Skip to content

Research

I develop uncertainty-aware modeling, optimization, and control methods for systems that must infer, decide, and act under imperfect information.

My current research program focuses on mechanistically grounded machine learning for adaptive biomedical sensing and closed-loop experimentation. The central question is how a measurement or experimental system can update its beliefs about an underlying physical or biomedical process and choose the next informative action under practical constraints.

Five-stage pipeline from sparse noisy measurements through forward model, probabilistic inference, adaptive decision-making, to closed-loop experimentation, with four methodological tags and three application domains
Research program overview — from measurements to adaptive closed-loop experimentation, supported by four transferable methodological pillars.
Three-panel comparison of sensing paradigms: fixed protocols with no feedback, black-box learning with opaque models and no feedback, and the proposed closed-loop paradigm with a measure-infer-decide-act feedback loop and calibrated uncertainty
Paradigm evolution — from fixed protocols through black-box learning to the closed-loop paradigm with calibrated uncertainty and human supervision.
Research map connecting four methodological pillars (uncertainty-aware inference, adaptive decision-making, closed-loop experimentation, mechanistic ML) to three application domains (energy and cyber-physical systems, biomedical sensing, closed-loop discovery) with trajectory arrows showing the progression from prior work to current research
Research map — the shared methodological base connects published prior work in energy systems to ongoing research in biomedical sensing and closed-loop discovery.

Methodological Pillars

Uncertainty-aware inference. Probabilistic reconstruction, state estimation, and uncertainty quantification from noisy or incomplete measurements. Methods draw on Bayesian inference, variational inference, and Markov-chain Monte Carlo (MCMC). This pillar spans synthetic prosumer-response identification with SINDYc and MCMC (published), maximum-likelihood grey-box building models (published), and ongoing radar-based physiological state estimation.

Adaptive decision-making. Measurement, waveform, sampling, and experiment selection framed as sequential decision problems under uncertainty. Methods include model predictive control (MPC), optimal experimental design, and information-theoretic acquisition. The same sequential-decision framework that drove target-adjusted MPC for microgrid frequency control now drives adaptive waveform design for biomedical sensing.

Closed-loop experimentation. Systems that connect sensing, inference, and decision into a feedback loop while preserving human supervision where required. This pillar extends from an integrated control architecture demonstrated on a physical laboratory microgrid (published) to human-supervised autonomous bioassay platforms for antibacterial screening (under review).

Mechanistic machine learning. Learned representations constrained by physical models, scientific priors, and domain knowledge. Rather than treating the system as a black box, we encode known structure (conservation laws, differential equations, sensor models) into the inference or control pipeline. Examples range from SINDYc with physical basis functions for prosumer modeling to radar forward models for physiological monitoring.

These four pillars are described in more detail on the Methods page. They apply across all current and prior application domains.

Current Platforms

Biomedical sensing (ongoing)

Radar-based physiological monitoring as an adaptive sensing testbed, combining physical measurement models, inverse problems, real-time constraints, and health-relevant signals such as respiration and cardiac motion. This work is conducted as part of the guest researcher appointment at Hangzhou Institute of Technology, Xidian University.

→ Project case study: Adaptive Radar-Based Physiological Sensing

Closed-loop discovery (ongoing; manuscript under review)

AntiSyn-AI and related biomedical collaborations, where uncertainty-aware active selection prioritizes informative experimental conditions under assay cost and safety constraints. The first manuscript describing this platform is currently under review.

→ Project case study: Closed-Loop Discovery with AntiSyn-AI

Established Outputs

The following peer-reviewed publications and software outputs demonstrate the methodological base across application domains:

Output Type Year Evidence
Target-adjusted MPC for microgrid frequency control Journal paper 2019 DOI: 10.1049/iet-rpg.2019.0487
Prosumer response estimation via SINDYc + MCMC Journal paper 2020 DOI: 10.3390/en13123183
Three-level hierarchical microgrid control Journal paper 2020 DOI: 10.1016/j.epsr.2020.106758
Grey-box model of building thermal dynamics Conference paper 2020 SINTEF Proceedings, Oslo
SINDYc and MCMC framework Open-source software 2020 Zenodo: 10.5281/zenodo.3911952
Modular Energy Hub Framework Open-source software 2016 GSoC Archive

See the Publications page for the complete record, including two manuscripts currently under review. Research software outputs are catalogued on the Software page.

Prior Foundations

My earlier work in microgrid control, grey-box modeling, and energy-system optimization provides the technical foundation for this research program. The same probabilistic–control toolkit—state estimation, prediction, constrained optimization, hardware validation—developed for energy grids is being extended to domains where uncertainty is higher and the stakes are clinical.

Selected published outputs:

  • Target-adjusted MPC for microgrid frequency control — a predictive control framework that accounts for uncertain renewable generation and load. Published in IET Renewable Power Generation (2019). [DOI: 10.1049/iet-rpg.2019.0487]
  • Prosumer response estimation using SINDYc + MCMC — data-driven discovery of prosumer behavior dynamics with Bayesian uncertainty quantification. Published in Energies (2020). [DOI: 10.3390/en13123183] [Software on Zenodo]
  • Three-level hierarchical microgrid control — coordinated scheduling, frequency control, and local tracking demonstrated on a grid-connected laboratory microgrid. Published in Electric Power Systems Research (2020). [DOI: 10.1016/j.epsr.2020.106758]
  • Grey-box building thermal modeling — probabilistic identification of building heat dynamics from measurement data. Conference paper, SINTEF Proceedings (2020).
  • Modular energy-hub modeling framework — open-source optimization framework (Python, Pyomo), developed during Google Summer of Code 2016 at Empa.

All published, peer-reviewed outputs are listed with DOIs on the Publications page. Prior work is presented as transferable evidence, not as the dominant research identity. See the Applications page for the full application map.

Status of Claims

  • Published: Four peer-reviewed journal papers and three conference papers, each with DOIs or stable links as shown above.
  • Under review: Two manuscripts (AntiSyn-AI and virus-host entity mining) listed separately on the Publications page; not counted as published.
  • Ongoing: Biomedical sensing and closed-loop discovery platforms, described in present-progress language tied to current appointments.
  • Proposed: Future directions such as general autonomous scientific experimentation remain future-looking and are not presented as completed achievements.
Last updated on