Anomaly Detection and Zero-Shot Learning with Error-Correcting Output Codes

Mohammed S. Khan,Lucas J. Lavalva,Mathew R. Marchiano,Scott Zockoll

Zero-shot learning (ZSL) is a process by which a machine learns to recognize previously unseen objects. We propose that ZSL can be used in an incremental learning setting such that a machine can detect unknown objects then integrate them incrementally. Class labels are represented by codewords from an error-correcting code. Using multi-label classification, anomalies are detected and encoded as new concepts based on the output codewords of the unseen objects and their Hamming distances with respect to seen objects. Preliminary results on a set of datasets show how application of error-correcting codes for anomaly detection outperforms current approaches.