G$^{2}$D: Boosting Multimodal Learning with Gradient-Guided Distillation

2025-06-26

AI Summary

Multimodal learning commonly suffers from modality imbalance—where dominant modalities monopolize optimization while weaker ones contribute insufficiently, thereby limiting collaborative representation learning. To address this, we propose a gradient-guided knowledge distillation framework coupled with a novel Dynamic Sequential Modality Prioritization (SMP) mechanism, the first to enable modalities to alternately assume optimization dominance during training, thereby enforcing substantial gradient participation from weaker modalities. We further design a customized loss function integrating unimodal supervision and multimodal alignment constraints. Evaluated across multiple real-world multimodal benchmarks, our method significantly enhances the contribution of weaker modalities and consistently outperforms state-of-the-art approaches on both classification and regression tasks. These results demonstrate its effectiveness and generalizability in modeling heterogeneous modalities collaboratively.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Multimodal learning aims to leverage information from diverse data modalities to achieve more comprehensive performance. However, conventional multimodal models often suffer from modality imbalance, where one or a few modalities dominate model optimization, leading to suboptimal feature representation and underutilization of weak modalities. To address this challenge, we introduce Gradient-Guided Distillation (G$^{2}$D), a knowledge distillation framework that optimizes the multimodal model with a custom-built loss function that fuses both unimodal and multimodal objectives. G$^{2}$D further incorporates a dynamic sequential modality prioritization (SMP) technique in the learning process to ensure each modality leads the learning process, avoiding the pitfall of stronger modalities overshadowing weaker ones. We validate G$^{2}$D on multiple real-world datasets and show that G$^{2}$D amplifies the significance of weak modalities while training and outperforms state-of-the-art methods in classification and regression tasks. Our code is available at https://github.com/rAIson-Lab/G2D.

Problem

• Addressing modality imbalance in multimodal learning models
• Enhancing weak modalities' contribution during model training
• Improving multimodal feature representation and performance

Innovation

• Gradient-Guided Distillation for multimodal learning
• Dynamic sequential modality prioritization technique
• Custom loss function fusing uni and multimodal objectives