Proposes using sparse autoencoders to extract class-conditioned concept vectors, then measuring logit stability under targeted perturbations as an interpretable OOD signal for deep networks in medical imaging.
Scaling out-of-distribution detection for real-world settings
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
abstract
Detecting out-of-distribution examples is important for safety-critical machine learning applications such as detecting novel biological phenomena and self-driving cars. However, existing research mainly focuses on simple small-scale settings. To set the stage for more realistic out-of-distribution detection, we depart from small-scale settings and explore large-scale multiclass and multi-label settings with high-resolution images and thousands of classes. To make future work in real-world settings possible, we create new benchmarks for three large-scale settings. To test ImageNet multiclass anomaly detectors, we introduce the Species dataset containing over 700,000 images and over a thousand anomalous species. We leverage ImageNet-21K to evaluate PASCAL VOC and COCO multilabel anomaly detectors. Third, we introduce a new benchmark for anomaly segmentation by introducing a segmentation benchmark with road anomalies. We conduct extensive experiments in these more realistic settings for out-of-distribution detection and find that a surprisingly simple detector based on the maximum logit outperforms prior methods in all the large-scale multi-class, multi-label, and segmentation tasks, establishing a simple new baseline for future work.
representative citing papers
Debiased negative mining via Monte-Carlo sampling from ID labels and unlabeled wild data improves OOD detection with VLMs and achieves new state-of-the-art results.
DLED reformulates open-set face forgery detection as an uncertainty estimation task and uses dual-level spatial-frequency evidence collection to identify novel fake categories, claiming 20% average gains over baselines.
ConjNorm reframes OOD detection score design as optimizing norm p in an exponential family density model via a Bregman divergence theorem, with a tractable Monte Carlo estimator, claiming SOTA gains on CIFAR-100 and ImageNet-1K.
Face-Trace attributes synthetic face images to known generators, rejects images from unseen generators via an energy score, and clusters the rejected images into groups corresponding to distinct unknown generators, including in an incremental setting.
Hyperspherical time-frequency representations learned via von Mises-Fisher likelihood improve OOD detection on UCR and UEA archives using k-NN and Mahalanobis scores over contrastive baselines.
VOLTA, consisting of a deep encoder with learnable prototypes plus cross-entropy and post-hoc temperature scaling, matches or exceeds ten UQ baselines in accuracy, achieves lower expected calibration error, and performs well on out-of-distribution detection across CIFAR, SVHN, and corruption shifts.
Benchmark across architectures and shift regimes finds OOD detector rankings shift with representation collapse; proposes NC-based shortlist predictor and PCA filter without extra OOD data.
citing papers explorer
-
When Confidence Lacks Concepts: Interpretable OOD Detection via Representation Perturbations
Proposes using sparse autoencoders to extract class-conditioned concept vectors, then measuring logit stability under targeted perturbations as an interpretable OOD signal for deep networks in medical imaging.
-
Debiased Negative Mining Improves Out-of-distribution Detection with Pre-trained Vision-Language Models
Debiased negative mining via Monte-Carlo sampling from ID labels and unlabeled wild data improves OOD detection with VLMs and achieves new state-of-the-art results.
-
Open Set Face Forgery Detection via Dual-Level Evidence Collection
DLED reformulates open-set face forgery detection as an uncertainty estimation task and uses dual-level spatial-frequency evidence collection to identify novel fake categories, claiming 20% average gains over baselines.
-
ConjNorm: Tractable Density Estimation for Out-of-Distribution Detection
ConjNorm reframes OOD detection score design as optimizing norm p in an exponential family density model via a Bregman divergence theorem, with a tractable Monte Carlo estimator, claiming SOTA gains on CIFAR-100 and ImageNet-1K.
-
Face-Trace: Open-Set Attribution and Progressive Discovery of Synthetic Face Generators
Face-Trace attributes synthetic face images to known generators, rejects images from unseen generators via an energy score, and clusters the rejected images into groups corresponding to distinct unknown generators, including in an incremental setting.
-
Learning Hyperspherical Time-Frequency Representations for Time-Series Out-of-Distribution Detection
Hyperspherical time-frequency representations learned via von Mises-Fisher likelihood improve OOD detection on UCR and UEA archives using k-NN and Mahalanobis scores over contrastive baselines.
-
VOLTA: The Surprising Ineffectiveness of Auxiliary Losses for Calibrated Deep Learning
VOLTA, consisting of a deep encoder with learnable prototypes plus cross-entropy and post-hoc temperature scaling, matches or exceeds ten UQ baselines in accuracy, achieves lower expected calibration error, and performs well on out-of-distribution detection across CIFAR, SVHN, and corruption shifts.
-
A Systematic Analysis of Out-of-Distribution Detection Under Representation and Training Paradigm Shifts
Benchmark across architectures and shift regimes finds OOD detector rankings shift with representation collapse; proposes NC-based shortlist predictor and PCA filter without extra OOD data.