Geodesic Multi-Modal Mixup for Robust Fine-Tuning

Changdae Oh, Junhyuk So, Hoyoon Byun, Yong Taek Lim, Minchul Shin, Jong June Jeon, Kyungwoo Song

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

8 Scopus citations

Abstract

Pre-trained multi-modal models, such as CLIP, provide transferable embeddings and show promising results in diverse applications. However, the analysis of learned multi-modal embeddings is relatively unexplored, and the embedding transferability can be improved. In this work, we observe that CLIP holds separated embedding subspaces for two different modalities, and then we investigate it through the lens of uniformity-alignment to measure the quality of learned representation. Both theoretically and empirically, we show that CLIP retains poor uniformity and alignment even after fine-tuning. Such a lack of alignment and uniformity might restrict the transferability and robustness of embeddings. To this end, we devise a new finetuning method for robust representation equipping better alignment and uniformity. First, we propose a Geodesic Multi-Modal Mixup that mixes the embeddings of image and text to generate hard negative samples on the hypersphere. Then, we fine-tune the model on hard negatives as well as original negatives and positives with contrastive loss. Based on the theoretical analysis about hardness guarantee and limiting behavior, we justify the use of our method. Extensive experiments on retrieval, calibration, few- or zero-shot classification (under distribution shift), embedding arithmetic, and image captioning further show that our method provides transferable representations, enabling robust model adaptation on diverse tasks. Code: https://github.com/changdaeoh/multimodal-mixup.

Original languageEnglish
Title of host publicationAdvances in Neural Information Processing Systems 36 - 37th Conference on Neural Information Processing Systems, NeurIPS 2023
EditorsA. Oh, T. Neumann, A. Globerson, K. Saenko, M. Hardt, S. Levine
PublisherNeural information processing systems foundation
ISBN (Electronic)9781713899921
StatePublished - 2023
Event37th Conference on Neural Information Processing Systems, NeurIPS 2023 - New Orleans, United States
Duration: 10 Dec 202316 Dec 2023

Publication series

NameAdvances in Neural Information Processing Systems
Volume36
ISSN (Print)1049-5258

Conference

Conference37th Conference on Neural Information Processing Systems, NeurIPS 2023
Country/TerritoryUnited States
CityNew Orleans
Period10/12/2316/12/23

Fingerprint

Dive into the research topics of 'Geodesic Multi-Modal Mixup for Robust Fine-Tuning'. Together they form a unique fingerprint.

Cite this