Predicted Embedding Power Regression for Large-Scale Out-of-Distribution Detection
This addresses the safety and performance issues for real-world machine learning systems by improving OOD detection at large scales, though it is incremental as it builds on existing methods like grouped softmax.
The paper tackles the problem of large-scale out-of-distribution (OOD) detection, where existing methods struggle, by developing a novel approach that calculates the probability of predicted class labels based on learned label distributions, resulting in statistically significant improvements in AUROC (84.2 vs 82.4) and AUPR (96.2 vs 93.7) across 14 datasets.
Out-of-distribution (OOD) inputs can compromise the performance and safety of real world machine learning systems. While many methods exist for OOD detection and work well on small scale datasets with lower resolution and few classes, few methods have been developed for large-scale OOD detection. Existing large-scale methods generally depend on maximum classification probability, such as the state-of-the-art grouped softmax method. In this work, we develop a novel approach that calculates the probability of the predicted class label based on label distributions learned during the training process. Our method performs better than current state-of-the-art methods with only a negligible increase in compute cost. We evaluate our method against contemporary methods across $14$ datasets and achieve a statistically significant improvement with respect to AUROC (84.2 vs 82.4) and AUPR (96.2 vs 93.7).