AI Researcher ο½ Time-Series + Biomedical AI + Quantum ML
Harvard Medical School & BIDMC, ex-OpenAI, ex-MediaTek, ex-Max Planck Institute
Passionate about transforming data into intelligence π§ β‘οΈ
- π§ͺ Researcher at Harvard Medical School & BIDMC (AI x Surgical Informatics).
- π§ Building the EEG Foundation Model for brain signal understanding (AI x EEG x Psychiatry).
- π DeFAI Researcher at Omnis Labs (AI x DeFi x Finance).
- π± Currently exploring Quantum Machine Learning, Quant Finance, DeFAI and Brain-Computer Interfaces.
- π Reviewer for ICML, KDD, MICCAI, ICASSP, IEEE TPAMI, TNNLS, and more.
- π Google Scholar ο½ Website
| Year | Venue | Title | Citations |
|---|---|---|---|
| 2025 | PLoS ONE | Res-VMamba: Fine-Grained Food Category Visual Classification Using Selective State Space Models with Deep Residual Learning | 53 |
| 2024 | IEEE SiPS | QEEGNet: Quantum Machine Learning for Enhanced Electroencephalography Encoding | 32 |
| 2024 | arXiv | Mind's Eye: Image Recognition by EEG via Multimodal Similarity-Keeping Contrastive Learning | 22 |
| 2023 | J. Affective Disorders | Prediction of Antidepressant Responses to Non-Invasive Brain Stimulation Using Frontal EEG Signals | 18 |
| 2023 | Ann. Biomed. Eng. | Intraoperative Detection of Surgical Gauze Using Deep Convolutional Neural Network | 17 |
| 2025 | arXiv | QASA: Quantum Adaptive Self-Attention for Quantum Transformer Models | 17 |
| 2025 | ICASSP | Quantum Multimodal Contrastive Learning Framework | 16 |
| 2025 | IEEE QCE | Quantum RL Trading Agent for Sector Rotation in the Taiwan Stock Market | 5 |
π Full list β Google Scholar Β· Citations recorded on 2026-03-19
| Year | Award |
|---|---|
| 2026 | NVIDIA GTC 2026 β Top 8 Finalist, GPU-Accelerated Chinese Brain-to-Text Interface (NVIDIA H100) |
| 2026 | Berkeley RDI AgentX-AgentBeats β 3rd Place in Healthcare Agent (1,300+ teams worldwide) |
| 2023 | 20th National Innovation Award β Certificate in Clinical Research category |
| 2019 | International Blockchain Olympiad (IBCOL) β World Finalist (100+ teams) |
| 2017 | Microsoft Imagine Cup β 2nd Place in Taiwan |
| 2016 | iGEM Competition β World Gold Medal + Best Applied Design + Best Part Collection + Best Presentation (300+ teams) |
-
Res-VMamba
Fine-grained food classification using state space models with deep residual learning. PLoS ONE, 53 citations. -
MUSE EEG
Mind's Eye: EEG-based image recognition using multimodal similarity-keeping contrastive learning. 22 citations. -
QEEGNet
Hybrid quantum-classical encoder model for EEG representation learning. IEEE SiPS 2024, 32 citations. -
QASA
Quantum Adaptive Self-Attention for Quantum Transformer Models. 17 citations. -
QCL
Quantum Multimodal Contrastive Learning for EEG-image analysis. ICASSP 2025, 16 citations.
-
LCM β Large Cognition Model Pretrained EEG Foundation Model for brain signal understanding. 14 citations.
-
FreqLens
Interpretable Frequency Attribution for Time Series Forecasting.
-
Necomimi
EEG-to-Image generation using diffusion models with neural cognitive alignment. 9 citations.
- DL-based Eye Tracking System
The new Gaze Heatmap System provides advanced gaze tracking with ETH-XGaze as the default model (best accuracy, supports MPS GPU acceleration). - Webcam Gaze Tracker, Gaze Label Tool & Video Annotation Tool
A web application that uses your webcam to track your eye gaze in real-time, featuring: Gaze-based Image Labeling powered by SAM (Segment Anything Model), Video Annotation Mode for recording gaze and audio commentary on videos. Built with WebGazer.js, ONNX Runtime Web, and Vite. - FewShot3DBrain
Frozen FOMO backbone + task-specific LoRA adapters for sequential few-shot continual learning on tumor segmentation (BraTS) and brain age regression (IXI). - Food2K-TW101
Comprehensive benchmark of fine-grained food classification models on Food2K, Taiwanese Food-101, and Food2K-TW datasets.
- PyTorch Fine-Grained Classification Template PyTorch fine-grained visual classification template implementing multiple architectures (MetaFormer, PRENet, CMAL-Net) for fine-grained image recognition tasks.
- ASR-LLM-Medical Prediction Pipeline
A comprehensive two-stage pipeline for emergency medical service (EMS) call analysis, combining ASR evaluation with LLM enhancement for medical term correction, information extraction, and emergency page generation. [Related]: Firefox Crawler for Broadcastify Audio Archives. - dDTF β Dynamic Directed Transfer Function
EEG channel connectivity and directionality analysis using dynamic Directed Transfer Function in the frequency domain.
- NTQIP NEI-6 Prediction
Decision-tree-based trauma outcome prediction on the TQIP NEI-6 dataset using demographics, GCS, vitals, and injury mechanism features. - TQIP NEI-6 Local Feature Analysis
Local feature engineering and analysis pipeline for TQIP NEI-6 trauma prediction. - Trauma Activation Matcher
Automated matching of trauma activation registry data with EMS audio recordings using time-window, demographics, and injury pattern matching.
- BTSPFormer
Transformer-based time-series model inspired by behavioral-timescale synaptic plasticity (BTSP), designed for predictive learning on damped signals. - DeFi Benchmark on Curve.fi
Predict Curve Pool Virtual Price changes using Random Forest, LSTM, Transformer and QNN (Quantum Neural Network) models. - Curve Pool Intelligent Prediction System Real-time prediction of Curve pool Virtual Price changes, intelligently identifying the best investment opportunities.
- Time-Series Forecasting Benchmark Framework Curated benchmark and evaluation framework for time-series forecasting, aggregating 9+ dataset/model repos (GiftEval, Chronos, UniTS, TFB, BasicTS, etc.).
- Quantum MNIST Diffusion
Quantum-enhanced diffusion models for image generation, bridging medical imaging and quantum computing. - QuantumRWKV
Quantum-Enhanced Channel Mixing in RWKV Models for Time Series Forecasting. - Vision-QRWKV
Vision QRWKV: Exploring Quantum-Enhanced RWKV Models for Image Classification. - QRL for Industry Sector Rotation
QCE25 Quantum Reinforcement Learning agent for industry sector rotation in the stock market. - QML for Text Gen
Quantum text generation using QKSAN, QRWKV, and QASA models with comprehensive evaluation pipeline. - ViQASA β QASA Vision Transformer
Hybrid classical-quantum Vision Transformer with quantum-inspired attention mechanism using Pennylane, evaluated on MNIST and Fashion-MNIST. - Diffusion QASA
Applying Quantum Adaptive Self-Attention (QASA) to diffusion models. - Quantum Attention
Exploration of quantum attention mechanisms for neural network architectures. - Quantum Molecular Prediction
Quantum and classical regression models (QNN, QASA, Transformer, XGBoost, etc.) for predicting molecular properties.
- EEG Dataset Survey Paper
A survey of public EEG datasets for machine learning and BCI. - Psycho Gundam
EEG-based real-time robotic control system using EMOTIV+ and deep learning. - Opensource EEG Dataset
Curated table of open-source EEG datasets with detailed specs (montage, sampling rate, subjects, classes) for ML and BCI research.
- ctext.org Chinese Poem Crawler
For scraping Chinese classical texts from ctext.org. - Chinese BERT Fine-tune
A BERT Chinese model fine-tuning project based on DeepSpeed framework, trained on TMMLU+ multi-domain Chinese Q&A dataset. - Whisper Audio Translate Batch audio transcription pipeline using Whisper (large-v2/v3/turbo) for EMS audio data processing.
- EMS Audio Synthesis Pipeline Synthetic speech generation pipeline for EMS radio call ASR fine-tuning, combining LLM text generation, multi-engine TTS synthesis (XTTS/Bark/Edge), and radio-noise augmentation.
- IELTS Speaking & Writing Practice Full-featured IELTS practice web app with timed recording, local Whisper ASR transcription, automatic band score estimation, pronunciation assessment, and PDF export.
- Python 101 in zh
Simple python tutorial for beginners in Chinese. - PyTorch Medical Image Tutorial
Three deep learning models (VGG16, ResNet18, ViT) on the PathMNIST dataset. - Deep Learning Introduction
A step-by-step PyTorch tutorial for deep learning beginners. - LLM Finetune Tutorial
Comprehensive examples and scripts for fine-tuning LLMs using DeepSpeed, NVIDIA NeMo, and Unsloth. - Soft Prompt VLM Fine-tuning
Tutorial on parameter-efficient fine-tuning of Vision-Language Models using learnable soft prompts (~600K params vs full 124M). - Text-Image Search with CLIP
Text-to-image and image-to-image search system based on CLIP with fine-tuning support and multiple dataset integration. - Laser Physics AI Tutor
RAG-based AI teaching assistant for NYCU laser physics course, featuring 8 learning modes, built with Gemini 2.5 Flash, Supabase pgvector, and Vercel AI SDK.
- Solana Demo Tutorial
A basic Solana smart contract demo using Rust and TypeScript, deployed on devnet. - Solana PDA Practice
A simple Solana program that manages user data accounts with messages that can be updated through PDA. - Anchor Solana PDA
A simple Solana program built with the Anchor framework demonstrating the initialization of a Program Derived Address (PDA). - Blockchain from Scratch
A practice project building blockchain systems from the ground up. - Uniswap V3 AMM Performance Backtesting System
A backtesting system for Uniswap V3-style AMM performance on Katana L2 chain, using historical event data from the WBTC-USDC pool for backtest analysis. - Quantum vs Classical ML in AMM Trading Strategies
Quantum vs Classical ML in AMM Trading Strategies. - Compass DeFAI DEX AI-driven DEX protocol for DeFAI (Decentralized Finance + AI).
- AMM LP Simulator Interactive AMM liquidity provider simulator with real-time impermanent loss calculation, dual-perspective P&L, compounding fee yield, and farming reward depreciation modeling.
- Project Compass Web3 DeFi AI-based portfolio and risk management platform integrating classical and quantum AI models for portfolio optimization across protocols (LIDO, AAVE, Uniswap, etc.).
- Option Simulator
A mathematical model-based American options backtesting system that simulates option prices using mathematical models rather than relying on historical market data. - Smart Money Concept Demo
Interactive Smart Money Concept (SMC) analysis tool built with Gemini AI and TypeScript. - Curve AI Backend FastAPI backend service for Curve Finance pool metrics syncing, historical APY data tracking, and strategy event management with Celery workers and Prometheus monitoring.
- Aster DEX Trading Bot Flask-based trading bot for Aster DEX, implementing normal grid, logarithmic grid, and volume-based automated trading strategies.
- Dao-Agent Claude Code Skill that generates structured agent prompts using Taoist ritual metaphors, with 6 template types, multi-agent workflows, prompt linting, and token estimation.
- Shikigami Claude Code Skill inspired by Japanese onmyoji mythology for generating agent prompts, featuring 6 agent templates, 5 multi-agent formations, and composable enhancement modules.
- Moltbook AI Temple Autonomous AI agent on Moltbook social platform simulating a traditional Taiwanese temple, providing Claude-powered fortune telling, divination, and 24/7 automated community engagement.
- EQ Grader Local AI-powered grading system for elementary school EQ worksheets, using Ollama (Qwen 2.5 14B) with a three-stage evaluation pipeline, running entirely on Apple Silicon.
- BugCan Pest Counter
An IoT project for agricultural pest counting and detection.
- βοΈ Email: m50816m50816@gmail.com
- πΌ LinkedIn: Michael Chi-Sheng Chen