I chose Harvard Medical School for two reasons. First, my co-supervisor, Assistant Professor Sai Sun, introduced me to Instructor Jing (Jill) Cai, who kindly agreed to host me. Second, after her RIEC talk in March 2024 about natural conversation, I wanted to learn these methods even though my main research is visual attention.

The most time-consuming step was the U.S. visa and other paperwork. After the visa arrived, housing and system access went smoothly. Before leaving, I prepared a Python workflow for neural data so I could start quickly at MGH. I proposed to analyze stereo-EEG (sEEG) hyperscanning data from natural conversation, and the Sydney Cash Lab approved my visit.

Because I was not an enrolled Harvard student, I could not apply for dorm housing. I found a sublet online and shared the third floor of a house with an American roommate.

I had visited Boston in 2023, so I adapted quickly. Summer was short; from October, the temperature dropped, and winter brought heavy snow. Harvard hosts many visiting students, so people in the lab were used to newcomers. The work style was flexible: I usually worked on-site 2–3 days per week and remotely on other days.

At the Thier Research Building, I joined group meetings and worked with researchers and clinicians studying epilepsy and cognition using invasive recordings. My main task was sEEG analysis for hyperscanning, linking brain activity to natural conversation. I learned about experimental design, artifact handling, and measures of coupling within and between participants. Daily international interactions also improved my scientific communication in English.

Outside the lab, I explored the Charles River area and museums. In October 2024, I went to Chicago for SfN, presented consolidated work from Tohoku University, and met researchers from many fields.

This six-month training greatly advanced my skills. I became comfortable with sEEG and built analysis pipelines for natural-conversation hyperscanning, including preprocessing and candidate metrics for within- and cross-participant coupling. The full analysis continues because the dataset is complex and the project aims for top-tier journals. Meanwhile, my Python and neuroscience skills improved a lot.

I also gained from wider exchanges: I presented at SfN 2024, and Dr. Cai highlighted our project’s direction in a BrainMap seminar, “Large Language Model-Driven Analysis of Human Language Communication: From Single Neurons to Local Field Potentials.” From August 2024 onward, this work helped me grow at the intersection of neuroscience and AI. Support from GP-DS was important and contributed to my JSPS DC2 success.

I had planned to improve my English even more, but progress was slower than I hoped. Still, the experience was valuable, and we now meet online regularly to keep pushing the project forward.