Research map / Label-efficient and robust learning / Self-supervised and contrastive

Rotation Has Two Sides: Evaluating Data Augmentation for Deep One-class Classification

ICLR 2024 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Representation evaluation and trade-offs

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Evaluates how data augmentation, especially rotation-based pseudo-anomaly generation, affects deep one-class classification with self-supervised representations.

Problem

Deep one-class classification needs representations that separate normal from anomalous data when only one class is seen in training; discriminative self-supervised methods relying on geometric transformations such as rotation are the focus.

Approach

Examines the dual role of rotation-based augmentation in self-supervised one-class classification, evaluating how such geometric transformations used to create pseudo-anomalies affect performance (abstract is truncated, so details are limited).

Closest papers by meaning

• Learning and Evaluating Representations for Deep One-Class Classification (ICLR 2021 · similarity 0.87)
• Self-Supervised Representation Learning by Rotation Feature Decoupling (CVPR 2019 · similarity 0.87)
• Few-Shot One-Class Classification via Meta-Learning (AAAI 2021 · similarity 0.85)
• Learning with 3D rotations, a hitchhiker's guide to SO(3) (ICML 2024 · similarity 0.85)
• Self-supervised Transformation Learning for Equivariant Representations (NeurIPS 2024 · similarity 0.83)
• MetaCoCo: A New Few-Shot Classification Benchmark with Spurious Correlation (ICLR 2024 · similarity 0.82)