ML Experimentation Platform for Time-Series Anomaly Detection
May 11, 2026
·
1 min read
Independent study. Reproducible PyTorch experiment framework evaluating 5 model architectures across 200+ time-series datasets on a Slurm HPC cluster. Fail-fast screening cut a 17 GPU-hour benchmark to 1.3 GPU-hours, with 600+ runs tracked in Weights & Biases.
Authors
Yoonmee Hwang
(she/her)
DevOps / MLOps Engineer
DevOps and Site Reliability Engineer with 5 years operating Kubernetes at
cloud-provider scale at NHN Cloud, a top-3 cloud service provider in South Korea.
Currently a Master of Software Engineering student at Carnegie Mellon University,
building CI/CD platforms and ML experimentation infrastructure.