Projects — Yiming Zhang

2025 – Present

Singapore MOE AcRF Tier 2 · 1M SGD · Key Researcher

A multi-pronged effort to push PPG from a single-purpose sensor into a general-purpose modality for wearable health AI.

PPG-Sport. The first bilateral-wrist PPG dataset (48 hours) for high-intensity sports, addressing the gap in non-periodic motion artifact research.
CP-PPG. A GAN-based framework that recovers distorted waveforms caused by poor skin-sensor contact, improving signal fidelity by 40%.
Motion-aware HR estimation. A weakly-supervised motion-proxy framework that maintains high accuracy in interference-heavy racket sports.
PulseLM. The first PPG-to-Text benchmark, aligning heterogeneous physiological signals with LLMs to enable cross-modal reasoning across 12+ clinical tasks.

Outputs to date: 1 top-tier SCI paper, 3 conference papers.

Industry–Academic Collaboration · 6.02M HKD · Project Lead

Led a large-scale clinical study using smartwatches with multi-wavelength PPG and single-lead ECG.

Led a 10+ person cross-functional team and oversaw end-to-end project execution: planning, resource allocation, field coordination, quality control, and final delivery.
Designed clinical trial protocols, ethical approval documents, and collaboration agreements, collecting high-quality data from ~7,000 participants across 7 hospitals and community health centers.
Developed interpretable, lightweight, high-performance AI for accurate blood pressure monitoring and vascular condition identification.

Outputs: 2 top-tier SCI papers.

National Key R&D Program · 5M CNY · Key Researcher

Developed an end-to-end stack — from interpretable models to ring-type prototypes — for reliable long-term wearable monitoring.

Interpretable blood pressure and core-temperature estimation models grounded in ECG/PPG/ABP correlations in human microcirculation.
Adaptive calibration algorithms achieving long-term tracking gains of SBP +36.6%, DBP +32.3%, and 92% screening accuracy for cardiovascular risks.
Ring-type wearable hardware prototypes with AI models optimized on NVIDIA Jetson for real-time, low-power edge inference.

Outputs: 5 SCI papers, 1 EI paper, 3 patents.