Vision Superalignment: Weak-to-Strong Generalization for Vision Foundation Models

TL;DR

This paper introduces a novel weak-to-strong supervision method for vision foundation models, demonstrating that using weaker models to guide stronger ones can surpass traditional fine-tuning and improve performance across various tasks.

Contribution

The paper proposes an adaptively adjustable loss function for weak-to-strong supervision in vision models, showing it outperforms strong-to-strong and fine-tuning methods in multiple scenarios.

Findings

Outperforms strong-to-strong generalization benchmarks

Surpasses fine-tuning with full datasets

Effective across few-shot, transfer, and noisy label learning

Abstract

Recent advancements in large language models have sparked interest in their extraordinary and near-superhuman capabilities, leading researchers to explore methods for evaluating and optimizing these abilities, which is called superalignment. In this context, our paper delves into the realm of vision foundation models, focusing on the concept of weak-to-strong generalization, which involves using a weaker model to supervise a stronger one, aiming to enhance the latter's capabilities beyond the former's limits. We introduce a novel and adaptively adjustable loss function for weak-to-strong supervision. Our comprehensive experiments span various scenarios, including few-shot learning, transfer learning, noisy label learning, and common knowledge distillation settings. The results are striking: our approach not only exceeds the performance benchmarks set by strong-to-strong generalization but also surpasses the outcomes of fine-tuning strong models with whole datasets. This compelling evidence underscores the significant potential of weak-to-strong generalization, showcasing its capability to substantially elevate the performance of vision foundation models. The code is available at https://github.com/ggjy/vision_weak_to_strong.