This was very fun work from my postdoc with Eva Naumann and @jmxpearson.bsky.social. Code: github.com/project-improv/improv.git. We update regularly as we implement new kinds of adaptive experiments with collaborators. Please reach out if you run into any issues or want to suggest new features!

This opens up new experimental paradigms where hypothesis testing is iterative and efficient, and causal interventions can be timed based on ongoing dynamics that can't be known in advance. Happy to chat if you are interested in applying real-time and adaptive paradigms to your own experiments!

The platform is designed for flexibility. Researchers can easily plug in their own models and compose analysis pipelines in Python. It handles the hard backend stuff (parallelization, fault tolerance) so you can focus on the science. Works across platforms and requires minimal code to get started.

We demonstrate this across multiple scenarios: using Bayesian optimization to map neural responses, predicting future population dynamics, and implementing closed-loop optogenetic photostimulation to target neurons based on their response properties discovered during an experiment.

Free link: www.nature.com/articles/s41...