Adrian Duszkiewicz

@adrian-du.bsky.social

Systems neuroscientist | Working out how the ๐Ÿง  generates ๐ŸŒ to find its ๐Ÿงญ | Lecturer (Asst Prof) at the University of Manchester | Big fan of ancient things ๐Ÿบ๐Ÿ“œ ๐Ÿ›๏ธ

A really interesting new two-photon miniscope study for all the retrosplenial geeks out there! Anterior and posterior retrosplenial cortex employ distinct strategies for egocentricโ€“allocentric transformation in spatial coding

PNAS

Proceedings of the National Academy of Sciences (PNAS), a peer reviewed journal of the National Academy of Sciences (NAS) - an authoritative source of high-impact, original research that broadly spans...

pnas.org

Happy to say our head-direction-cell-in-3D is finally out! doi.org/10.1038/s420... Confirms that the brain's compass uses a dual-axis rule (like an iphone does) to maintain a constant estimate of horizontal heading even when the head is not horizontal. Blogpost explanation here: shorturl.at/9VdGM

How the brain adapts its 2D compass for a 3D world

The brainโ€™s neural compass seems 2D, detecting only left and right head-turns. However, if the head is tilted in 3D, its rotation around the vertical axis also covertly changes its horizontal facing ...

communities.springernature.com

cool review on a developmental approach to #NeuroAI Turns out the best benchmark might just be our kids ๐Ÿ‘ถ

Vlad Ayzenberg@vayzenb.bsky.social ยท 2w ago

Excited to share our review in @cp-neuron.bsky.social with @lauriebayet.bsky.social and @mickbonner.bsky.social! We describe how implementing principles from child development can advance the mechanistic plausibility and capacities of AI models We packed A LOT into this review, here's a quick ๐Ÿงต

Our Impact Scholars @neuromatch.bsky.social project on ๐—ต๐—ผ๐˜„ ๐—ฟ๐—ฒ๐˜„๐—ฎ๐—ฟ๐—ฑ ๐˜ƒ๐˜€ ๐—ฒ๐˜…๐—ฝ๐—ฒ๐—ฟ๐—ถ๐—ฒ๐—ป๐—ฐ๐—ฒ ๐˜€๐—ต๐—ฎ๐—ฝ๐—ฒ ๐—ป๐—ฒ๐˜‚๐—ฟ๐—ฎ๐—น ๐—ฝ๐—ผ๐—ฝ๐˜‚๐—น๐—ฎ๐˜๐—ถ๐—ผ๐—ป ๐—ฑ๐˜†๐—ป๐—ฎ๐—บ๐—ถ๐—ฐ๐˜€ ๐—ถ๐—ป ๐—บ๐—ผ๐˜‚๐˜€๐—ฒ ๐˜ƒ๐—ถ๐˜€๐˜‚๐—ฎ๐—น ๐—ฐ๐—ผ๐—ฟ๐˜๐—ฒ๐˜… is out! โ€จ With ๐—๐—ผ๐—ต๐—ป ๐— ๐—ฎ๐—ฑ๐—ฟ๐—ถ๐—ฑ-๐—–๐—ฎ๐—ฟ๐˜ƒ๐—ฎ๐—ท๐—ฎ๐—น @johnmadridcarvajal.bsky.social, ๐—ž๐—ฎ๐˜๐—ท๐—ฎ ๐—ž๐—ฎ๐˜‚๐—ฟ๐—ถ๐—ป๐—ธ๐—ผ๐˜€๐—ธ๐—ถ, ๐—œ๐˜€๐—ต๐—ฎ๐˜๐—ฝ๐—ฟ๐—ฒ๐—ฒ๐˜ ๐—ฆ๐—ถ๐—ป๐—ด๐—ต, ๐—ฅ๐—ผ๐—ต๐—ถ๐˜ ๐—๐—ฒ๐˜€๐˜„๐—ฎ๐—ป๐˜๐—ต. ๐Ÿง  ๐Ÿ‘‰ zenodo.org/records/2064... ๐Ÿงต๐Ÿ‘‡ 1/8

Schematic of the experimental setup and our analysis. Mice experienced visual corridors containing leaf or circle patterns, with reward available only for one pattern in the rewarded cohort and no reward in the unrewarded cohort. Large-scale calcium imaging sampled neurons across V1 and higher visual areas before and after learning. Neural activity was organised into neuron ร— time ร— trial tensors and analysed with tensor component analysis. Example factors show differences between rewarded and unrewarded mice before versus after learning, including stronger separation of leaf and circle trials after rewarded learning.

The human brain is strikingly modular: distinct networks for language, formal reasoning, social reasoning, physical reasoning. Is this fundamental to intelligent systems, or an accident of evolution? In our new preprint, we find the same modular organization emerges in LLMs.