Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Survey on Compositional Learning of AI Models: Theoretical and Experimental Practices

Sania Sinha, Tanawan Premsri, Parisa Kordjamshidi

TL;DR

A survey of the literature on compositional learning of AI models and the connections made to cognitive studies identifies abstract concepts of compositionality in cognitive and linguistic studies and connects these to the computational challenges faced by language and vision models in compositional reasoning.

Abstract

Compositional learning, mastering the ability to combine basic concepts and construct more intricate ones, is crucial for human cognition, especially in human language comprehension and visual perception. This notion is tightly connected to generalization over unobserved situations. Despite its integral role in intelligence, there is a lack of systematic theoretical and experimental research methodologies, making it difficult to analyze the compositional learning abilities of computational models. In this paper, we survey the literature on compositional learning of AI models and the connections made to cognitive studies. We identify abstract concepts of compositionality in cognitive and linguistic studies and connect these to the computational challenges faced by language and vision models in compositional reasoning. We overview the formal definitions, tasks, evaluation benchmarks, various computational models, and theoretical findings. Our primary focus is on linguistic benchmarks and combining language and vision, though there is a large amount of research on compositional concept learning in the computer vision community alone. We cover modern studies on large language models to provide a deeper understanding of the cutting-edge compositional capabilities exhibited by state-of-the-art AI models and pinpoint important directions for future research.

A Survey on Compositional Learning of AI Models: Theoretical and Experimental Practices

TL;DR

A survey of the literature on compositional learning of AI models and the connections made to cognitive studies identifies abstract concepts of compositionality in cognitive and linguistic studies and connects these to the computational challenges faced by language and vision models in compositional reasoning.

Abstract

Compositional learning, mastering the ability to combine basic concepts and construct more intricate ones, is crucial for human cognition, especially in human language comprehension and visual perception. This notion is tightly connected to generalization over unobserved situations. Despite its integral role in intelligence, there is a lack of systematic theoretical and experimental research methodologies, making it difficult to analyze the compositional learning abilities of computational models. In this paper, we survey the literature on compositional learning of AI models and the connections made to cognitive studies. We identify abstract concepts of compositionality in cognitive and linguistic studies and connect these to the computational challenges faced by language and vision models in compositional reasoning. We overview the formal definitions, tasks, evaluation benchmarks, various computational models, and theoretical findings. Our primary focus is on linguistic benchmarks and combining language and vision, though there is a large amount of research on compositional concept learning in the computer vision community alone. We cover modern studies on large language models to provide a deeper understanding of the cutting-edge compositional capabilities exhibited by state-of-the-art AI models and pinpoint important directions for future research.
Paper Structure (20 sections, 1 figure, 2 tables)

This paper contains 20 sections, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Compositional Learning Facets
  3. Measures of Compositionality
  4. Compositionality as Function Properties
  5. Compositionality and Continual Learning
  6. Compositionality and Emergent Intelligence
  7. Abstract Tasks and Datasets
  8. Systematicity or Novel Composition
  9. Productivity or Length Generalization
  10. Other Generalization Criteria
  11. Empirical Findings: Compositional Learning Models
  12. Basic Neural Models
  13. Transformer-based Architectures
  14. Neurosymbolic Architectures
  15. Theoretical Findings: Mathematical Formulations of Compositionality
  16. ...and 5 more sections

Figures (1)

  • Figure 1: Outline of covered concepts in this survey, related to the structure of the paper. We structure our study of compositional learning by dividing it into four parts of compositional learning facets, datasets, compositional learning models, and evaluation methods from both empirical and theoretical perspectives. The topics have the respective sections associated with them. The main areas of required research and future direction are included in the descriptions in the Evaluation boxes, which are further discussed in Section \ref{['discussion']}.