Coding Competition

IEEE IUS 2026 QUASOS Challenge                                   Mar - May 2026 

PWI-Net: Quantitative Speed-of-Sound Estimation from Ultrasound RF Data

Developed PWI-Net, a feed-forward deep-learning framework that reconstructs spatially varying speed-of-sound maps directly from multi-angle, beamformed radio-frequency ultrasound wavefields. The method provides a learned alternative to computationally intensive iterative full-waveform inversion.

My contribution: Led the deep-learning development, including direction-aware encoders adapted to axial–lateral RF sampling, multi-angle plane-wave feature fusion, and structure-aware objectives combining SSIM, gradient consistency, and geometric inclusion priors.

Team: Scott Schoen Jr. (Lead Author), Guangyi Zhang (Deep Learning Lead), and Ion Candel

Result: Accepted for a lecture presentation in the QUASOS Challenge Winners Session at IEEE IUS 2026. The symposium will be held October 4-8, 2026, in Raleigh, North Carolina.

Status: Manuscript in preparation; code release planned following publication.

NeurIPS 2025 EEG Foundation Challenge     Sept - Nov 2025

Enhancing CBraMod with Learnable EEG Rhythm Weighting

Developed an enhanced CBraMod-based EEG foundation model with learnable weighting of canonical EEG frequency rhythms. The model combines robust signal normalization, spectral representations, channel-aware attention, and masked pretraining for cross-subject EEG decoding.

My contribution: Completed the project as a solo competitor, including model design, implementation, training, experimentation, and competition submission.

Team: Solo

Result:

• 3rd of 110 teams in the warm-up phase

• 9th of 184 teams in the Challenge 2 final phase

Explainable AI in Personalized Cancer Medicine                                    Dec 2024 – Jan 2025

Developed an interpretable regression-tree approach for predicting cancer-drug response from gene-expression profiles. The project emphasized transparent decision pathways and biologically meaningful features for personalized treatment prediction.

My role: Mentor

Team: Lin Yang, Guangyi Zhang, Hansen He (PI), and Gregory Schwartz (PI)

Result: 1st on the competition’s Kaggle leaderboard.

The competition focused on explainable models for drug-response prediction, with the leading teams invited to present at IEEE SSCI 2025.

EPiC 2023 Challenge April 2023

Robust End-to-End Emotion Recognition from ECG Using Transformers

Developed an end-to-end deep-learning pipeline for predicting valence and arousal from ECG signals. The approach combined physiological-signal preprocessing, one  dimensional convolutional representations, Transformer modeling, and pretraining/retraining strategies across multiple evaluation scenarios.

Team: Guangyi Zhang and Ali Etemad (PI)

Result: 2nd Place Winner in the Emotion Physiology and Experience Collaboration Challenge.

Awards: Top Performance Award and Completion Award