Understanding Transferable Representation Learning and Zero-shot Transfer in CLIP

TL;DR

This paper provides a theoretical analysis of CLIP's transferable representations and zero-shot transfer capabilities, introduces a new CLIP-inspired method, and demonstrates improved performance on benchmark datasets.

Contribution

It offers the first formal study of CLIP's transfer learning mechanisms and proposes a novel approach that outperforms existing methods.

Findings

Features from different modalities get aligned in CLIP.

Theoretical insights explain CLIP's zero-shot transfer performance.

Proposed method achieves better results than CLIP on benchmarks.

Abstract

Multi-modal learning has become increasingly popular due to its ability to leverage information from different data sources (e.g., text and images) to improve the model performance. Recently, CLIP has emerged as an effective approach that employs vision-language contrastive pretraining to learn joint image and text representations and exhibits remarkable performance in zero-shot learning and text-guided natural image generation. Despite the huge practical success of CLIP, its theoretical understanding remains elusive. In this paper, we formally study transferrable representation learning underlying CLIP and demonstrate how features from different modalities get aligned. We also analyze its zero-shot transfer performance on the downstream tasks. Inspired by our analysis, we propose a new CLIP-type approach, which achieves better performance than CLIP and other state-of-the-art methods on benchmark datasets.