How do we combine "improved models of naturalistic behaviour" with "precise predictions of the neural computations"? Top-down models have been better at capturing phenomenology than making precise predictions. Should these be fit together (e.g., top-down objectives regularized by real neurons)?

Thanks for this great piece! It's kind of ironic that theory has become such a reductionist force in neuroscience. 20 years ago, theorists were a big part of the push into natural scenes (e.g., www.rctn.org/bruno/papers...).

• When:

📅 First Thursday of every month, 9:00-11:00 AM Pacific Time (starting August 7, 2025)

• Where:

📍 BWW, Berkeley (breakfast provided!) + 💻🌎 Also on Zoom

(Here's the sign up form again: forms.gle/o5DXD4WMdhTg...)

🧵[4/5]

Congrats, Matteo!

I agree with @hakwan.bsky.social that it's great for trainees to hear critical questions at meetings. These have disappeared, especially in the US. Maybe it's the drop in NIH funding rates in the 2000s. People don't want enemies. The worst are the meetings hosted by funders: people just nod & smile.

Curious what was different during active vision. Coarse to fine was pretty much entirely predicted by tuning, so not really different at all