Anomaly Detection on Remote Sensing with Ray + Horovod
summary
Uses a U-Net frame prediction model that ingests t−k..t−1 frames to predict frame t and flags pixels where prediction versus ground truth error exceeds a threshold as anomalies. Training runs distributed with Ray + Horovod on IBM Cloud Code Engine serverless nodes to handle terabytes of Sentinel data and volatile daily input. Pipeline needs only a Ray cluster YAML and a small modified training script. Example: NDVI anomaly pinpoints the Woolsey wildfire.
AI-written summary · about
chapters (5)
related talks
26mAnomaly Detection System for the CERN Cloud
35mAnomaly Detection in Smart Buildings using Federated Learning
27mDistributed Deep Learning with Horovod on Ray
38mAnomaly detection in the cloud: how to detect issues in data [...]
43mDetecting Anomalies in Streaming Data
39mMastering real-time anomaly detection with open source tools
29mAnomaly Detection
48mPaths to anomaly detection w/ TIBCO data science, streaming on AWS