Research portfolio · Share page

Wearable health sensing, physiological signals, and reliable deployed streams.

I work on wearable-compatible physiological AI: PPG / BVP, rPPG, BCG, ECG, sleep and cardiorespiratory signals, motion-robust modeling, and confidence-aware reporting for mobile health systems.

Download portfolio PDFView project evidence

01

CITRUS thesis · Wearable PPG/BVP

Motion-robust wearable heart-rate streams

CITRUS studies why ring PPG/BVP heart-rate monitoring fails under motion, then decomposes the deployed stream into candidate disambiguation, causal decoding, reliability estimation, and accept / hold / reject reporting.

  • Candidate-based HR estimation rather than direct scalar regression.
  • Causal Viterbi decoding using current and past windows only.
  • Reliability-aware reporting for coverage-error trade-offs.

Experimental architecture and comparison roles

01

RingDatasetV2

primary benchmark

02

PPG-DaLiA

external wrist-BVP validation

03

Sports cohort

high-motion boundary

04

MMPD / rPPG

cross-modality boundary

Layer 1

DSP candidates

Layer 2

CITRUS scorer

Layer 3

Causal Viterbi

Layer 4

Reliability layer

Layer 5

Accept / hold / reject

Candidate selection replaces direct scalar regression for corrupted PPG/BVP windows.

Causal decoding uses current and past windows for real-time deployment.

Reliability reporting separates full-coverage error from usable emitted streams.

54
participants
49.76 h
ring recordings
10.91 BPM
motion MAE
0.898
error-risk AUC

02

Scientific Data 2026 · HealthRing dataset

HealthRing: Physiology Dataset for Health Sensing on Rings

My Scientific Data paper presents HealthRing, a ring-based physiology dataset for wearable health sensing. It grounds the portfolio in published dataset work: optical ring layouts, synchronized PPG and accelerometry, and clinical references for cardiovascular and respiratory signals.

  • Published in Scientific Data on 2026-06-10.
  • DOI: 10.1038/s41597-026-07289-x.
  • Dataset work connecting hardware, protocol design, physiological references, and model evaluation.
Nature article→DOI→
HealthRing ring hardware and sensor layout
Sci Data
journal
2026
published
PPG/ACC
signals
HR/RR
references

03

BIBM 2026 draft · Sleep staging

How far can wearable-compatible signals go?

A controlled decomposition of non-EEG sleep staging. The work asks what consumer and laboratory cardiorespiratory signals can honestly recover about sleep architecture, and when a system should abstain.

  • Four layers: signal source, physiological representation, temporal prior, decision.
  • Confidence-aware reporting separates reliable epochs from uncertain epochs.
  • Frames consumer sleep staging as physiological trend monitoring, not EEG-equivalent scoring.
Four-layer framework for non-EEG sleep staging
0.253
naive non-EEG kappa
0.485
physiology-aware kappa
0.616
kappa at 50% coverage
195
SHHS subjects

Knowledge base, CV, and sharing

This page is the shareable layer. The PDF is a compact attachment for advisor outreach; the CV page remains the formal resume; the research notes and project pages form the longer-running knowledge base.

Portfolio PDF →CV page →Knowledge notes →