ReconBoost: Boosting Can Achieve Modality Reconcilement

TL;DR

ReconBoost introduces a modality-alternating learning approach for multi-modal models, addressing modality competition by dynamically updating fixed modalities and employing regularization techniques inspired by gradient boosting, leading to improved multi-modal learning performance.

Contribution

The paper proposes ReconBoost, a novel modality-reconcilement method that alternates learning between modalities and incorporates regularization, inspired by gradient boosting, to enhance multi-modal learning.

Findings

Outperforms existing multi-modal methods on six benchmarks.

Effectively mitigates modality competition and overfitting.

Demonstrates the benefit of alternating modality updates with regularization.

Abstract

This paper explores a novel multi-modal alternating learning paradigm pursuing a reconciliation between the exploitation of uni-modal features and the exploration of cross-modal interactions. This is motivated by the fact that current paradigms of multi-modal learning tend to explore multi-modal features simultaneously. The resulting gradient prohibits further exploitation of the features in the weak modality, leading to modality competition, where the dominant modality overpowers the learning process. To address this issue, we study the modality-alternating learning paradigm to achieve reconcilement. Specifically, we propose a new method called ReconBoost to update a fixed modality each time. Herein, the learning objective is dynamically adjusted with a reconcilement regularization against competition with the historical models. By choosing a KL-based reconcilement, we show that the proposed method resembles Friedman's Gradient-Boosting (GB) algorithm, where the updated learner can correct errors made by others and help enhance the overall performance. The major difference with the classic GB is that we only preserve the newest model for each modality to avoid overfitting caused by ensembling strong learners. Furthermore, we propose a memory consolidation scheme and a global rectification scheme to make this strategy more effective. Experiments over six multi-modal benchmarks speak to the efficacy of the method. We release the code at https://github.com/huacong/ReconBoost.