Jarrod Haas

Jarrod Haas, William Yolland, Bernhard Rabus

We propose a simple modification to standard ResNet architectures--L2 normalization over feature space--that substantially improves out-of-distribution (OoD) performance on the previously proposed Deep Deterministic Uncertainty (DDU) benchmark. We show that this change also induces early Neural Collapse (NC), an effect linked to better OoD performance. Our method achieves comparable or superior OoD detection scores and classification accuracy in a small fraction of the training time of the benchmark. Additionally, it substantially improves worst case OoD performance over multiple, randomly initialized models. Though we do not suggest that NC is the sole mechanism or a comprehensive explanation for OoD behaviour in deep neural networks (DNN), we believe NC's simple mathematical and geometric structure can provide a framework for analysis of this complex phenomenon in future work.

Jarrod Haas, William Yolland, Bernhard Rabus

We demonstrate that L2 normalization over feature space can produce capable performance for Out-of-Distribution (OoD) detection for some models and datasets. Although it does not demonstrate outright state-of-the-art performance, this method is notable for its extreme simplicity: it requires only two addition lines of code, and does not need specialized loss functions, image augmentations, outlier exposure or extra parameter tuning. We also observe that training may be more efficient for some datasets and architectures. Notably, only 60 epochs with ResNet18 on CIFAR10 (or 100 epochs with ResNet50) can produce performance within two percentage points (AUROC) of several state-of-the-art methods for some near and far OoD datasets. We provide theoretical and empirical support for this method, and demonstrate viability across five architectures and three In-Distribution (ID) datasets.

Jarrod Haas

Out-of-Distribution (OoD) detection has developed substantially in the past few years, with available methods approaching, and in a few cases achieving, perfect data separation on standard benchmarks. These results generally involve large or complex models, pretraining, exposure to OoD examples or extra hyperparameter tuning. Remarkably, it is possible to achieve results that can exceed many of these state-of-the-art methods with a very simple method. We demonstrate that ResNet18 trained with Supervised Contrastive Learning (SCL) produces state-of-the-art results out-of-the-box on near and far OoD detection benchmarks using only Euclidean distance as a scoring rule. This may obviate the need in some cases for more sophisticated methods or larger models, and at the very least provides a very strong, easy to use baseline for further experimentation and analysis.