Mohammad Nikbakht

Making sense of the physical world.

ML engineer at Apple · PhD, Georgia Tech

I build machine learning that connects signals to experiences, from health sensing to intuitive interactions.

MacPaint avatar of Mohammad Nikbakht
ResearchGeorgia Tech · 2019-2024
Georgia Tech GT logo in Macintosh style

Research at Georgia Tech

PhD in Electrical & Computer Engineering · 2019-2024

Wearable systems and machine learning for health.

Google Scholar

Scholar entries
21
Citations
248
h-index
9
i10-index
9

Updated Nov 6, 2026

Selected research projects

MacPaint concept showing exactly three sensors: one vertical chest sensor and one angled sensor on each clavicle; a compact phone visualizes three stethoscope heads listening to a heart with example HR, PEP, and LVET values

SeismoNet

Multi-view cardiac sensing

Miniature synchronized wearables capture cardiac mechanics from multiple chest positions. Google Glass and a phone display SCG signals and derived measures such as heart rate, pre-ejection period, and left ventricular ejection time.

Read the paper
MacPaint concept of a tennis player wearing the KneeMS recorder on the thigh, connected by a ribbon cable to a round acoustic sensing head at the knee

KneeMS

Listening to the knee in motion

A compact wireless system that listens to knee acoustic emissions. The small recording unit sits on the lower thigh; a short ribbon cable connects it to a separate sensing head at the knee to study changes in joint health.

Best Paper · IEEE BSN 2023Read the paper
MacPaint block diagram with synthetic subject ID and cardiac timing parameters feeding a synthetic SCG generator and a seismocardiogram output

Synthetic cardiac signal generation

Transformer-based modeling of cardiac mechanics

A transformer-based generative model that produces seismocardiograms with control over cardiac timing and participant-specific morphology. The model generates physiologically varied signals while retaining control over key cardiac features.

Read the paper
MacPaint concept of a chest-worn sensor on a walking person, a seismocardiogram dominated by large walking-motion artifacts entering an SCG denoising block, and a clean seismocardiogram emerging

Cardiac signal denoising

Cleaner wearable signals during walking

A residual U-Net reduces motion noise in chest-worn seismocardiogram recordings during walking, preserving cardiac information for timing estimation and health assessment.

Read the paper