ModalImmune: Immunity Driven Unlearning via Self Destructive Training

2026-02-18

AI Summary

This work addresses the performance degradation of multimodal systems under partial or complete modality loss during real-world deployment. To enhance robustness, the authors propose a modality-immune training framework that deliberately masks specific modalities in a controlled manner during training, compelling the model to learn resilient joint representations. The framework innovatively integrates spectral-adaptive collapse regularization, information-gain-guided intervention, curvature-aware gradient masking, and a Neumann-truncated hypergradient-based auto-tuning strategy. Evaluated on standard multimodal benchmarks, the approach significantly improves robustness against missing or corrupted modalities while maintaining strong convergence stability and reconstruction fidelity.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Multimodal systems are vulnerable to partial or complete loss of input channels at deployment, which undermines reliability in real-world settings. This paper presents ModalImmune, a training framework that enforces modality immunity by intentionally and controllably collapsing selected modality information during training so the model learns joint representations that are robust to destructive modality influence. The framework combines a spectrum-adaptive collapse regularizer, an information-gain guided controller for targeted interventions, curvature-aware gradient masking to stabilize destructive updates, and a certified Neumann-truncated hyper-gradient procedure for automatic meta-parameter adaptation. Empirical evaluation on standard multimodal benchmarks demonstrates that ModalImmune improves resilience to modality removal and corruption while retaining convergence stability and reconstruction capacity.

Problem

• multimodal systems
• modality loss
• robustness
• input channel failure
• deployment reliability

Innovation

• modality immunity
• self-destructive training
• gradient masking
• hyper-gradient adaptation
• multimodal robustness