Interpreting CLIP with Sparse Linear Concept Embeddings (SpLiCE)

TL;DR

This paper introduces SpLiCE, a novel method that transforms dense CLIP embeddings into sparse, interpretable linear combinations of concepts, enhancing transparency without sacrificing performance.

Contribution

The work presents a task-agnostic, training-free approach to interpret and replace dense CLIP embeddings with sparse, concept-based representations, improving interpretability.

Findings

SpLiCE effectively decomposes CLIP embeddings into human-interpretable concepts.

The method maintains high downstream task performance.

SpLiCE enables detection of spurious correlations and model editing.

Abstract

CLIP embeddings have demonstrated remarkable performance across a wide range of multimodal applications. However, these high-dimensional, dense vector representations are not easily interpretable, limiting our understanding of the rich structure of CLIP and its use in downstream applications that require transparency. In this work, we show that the semantic structure of CLIP's latent space can be leveraged to provide interpretability, allowing for the decomposition of representations into semantic concepts. We formulate this problem as one of sparse recovery and propose a novel method, Sparse Linear Concept Embeddings, for transforming CLIP representations into sparse linear combinations of human-interpretable concepts. Distinct from previous work, SpLiCE is task-agnostic and can be used, without training, to explain and even replace traditional dense CLIP representations, maintaining high downstream performance while significantly improving their interpretability. We also demonstrate significant use cases of SpLiCE representations including detecting spurious correlations and model editing.