Reproduce “CLAD: Efficient Log Anomaly Detection Directly on Compressed Representations” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Benzhao Tang, Shiyu Yang
Researchers at Guangzhou University developed CLAD, a deep learning framework that performs log anomaly detection directly on compressed byte streams, bypassing decompression and parsing overheads. CLAD achieved an average F1-score of 0.9909 across five benchmark datasets, outperforming existing methods that require uncompressed or parsed log data.
Key insights
- Anomalous log entries induce systematic structural deviations within compressed byte streams, making direct anomaly detection feasible without prior decompression or parsing.
- A multi-scale deep learning architecture, combining dilated CNNs and a hybrid Transformer–mLSTM encoder, effectively extracts fine-to-coarse-grained structural patterns from opaque compressed byte sequences.
- A specialized two-stage training strategy using masked feature prediction and joint focal-contrastive learning robustly addresses the challenges of learning from opaque compressed data and severe class imbalance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.