Diffusion Based for Sequence Anomaly Detection

Welcome to the official repository for Diffusion Based for Sequence Anomaly Detection. This repository hosts the original implementation of our sequence anomaly detection model.

Overview

The system is structured into distinct components. Specifically, it includes:

• TSA_training: TSA baseline embedding algorithm for modeling the ShipTrack Sequence
• LLM_SimCSE: SimCSE algorithm to finetune nomic embedding to adapt to ShipTrack domain
• DTE: Diffusion time estimation model training for anomaly detection.

Contents

1. TSA training
2. LLM_SimCSE
3. DTE

TSA training

This directory includes the code to train the TSA on the ShipTrack sequence through PyTorch Lightning

LLM_SimCSE

This directory includes the code to use SentenceTransformer to finetune Nomic on the ShipTrack sequence after text transformation.

DTE

This directory includes the code to train our main anomaly detection algorithm - DTE and other baseline algorithms such as DDPM.