Pith. sign in

REVIEW 5 cited by

SSD: A Unified Framework for Self-Supervised Outlier Detection

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2103.12051 v1 pith:YHZE32RR submitted 2021-03-22 cs.CV cs.AIcs.LG

SSD: A Unified Framework for Self-Supervised Outlier Detection

classification cs.CV cs.AIcs.LG
keywords datadetectiondetectorstrainingunlabeledframeworkperformancedetector
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We ask the following question: what training information is required to design an effective outlier/out-of-distribution (OOD) detector, i.e., detecting samples that lie far away from the training distribution? Since unlabeled data is easily accessible for many applications, the most compelling approach is to develop detectors based on only unlabeled in-distribution data. However, we observe that most existing detectors based on unlabeled data perform poorly, often equivalent to a random prediction. In contrast, existing state-of-the-art OOD detectors achieve impressive performance but require access to fine-grained data labels for supervised training. We propose SSD, an outlier detector based on only unlabeled in-distribution data. We use self-supervised representation learning followed by a Mahalanobis distance based detection in the feature space. We demonstrate that SSD outperforms most existing detectors based on unlabeled data by a large margin. Additionally, SSD even achieves performance on par, and sometimes even better, with supervised training based detectors. Finally, we expand our detection framework with two key extensions. First, we formulate few-shot OOD detection, in which the detector has access to only one to five samples from each class of the targeted OOD dataset. Second, we extend our framework to incorporate training data labels, if available. We find that our novel detection framework based on SSD displays enhanced performance with these extensions, and achieves state-of-the-art performance. Our code is publicly available at https://github.com/inspire-group/SSD.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Tackling the Noisy Elephant in the Room: Label Noise-robust Out-of-Distribution Detection via Loss Correction and Low-rank Decomposition

    cs.LG 2025-09 reject novelty 6.0

    NOODLE corrects noisy labels with a transition matrix and cleans features via low-rank sparse decomposition, improving OOD detection under label noise.

  2. UniAlign: A Model-Agnostic Framework for Robust Network Traffic Classification under Distribution Shifts

    cs.LG 2026-05 unverdicted novelty 5.0

    UniAlign improves robustness of deep learning NTC models under distribution shifts via domain alignment fine-tuning and stable ensembling, yielding 2.51% accuracy and 2.71% F1 gains over standard training on three pub...

  3. TINS: Test-time ID-prototype-separated Negative Semantics Learning for OOD Detection

    cs.CV 2026-05 unverdicted novelty 5.0

    TINS improves OOD detection by learning negative semantics at test time with ID-prototype separation, cutting average FPR95 from 14.04% to 6.72% on the Four-OOD benchmark with ImageNet-1K.

  4. HEDP: A Hybrid Energy-Distance Prompt-based Framework for Domain Incremental Learning

    cs.AI 2026-05 unverdicted novelty 5.0

    HEDP uses energy regularization inspired by Helmholtz free energy plus hybrid energy-distance weighting in prompts to improve domain selection and achieve a 2.57% accuracy gain on benchmarks like CORe50 while mitigati...

  5. Benchmarking Vision Foundation Models for Input Monitoring in Autonomous Driving

    cs.CV 2025-01 unverdicted novelty 5.0

    Vision foundation model embeddings with density modeling outperform state-of-the-art methods for unsupervised semantic and covariate shift detection in autonomous driving inputs.