BERT Embeddings: A Modern Machine-learning Approach for Detecting Malware from Command Lines (Part 2 of 2)
CrowdStrike data science researchers recently explored and experimented with the use of Bidirectional Encoder Representation from Transformers (BERT) for embedding command lines, focusing on anomaly detection, but without detailing the model itself. Diving deeper into that research, CrowdStrike researchers explain the reasons for using BERT for command line representation and how to train the model and assess its performance.