Log Anomaly Detection Using Adaptive Universal Transformer
Sergio Ryan Wibisono, Achmad Imam Kistijantoro · 2019
This work shows deep learning log anomaly detection mechanism based on modifications of DeepLog. Instead of using LSTM as implemented on DeepLog, we use adaptive universal transformer model to allow learning long input sequences, which is difficult to do with LSTM. Our experiments show that by utilizing adaptive universal transformer in DeepLog we can train the model using arbitrarily longer sequences that result in good accuracy in detecting log anomaly.