On the Importance of Regularisation and Auxiliary Information in OOD Detection

Neural networks are often utilised in critical domain applications (e.g. self-driving cars, financial markets, and aerospace engineering), even though they exhibit overconfident predictions for ambiguous inputs. This deficiency demonstrates a fundamental flaw indicating that neural overfit on spurious correlations. To address this problem work we present two novel objectives improve the ability of network to detect out-of-distribution samples therefore avoid We empirically demonstrate our methods outperform baseline perform better than majority existing approaches while still maintaining competitive performance against rest. Additionally, robustness approach common corruptions importance regularisation auxiliary information detection.