Lab Retreat — Presenting My Research Proposal

In September, our research lab held its annual retreat. It was an intensive few days of research presentations, discussions, and planning — and also my first time formally presenting a research proposal to the group.

What's a Lab Retreat?

In Japanese research labs, a "gassyuku" (retreat) is a tradition where the entire lab gathers — usually off-campus — for a concentrated period of research presentations and discussions. Each member presents their current progress or upcoming plans, and everyone provides feedback.

It's different from the weekly seminar. The retreat is more immersive. You spend the whole day thinking about research, and the informal setting encourages deeper conversations.

My Research Proposal

I presented my research proposal, which focuses on improving software testing efficiency using generative AI techniques. Specifically, I'm exploring how diffusion models and GANs can be applied to generate better test inputs for fuzzing tools.

The presentation covered:

• Problem statement — why traditional fuzzing is inefficient at generating targeted test inputs
• Proposed approach — leveraging generative models to produce inputs that maximize code coverage
• Related work — existing research at the intersection of AI and software testing
• Preliminary plan — milestones and expected timeline

Getting feedback from lab members with different research backgrounds was incredibly valuable. Some pointed out aspects I hadn't considered, and the discussion helped refine my approach.

FOSE2025 on the Horizon

After the retreat, I also confirmed my plan to present a poster at FOSE2025 (Foundation of Software Engineering workshop) in early November. This will be my first academic presentation outside the lab, so I'm both excited and nervous.

The poster will present an early-stage version of my research — a chance to get feedback from the broader software engineering community in Japan.

Takeaways

The retreat reminded me of a few things:

• Presenting early is better than presenting perfect — feedback on rough ideas is more useful than polishing in isolation
• Different perspectives matter — someone from a completely different subfield can sometimes see the flaw (or opportunity) you've been missing
• Research is a social activity — the best ideas often emerge from conversations, not just papers

Looking forward to FOSE2025. I'll write about that experience when the time comes.

Feel free to reach out via email or GitHub if you're interested in software testing or AI for SE research.