Outlier Synthesis

Outlier synthesis focuses on generating artificial outlier data to improve the ability of machine learning models to detect anomalies and out-of-distribution (OOD) samples, particularly in scenarios with limited or no labeled outlier data. Current research emphasizes techniques like virtual outlier synthesis, often integrated with self-supervised learning or prompt-based methods, and leverages model architectures such as autoencoders and large vision-language models (e.g., CLIP) to generate synthetic outliers that effectively regularize model decision boundaries. This approach is crucial for enhancing the robustness and safety of machine learning systems in various applications, including autonomous driving, object detection, and class-incremental learning, by mitigating overconfidence in predictions and improving anomaly detection accuracy.

Papers

October 28, 2024

zGAN: An Outlier-focused Generative Adversarial Network For Realistic Synthetic Data Generation
Azizjon Azimi, Bonu Boboeva, Ilyas Varshavskiy, Shuhrat Khalilbekov, Akhlitdin Nizamitdinov, Najima Noyoftova, Sergey Shulgin
Anomaly Detection Synthetic Data Real World Synthetic Data Generation GAN Architecture Outlier Explanation Outlier Synthesis

July 22, 2024

Towards Open-World Object-based Anomaly Detection via Self-Supervised Outlier Synthesis
Brian K. S. Isaac-Medina, Yona Falinie A. Gaus, Neelanjan Bhowmik, Toby P. Breckon
Anomaly Detection Open World Open World Object Detection Distribution Object Outlier Synthesis Detection Backbone

March 30, 2024

CLIP-driven Outliers Synthesis for few-shot OOD detection
Hao Sun, Rundong He, Zhongyi Han, Zhicong Lin, Yongshun Gong, Yilong Yin
Patch Based Distribution Sample U CLIP Update Shot Out of Distribution Detection Outlier Synthesis

February 6, 2024

OVOR: OnePrompt with Virtual Outlier Regularization for Rehearsal-Free Class-Incremental Learning
Wei-Cheng Huang, Chun-Fu Chen, Hsiang Hsu
Class Incremental Learning Large Pre Trained Model Rehearsal Based Outlier Synthesis Rehearsal Free Class Incremental Learning

October 2, 2023

LS-VOS: Identifying Outliers in 3D Object Detections Using Latent Space Virtual Outlier Synthesis
Aldi Piroli, Vinzenz Dallabetta, Johannes Kopp, Marc Walessa, Daniel Meissner, Klaus Dietmayer
3D Object Detection 3D Object 3D Object Detector State of the Art Object Outlier Synthesis

August 4, 2023

Synthetic outlier generation for anomaly detection in autonomous driving
Martin Bikandi, Gorka Velez, Naiara Aginako, Itziar Irigoien
Autonomous Driving Anomaly Detection Semantic Segmentation Model Outlier Detection Outlier Synthesis

March 6, 2023

Non-Parametric Outlier Synthesis
Leitian Tao, Xuefeng Du, Xiaojin Zhu, Yixuan Li
Distribution Detection Gaussian Distribution Outlier Synthesis

February 2, 2022

VOS: Learning What You Don't Know by Virtual Outlier Synthesis
Xuefeng Du, Zhaoning Wang, Mu Cai, Yixuan Li
Distribution Detection Outlier Data Synthetic Outlier Outlier Synthesis