Sunday, May 17, 2026, 12:45 – 2:15 pm, Blue Heron

Speakers: Ruolin Wang, Mayukh Deb, Apurva Ratan Murty

This satellite event will show how in-silico experimentation can be used as practical framework for benchmarking computational models of the human brain. The core idea is simple: many experiments in vision science can now be instantiated entirely in-silico. While the possibility of running in-silico studies is widely discussed, the systematic re-simulation of past experiment studies (using the original stimuli and analyses etc) has been lacking. This satellite event builds directly on our recent efforts (Wang, Deb et al. in prep) that show how prior experiments can be replayed in silico and used to evaluate which findings replicate across candidate brain models and which do not.

The session will be divided into three parts. In the first session (~1h), we will introduce the in-silico replication framework (30 min) and walk through a complete experimental simulation pipeline (30 min). We will show concretely how one should go about simulating studies in models, how prior studies can be replicated computationally, and discuss best practices for quantifying replication success across models. The second session (~30 min) will be hands-on. Participants will be invited to run their own experiments in silico using a drag-and-drop executable modeling framework which we will introduce. Attendees are encouraged to test prior results, current experimental designs, or even untested ideas on several computational models. This session is designed as a period of open exploration In the final session (~30 min), participants will discuss their findings in an open forum. We will identify and celebrate successful replications. But even more importantly, we hope to expose gaps and loopholes that current models exploit. By drawing on clever experimental manipulations from the cognitive vision neuroscience community, this session aims to stress-test (and even break) models in ways that are theoretically informative. Take together, the session aims to turn in-silico replications on computational models into a scalable community-driven adversarial tool for accelerating theory building and model development in vision science.

We additionally would like to point out another VSS Satellite event called “Re:Vision: A community replication and generalization initiative for fMRI research of visual stimuli” that will complement the model benchmarking session.