🌿 Nature Exposure and Cognitive Aging

National Taiwan University & Academia Sinica 2021 – Present PhD Research

Project: Neural mechanisms of forest environmental information and their use in mitigating cognitive aging

  • Built virtual reality (VR) experiments in Python to measure how nature exposure benefits cognition in older adults.
  • As a visiting scholar at Brandeis University, evaluated the cognitive effects of short-term nature exposure through video.
  • Resampled raw video to the scanner's sampling period (TR = 0.752 s, 900 time points) and decomposed it into low-level visual, temporal, semantic and object-level feature series, aligned with the fMRI time series as testable regressors.
  • Signal processing: motion energy from adjacent-frame differences, Shannon entropy for spatial and temporal unpredictability, sliding windows to track semantic change, and power spectral density (PSD) to compare the temporal-frequency structure of the two stimulus types.
  • Presented at the Psychonomic Society Annual Meeting, New York (2024).

Transferable methods: signal processing and multimodal feature engineering (sampling alignment, differencing, entropy and spectral analysis) that turn unstructured signals into modelable time series. The same approach drives virtual metrology on CMP process sensor sequences and the topic-entropy time series for chat groups, where entropy predicts topic shifts in the next window.

  • Python
  • VR Experimental Design
  • Signal Processing
  • OpenCV
  • YOLOv8
  • Multimodal Time-Series

Related code ↗

🌐 Culture, Value, and Decision Making

National Taiwan University 2019 – 2021 Research Assistant

Project: Cultural influences on learning and decision making (a Taiwan–U.S. cross-national study)

  • Designed a cross-national experiment workflow in Python that runs in sync at both sites.
  • Built a database with Shell and SQL to make cross-server data sharing and version control possible.
  • Validated and calibrated cross-site multiband fMRI signals: the same participants repeated visual and motor tasks on the same model of 3T scanner in Taiwan and the U.S., and ANOVA separated participant, task and site effects. Without first correcting systematic differences between scanner sites, no later group comparison holds up.
  • Publications: Biological Psychology (2021); a chapter in Clinical Cultural Neuroscience (Oxford University Press, 2019); Cognitive Neuroscience Society Annual Meeting, Virtual (2020).

Transferable methods: cross-site data integration with batch-effect correction, plus data pipelines and QC workflows: the same problem as integrating data across fabs, tools or sources. Studying how people decide under probability and value carries straight into the multi-agent investment research system, where every judgment is a probability evaluated with Brier scores and calibration curves, not just right or wrong.

  • Python
  • Shell
  • SQL
  • fMRI Signal Calibration
  • Cross-Site Data Integration
  • Data QC Pipelines

DOI ↗

🧩 ADHD and Neuroimaging

National Taiwan University 2015 – 2018 Master's Research

Project: Differential neural processing of value during decision making in adults with ADHD and healthy controls

  • Led the team through recruitment of participants with ADHD and MRI scan scheduling.
  • Developed automated pipelines in MATLAB and R to process and analyze behavioral and fMRI data.
  • Publication: Journal of Psychiatry and Neuroscience (2023).

Transferable methods: statistical modeling and multiple-comparison correction with small, high-dimensional, imbalanced samples, and end-to-end automated analysis pipelines. These became habits in every later project: every rate in the RAG system carries a confidence interval, the quant backtests use a Deflated Sharpe that accounts for the number of trials, and conversation forecasting uses a block bootstrap to estimate the uncertainty of its gains.

  • MATLAB
  • R
  • Statistical Modeling
  • Experimental Design
  • Pipeline Development

DOI ↗

The full list is on the Publications page; results from applying these methods to industry data are on the Projects page.