wangw3943

Implemented multi-GPU deep learning pipelines for scRNA-seq representation learning using soft similarity, leveraging CUDA for acceleration on datasets exceeding 200 GB. Engineered custom regularizers for feature separation and compactness, adapted the VICReg architecture to improve cross-sample synchronization, developing and debugging in VS Code. Performed unit-testing to validate ablation experiments and contrastive learning modules. Achieved state-of-the-art performance (scIB benchmarks: biological conservation and batch effect removal) while maintaining scalability on Brown’s OSCAR HPC cluster.

Directed research on strategies to enhance reward inference in inverse reinforcement learning, implementing prototype algorithms in PyTorch and CUDA. Conducted an in-depth survey of contemporary IRL frameworks including MaxEnt IRL, Bayesian IRL, and adversarial methods. Explored links to feature construction, diffusion models, and multi-agent adversarial game theory.

Developed a C++ inference bot in Visual Studio 2022 to deduce player-held cards in the board game Clue using partial game state information. Designing a custom reinforcement learning environment (independent of Gym) for multi agent training.

Designed and deployed a VAE in PyTorch for anomaly detection in large-scale hospital electronic health records (MIMIC dataset, hundreds of GB) to flag patients with opioid overdose risk. Led stakeholder collaboration with NYU’s InnoVention program, emergency department physicians, nurses, and pain specialists to evaluate clinical integration, developing in VS Code and training on OSCAR HPC.