[Paper Review] Getting aligned on representational alignment

Ilia Sucholutsky, Lukas Muttenthaler+31 more|arXiv (Cornell University)|Oct 18, 2023

Bioinformatics and Genomic Networks34 citations

TL;DR

The paper proposes a unifying framework for representational alignment across cognitive science, neuroscience, and machine learning, surveys prior work through this lens, and outlines open problems to catalyze cross-disciplinary collaboration.

ABSTRACT

Biological and artificial information processing systems form representations of the world that they can use to categorize, reason, plan, navigate, and make decisions. How can we measure the similarity between the representations formed by these diverse systems? Do similarities in representations then translate into similar behavior? If so, then how can a system's representations be modified to better match those of another system? These questions pertaining to the study of representational alignment are at the heart of some of the most promising research areas in contemporary cognitive science, neuroscience, and machine learning. In this Perspective, we survey the exciting recent developments in representational alignment research in the fields of cognitive science, neuroscience, and machine learning. Despite their overlapping interests, there is limited knowledge transfer between these fields, so work in one field ends up duplicated in another, and useful innovations are not shared effectively. To improve communication, we propose a unifying framework that can serve as a common language for research on representational alignment, and map several streams of existing work across fields within our framework. We also lay out open problems in representational alignment where progress can benefit all three of these fields. We hope that this paper will catalyze cross-disciplinary collaboration and accelerate progress for all communities studying and developing information processing systems.

Motivation & Objective

• Provide a common language to describe representational alignment research across cognitive science, neuroscience, and machine learning.
• Summarize how prior work fits into a single framework with five key components.
• Identify and articulate open problems and challenges where progress benefits multiple fields.
• Encourage cross-disciplinary communication and collaboration in studying representational alignment.
• Offer a baseline framework that can accommodate measuring, bridging, and increasing alignment across systems.

Proposed method

• Survey literature from cognitive science, neuroscience, and machine learning to map existing work onto the proposed framework.
• Formalize representation spaces, data, measurements, embeddings, and alignment functions within a unified notation.
• Characterize measurement, bridging, and increasing representational alignment as three central challenges.
• Differentiate alignment function properties (similarity vs dissimilarity, descriptive vs differentiable, symmetric vs directional).
• Illustrate framework with representative examples and discuss open problems across disciplines.

Experimental results

Research questions

• RQ1 How can we consistently measure the degree of representational alignment between two systems across diverse domains?
• RQ2 How can representations from different systems be bridged into a shared space to enable direct comparison and integration?
• RQ3 How can we increase representational alignment through transformations or learning while ensuring generalization to new data?
• RQ4 What decision criteria (data, embeddings, and measurement choices) influence inferences about alignment across fields?
• RQ5 What are the open problems and risks in representational alignment that transcend cognitive science, neuroscience, and machine learning?

Key findings

• A unifying framework can describe representational alignment across cognitive science, neuroscience, and machine learning with five core components (data, systems, measurements, embeddings, and alignment function).
• Representational alignment research can be categorized into three challenges: measuring alignment, bridging representational spaces, and increasing alignment.
• Alignment measures can be similarity-based or dissimilarity-based, descriptive or differentiable, and symmetric or directional, each enabling different inferences.
• Different fields employ various alignment functions (e.g., representational similarity analysis in neuroscience; Pearson/cosine similarity in descriptive analyses; KL divergence in directional analyses).
• The framework facilitates cross-disciplinary interpretation of a wide range of prior studies and highlights open problems for collaboration.