Tony Ha

I research about neural image/video compression methods including vector quantization, neural codec, feature tokenization, and generative reconstruction.

I develop end-to-end conversational AI systems for human-robot interaction. My work includes optimizing the system pipeline for speech-to-text, video live captioning and summarization, LLMs, text-to-speech synthesis, and face detection to enable seamless multimodal communication.

* MINIMAP: Magnetic resonance imaging as non-invasive method for the assessment of pancreatic fibrosis * Project description: NIH-funded project which involves multiple hospital institutions (Indiana University, Mayo, University of Pittsburgh...) to develop a quantitative scoring system as a biomarker of pancreatic fibrosis * Developed pancreatic image segmentation algorithm (U-Net variants) given abdominal MRI

* Project description: Develop a decision-support system for resource-allocation based on accurate workload forecasting * Developed institutional and individual-level inflow/outflow/census prediction algorithm (Attention-based MLP) to forecast the hospital workload given de-identified electronic health record (EHR)

* Worked as full-time researcher in healthcare IT field * Involved in various projects about medical informatics (ECG signal, radiology image, electronic health record, NLP in medical field, etc.) * Published several papers in medical informatics journals (ORCID #: 0000-0002-6596-5116)