project 2

VUS Range based Time Series-Anomaly Detection metric:

Achieved 5X optimization of VUS-Anomaly Detection (AD) metric by strategic enhancements in calculation methodologies

• Provided algorithmic and conceptual improvements, for reducing time and space complexity by a quadratic factor
• Obtained a fivefold runtime reduction using efficient data structures, producing runtimes similar to the Range-AUC metric
• Created run-time and robustness study for Synthetic and TSB-UAD Benchmark by balancing loads across 3 Dino servers using 10+ AD models including Isolation Forest, Robust covariance, SVMs & synthetically generated near-perfect/worst models

Series2Graph Time Series Classification:

• Employed Series2Graph to obtain time series features as graphs for UCR-2018 to produce Explainable-Graph classification
• Currently tuning the modified Graph Convolution Networks (GCNs) for Weighted-Directed Graphs using torch geometric