AI summaryⓘ
The authors focus on improving how multimodal models (which use multiple types of data) detect when they make mistakes, a big deal in critical areas like self-driving cars and healthcare. They noticed that when these models fail, their confidence drops compared to individual data types, calling this 'confidence degradation.' To fix this, the authors created a new training method called Adaptive Confidence Regularization that teaches the model to spot this drop in confidence. They also invented a way to create tricky examples for training by swapping features between data modes, helping the model learn to recognize uncertain predictions better. Their tests on various datasets and modalities showed that their method consistently improves the model’s reliability.
multimodal modelsfailure detectionconfidence degradationAdaptive Confidence Regularizationmultimodal feature swappingoutlier synthesispredictive confidenceself-driving vehiclesmedical diagnosticsmodel reliability
Authors
Moru Liu, Hao Dong, Olga Fink, Mario Trapp
Abstract
The deployment of multimodal models in high-stakes domains, such as self-driving vehicles and medical diagnostics, demands not only strong predictive performance but also reliable mechanisms for detecting failures. In this work, we address the largely unexplored problem of failure detection in multimodal contexts. We propose Adaptive Confidence Regularization (ACR), a novel framework specifically designed to detect multimodal failures. Our approach is driven by a key observation: in most failure cases, the confidence of the multimodal prediction is significantly lower than that of at least one unimodal branch, a phenomenon we term confidence degradation. To mitigate this, we introduce an Adaptive Confidence Loss that penalizes such degradations during training. In addition, we propose Multimodal Feature Swapping, a novel outlier synthesis technique that generates challenging, failure-aware training examples. By training with these synthetic failures, ACR learns to more effectively recognize and reject uncertain predictions, thereby improving overall reliability. Extensive experiments across four datasets, three modalities, and multiple evaluation settings demonstrate that ACR achieves consistent and robust gains. The source code will be available at https://github.com/mona4399/ACR.