DiffCLIP: Differential Attention Meets CLIP
This work improves multi-modal representations for vision-language tasks, but it is incremental as it builds on existing CLIP architectures.
The authors tackled the problem of enhancing vision-language models by integrating differential attention into CLIP, resulting in superior performance on tasks like zero-shot classification and retrieval with minimal computational overhead.
We propose DiffCLIP, a novel vision-language model that extends the differential attention mechanism to CLIP architectures. Differential attention was originally developed for large language models to amplify relevant context while canceling out noisy information. In this work, we integrate this mechanism into CLIP's dual encoder (image and text) framework. With minimal additional parameters, DiffCLIP achieves superior performance on image-text understanding tasks. Across zero-shot classification, retrieval, and robustness benchmarks, DiffCLIP consistently outperforms baseline CLIP models. Notably, these gains come with negligible computational overhead, demonstrating that differential attention can significantly enhance multi-modal representations without sacrificing efficiency. Code can be found at https://github.com/hammoudhasan/DiffCLIP.