Confidence Scoring Using Whitebox Meta-models with Linear Classifier Probes
This addresses the need for reliable confidence estimation to bridge the gap between experimental and real-world AI applications, though it appears incremental as it builds on existing probing and meta-model concepts.
The paper tackles the problem of confidence scoring for deep neural networks by proposing a two-model paradigm with a meta-model that learns confidence scores from linear classifier probes inserted between base model layers. The approach outperforms baselines in filtering tasks on CIFAR-10 and CIFAR-100 datasets with and without noise.
We propose a novel confidence scoring mechanism for deep neural networks based on a two-model paradigm involving a base model and a meta-model. The confidence score is learned by the meta-model observing the base model succeeding/failing at its task. As features to the meta-model, we investigate linear classifier probes inserted between the various layers of the base model. Our experiments demonstrate that this approach outperforms various baselines in a filtering task, i.e., task of rejecting samples with low confidence. Experimental results are presented using CIFAR-10 and CIFAR-100 dataset with and without added noise. We discuss the importance of confidence scoring to bridge the gap between experimental and real-world applications.