Skip to content
View umangdobhal's full-sized avatar

Block or report umangdobhal

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
umangdobhal/README.md

Hi there 👋,

LinkedIn Banner

Umang Dobhal

ML Engineer · Diffusion Models · Sensor AI · Time-Series · Human Activity Recognition
M.Sc. Human Intelligence Systems @ Kyushu Institute of Technology


What I Work On

Real-world sensor data is noisy, imbalanced, and expensive to label. My research builds generative models — specifically diffusion models that produce synthetic time-series data realistic enough to actually improve downstream classifiers.

  • Temporal TabDDPM — Extended tabular diffusion models with Conv1D temporal adapters for multivariate accelerometer synthesis. Published in IJABC (ABC 2026).
  • WEAR Dataset Challenge — Two-stage CNN + Reservoir Computing pipeline. 5th globally, outperforming all gradient boosting baselines. ACM UbiComp HASCA 2025.
  • Smart Water Detection — Anomaly detection on a Jetson Nano using reconstructive Reservoir Computing. 96.33% F1. Fully deployed.
  • Kyushu River Forecasting — Spatiotemporal GNN (GRU + GCN) for 7-day discharge forecasting across 10 gauging stations.

Tech Stack

Python PyTorch Scikit-learn NumPy Pandas Streamlit Jupyter Git


Publications

Year Title Venue DOI / Link
2026 Extending Tabular DDPM for Time-Series Data Generation IJABC (ABC 2026) Paper
2025 Two-Stage Reservoir Computing for HAR (WEAR Challenge) ACM UbiComp HASCA 2025 Paper
2025 Sample Selection Strategy for Synthetic Gesture Data UCAmI 2025 (Springer) Paper
2024 Synthetic Skeleton Data Generation Using LLM for Nurse HAR ACM UbiComp 2024 Paper

Connect

LinkedIn Portfolio

Pinned Loading

  1. kyushu-river-stgnn kyushu-river-stgnn Public

    7-day river discharge forecasting across 10 Kyushu gauging stations using a Spatiotemporal GNN (GRU + GCN). PyTorch · Streamlit.

    Jupyter Notebook 1

  2. smart-water-detection smart-water-detection Public

    Anomaly detection for water contamination using reconstructive Reservoir Computing on a Jetson Nano. 96.33% F1. Fully deployed hardware system.

    Python

  3. SHLTutorial SHLTutorial Public

    Jupyter Notebook

  4. GazeLab GazeLab Public

    Swift

  5. umangdobhal.github.io umangdobhal.github.io Public

    TypeScript