Explore robust deep learning methods for anomaly detection in this 24-minute conference talk from KDD 2020. Delve into various techniques, including video surveillance, spectral methods, PCA, and auto-encoders. Learn about matrix factorization approaches and robust convolutional auto-encoders (RCAE). Compare conventional and deep learning-based anomaly detection methods through experiments on diverse datasets. Discover the performance of non-inductive anomaly detection and image de-noising capabilities of RCAE versus RPCA. Gain insights from speakers Raghavendra Chalapathy, Khoa Nguyen, and Sanjay Chawla as they present their findings and conclusions on advanced anomaly detection techniques.
KDD 2020: Robust Deep Learning Methods for Anomaly Detection
Association for Computing Machinery (ACM) via YouTube
Overview
Syllabus
Intro
Anomaly Detection: Video Surveillance.
Anomaly Detection: By Spectral Techniques
Anomaly Detection: PCA
Conventional Anomaly Detection Techniques
Matrix Factorization Approach: PCA
Auto-encoders for anomaly detection.
Comparison: Conventional Anomaly Detection Methods
Robust (convolution) Auto-Encoders RCAE
RCAE Vs Robust PCA (1)
Training RCAE (1)
Summary of Datasets
Anomaly Detection: Methods Compared
Experiment Settings
Methodology
Non Inductive: Top anomalous Images Detected USPS : 220 images of '1's, and 11 images of 7 (anomalous)
Non Inductive Anomaly Detection: Performance
Image De-noising Capability: RCAE vs RPCA
Conclusion
Taught by
Association for Computing Machinery (ACM)
