CAT-GS: Balanced Multimodal Learning via Calibrated Gating and Fusion Surgery

2026-08-24

AI Summary

本文提出CAT-GS方法，通过校准、自适应阈值门控和融合手术解决多模态神经网络训练中的模态不平衡、不稳定门控及融合干扰问题。

Abstract

End-to-end training of multimodal neural networks often exhibits unstable neural dynamics characterized by three coupled failure modes that degrade learning: (i) modality imbalance, where one branch dominates gradient-based optimization; (ii) unstable gating, where noisy confidence cues induce erratic modality selection; and (iii) fusion interference, where modality-specific gradients conflict at the shared fusion layer. We propose CAT-GS (Calibrated, Adaptive, Thresholded Gating with Fusion Surgery), a neural dynamics-based optimization controller for intelligent computing applications. CAT-GS operates during backpropagation without modifying model architectures, fusion modules, or task losses. Through calibration of teacher-derived reliability via temperature scaling and EMA smoothing, CAT-GS stabilizes neural dynamics using a margin-thresholded policy to switch between warm-up dropout, weak-modality prioritization, and weak-biased blending, stabilizes gradient magnitudes under aggressive gating via capped gradient-budget renormalization, and applies fusion-only PCGrad to reduce destructive cross-modal interference at the primary shared bottleneck. We evaluate CAT-GS on audio--visual multimodal pattern recognition benchmarks (CREMA-D, AV-MNIST, and VGGSound), a tri-modal setting (UR-FUNNY), controlled synthetic data (CG-MNIST), and additional cross-domain benchmarks (AVE and CMU-MOSI). CAT-GS improves or matches fused multimodal accuracy against strong imbalance-aware baselines (including OGM-GE, G$^2$D, and UMT) across settings, and yields smoother gating behavior with fewer conflicting fusion gradients.

Problem

• multimodal neural networks
• unstable neural dynamics
• modality imbalance
• gating instability
• fusion interference

Innovation

• Calibrated Gating
• Fusion Surgery
• Gradient Budget Renormalization
• PCGrad