Back to Projects
Complete
Featured
Embedded Log Anomaly Detection
Python
Sentence Transformers
BERT
Machine Learning
Anomaly Detection
NLP
About this project
A machine learning pipeline for prioritizing anomalous embedded-system test logs using Sentence Transformer embeddings and cosine similarity. Built for a CS 589 project, it compares several BERT-based encoders, normalizes timestamp-heavy data, and fine-tunes all-MiniLM-L6-v2 to separate abnormal and normal test runs without exposing proprietary log content.
Highlights
- Preprocessed months of embedded-system test logs for semantic comparison
- Compared all-MiniLM-L6-v2, LaBSE, and msmarco Sentence Transformer models
- Built pairwise cosine-similarity matrices to rank the most and least similar runs
- Fine-tuned all-MiniLM-L6-v2 with labeled log pairs and CosineSimilarityLoss
- Experimented with warmup, weight decay, and longer training schedules
- Evaluated anomaly ranking through normal/abnormal tail composition and TP/FN framing
- Kept the workflow local to protect sensitive source logs