusses the limitations of current machine learning approaches in generalization and reasoning, proposing new directions to enhance their adaptability and robustness across various domains.

Contribution

It identifies key shortcomings of modern machine learning in vision and reinforcement learning and suggests future research directions to address these issues.

Findings

Current ML models struggle with data scarcity and domain shifts.

Machine learning methods have limited reasoning capabilities.

Proposed exploration of new approaches to improve adaptability.

Abstract

While great advances are made in pattern recognition and machine learning, the successes of such fields remain restricted to narrow applications and seem to break down when training data is scarce, a shift in domain occurs, or when intelligent reasoning is required for rapid adaptation to new environments. In this work, we list several of the shortcomings of modern machine-learning solutions, specifically in the contexts of computer vision and in reinforcement learning and suggest directions to explore in order to try to ameliorate these weaknesses.