- MedPodGPT Research: It is surprising to see that leveraging medical podcast data into LLM significantly boosts its ability to process medical information, making it an invaluable tool for education and research [1].
- Image Quality Assessment Research: Mimicking the long-dependency perception of Human Visual System, the Non-local Modeling enhances image quality assessment [1].
- EEG Research: By cooperating with the functional topological relationship of electrodes, the decoding performance of raw EEG signals is improved [1, 2].
Source codes of all my works will be shared on GitHub, the trained models and datasets will be released on ๐ค Hugging Face, and some of my applied research on Vision and Language will be developed as Web or Android/iOS mobile apps.
Boston University: mostly Multimodal Foundation Models and Generative AI research
Topic 1: Large Language Models (LLMs):
-
๐ฅ ๐๐๐ง๐๐ก๐ฆ๐๐ซ๐ค ๐๐๐ญ๐๐ฌ๐๐ญ - GSM8K-Consistency Benchmark (available on ๐ค Hugging Face)
-
๐จ ๐๐จ๐จ๐ฅ๐ค๐ข๐ญ - PromptCraft: A Prompt Perturbation Toolkit and the released PyPI Package
-
๐งฐ ๐๐จ๐ฅ๐ฅ๐๐๐ญ๐ข๐จ๐ง๐ฌ - Quantized Large Language Models
-
๐ ๐๐๐ฉ๐๐ซ ๐๐ฎ๐ซ๐ฏ๐๐ฒ - Awesome LLM Self-Consistency
-
๐ ๐๐๐ฉ๐๐ซ ๐๐ฎ๐ซ๐ฏ๐๐ฒ - Awesome Semantic Textual Similarity (STS)
LLMs Research Topic Description ๐๐๐ ๐ ๐๐๐ผ๐น๐๐๐ถ๐ผ๐ป ๐ฏ๐ ๐๐ฑ๐๐ฒ๐ฟ๐๐ฎ๐ฟ๐ถ๐ฎ๐น ๐๐ฒ๐ฎ๐ฟ๐ป๐ถ๐ป๐ด CoEvolve LLMs: Co-Evolution of LLMs via Adversarial Learning ๐๐ถ๐ฑ๐ถ๐ฟ๐ฒ๐ฐ๐๐ถ๐ผ๐ป๐ฎ๐น ๐๐ผ๐ง Bidirectional Chain-of-Thought (Bi-CoT) Prompting ๐ฃ๐ฟ๐ผ๐บ๐ฝ๐ ๐๐ผ๐ป๐๐ถ๐๐๐ฒ๐ป๐ฐ๐ ๐ฎ๐ป๐ฑ ๐ฆ๐ฒ๐น๐ณ-๐๐ผ๐ป๐๐ถ๐๐๐ฒ๐ป๐ฐ๐ Prompt Consistency (PC) ๐๐ถ๐ป๐ฒ-๐ง๐๐ป๐ถ๐ป๐ด ๐๐๐ ๐ Fine-Tune LLaMA2 (7B) with LoRA Adapter on MetaMath
and
The Trained LoRA models for inference on ๐ค Hugging Face
Topic 2: Large Multimodal Models (LMMs/MM-LLMs): We will be deploying all our Multimodal Foundation Models on our MedPodGPT platform.
- ๐ ๐๐จ๐๐ญ๐ฐ๐๐ซ๐ - ๐๐๐๐๐จ๐๐๐๐ ๐๐ง๐ฅ๐ข๐ง๐ ๐๐ฅ๐๐ญ๐๐จ๐ซ๐ฆ and our preprint are available online!
- ๐ ๐๐๐ฉ๐๐ซ ๐๐ฎ๐ซ๐ฏ๐๐ฒ - Awesome Large Vision-Language Model (LVLM/MM-LLM)
Topic 3: Medical Imaging with AI:
- ๐ผ๏ธ ๐๐ง๐ฅ๐ข๐ง๐ ๐๐๐๐ข๐๐๐ฅ ๐๐ฆ๐๐ ๐ข๐ง๐ ๐๐ข๐ฌ๐ฎ๐๐ฅ๐ข๐ณ๐๐ซ - Online Medical Imaging Visualizer (based on Papaya)
City University of Hong Kong: mostly Computer Vision research
- ๐๐จ๐ง-๐ฅ๐จ๐๐๐ฅ ๐๐จ๐๐๐ฅ๐ข๐ง๐ - Non-local Modeling for Image Quality Assessment
- ๐๐จ๐ง-๐ฅ๐จ๐๐๐ฅ ๐๐จ๐๐๐ฅ๐ข๐ง๐ - Context-aware Non-local Compensation for Image Quality Assessment
Northeast Electric Power University: mostly Deep Learning research
- ๐ฅ ๐๐ข๐๐ซ๐๐ซ๐ฒ - EEG Deep Learning Library
- ๐๐๐จ๐ฎ๐ญ ๐๐๐ ๐๐ง๐ ๐๐๐ - EEG Motor Imagery Signals Classification via CNN
- ๐๐จ๐ง๐๐ซ ๐๐ฆ๐๐ ๐ ๐๐๐ ๐ฆ๐๐ง๐ญ๐๐ญ๐ข๐จ๐ง - Sonar Image Segmentation via Entropy Method
Philips Research: mostly Natural Language Processing research
- Medical Named Entity Recognition
- Medical Concept Mapping
- Deploy PyTorch NER Model with Flask and Docker as Web App
- Dynamic Webs Crawlering in Python
Others: