Quantifying Modality Contributions via Disentangling Multimodal Representations

2025/11/22 by Amit, Padegal, Kashyap, Omkar Mahesh, Rayasam, Namitha +2
#Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.2511.19470

Abstract

Quantifying modality contributions in multimodal models remains a challenge, as existing approaches conflate the notion of contribution itself. Prior work relies on accuracy-based approaches, interpreting performance drops after removing a modality as indicative of its influence. However, such outcome-driven metrics fail to distinguish whether a modality is inherently informative or whether its value arises only through interaction with other modalities. This distinction is particularly important in cross-attention architectures, where modalities influence each other's representations. In this work, we propose a framework based on Partial Information Decomposition (PID) that quantifies modality contributions by decomposing predictive information in internal embeddings into unique, redundant, and synergistic components. To enable scalable, inference-only analysis, we develop an algorithm based on the Iterative Proportional Fitting Procedure (IPFP) that computes layer and dataset-level contributions without retraining. This provides a principled, representation-level view of multimodal behavior, offering clearer and more interpretable insights than outcome-based metrics.

Citations

• MultiSHAP: A Shapley-Based Framework for Explaining Cross-Modal Interactions in Multimodal AI Models
• SmolVLM: Redefining small and efficient multimodal models
• Attention, Please! PixelSHAP Reveals What Vision-Language Models Actually Focus On
• PaliGemma 2: A Family of Versatile VLMs for Transfer
• TokenSHAP: Interpreting Large Language Models with Monte Carlo Shapley Value Estimation
• Enhancing multimodal cooperation via sample-level modality valuation
• Quantifying & Modeling Multimodal Interactions: An Information Decomposition Framework
• PMR: Prototypical Modal Rebalance for Multimodal Learning
• MultiViz: Towards Visualizing and Understanding Multimodal Models
• Balanced Multimodal Learning via On-the-fly Gradient Modulation
• Modality Competition: What Makes Joint Training of Multi-modal Network Fail in Deep Learning? (Provably)
• DIME: Fine-grained Interpretations of Multimodal Models via Disentangled Local Explanations
• Rethinking Attention-Model Explainability through Faithfulness Violation Test
• Perceptual Score: What Data Modalities Does Your Model Perceive?
• Vision-and-Language or Vision-for-Language? On Cross-Modal Influence in Multimodal Transformers
• Attention is not not Explanation
• Is Attention Interpretable?
• A Corpus for Reasoning About Natural Language Grounded in Photographs
• Defoiling Foiled Image Captions
• A Unified Approach to Interpreting Model Predictions
• Axiomatic Attribution for Deep Networks
• Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering
• Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization
• Nonnegative Decomposition of Multivariate Information

Related

• Enhancing multimodal cooperation via sample-level modality valuation
• Quantifying & Modeling Multimodal Interactions: An Information Decomposition Framework
• What are You Looking at? Modality Contribution in Multimodal Medical Deep Learning
• On Modality Bias Recognition and Reduction
• Modality-based Factorization for Multimodal Fusion