
Can we find shared neural spaces across subjects? Yes! We introduce Multi-encoder-decoder-VAE (MED-VAE), a method for model-guided cross-subject shared spaces. Fantastic work led by @pangeliki #CCN2026 (selected for a talk) 🧵 1/6

Computational neuroscientist bringing machine and neural learning closer together @OxNeuro @OxfordDPAG | #firstgen | @ERC_Research Grantee 🇵🇹🇪🇺🇬🇧🌳

Can we find shared neural spaces across subjects? Yes! We introduce Multi-encoder-decoder-VAE (MED-VAE), a method for model-guided cross-subject shared spaces. Fantastic work led by @pangeliki #CCN2026 (selected for a talk) 🧵 1/6

Out now in Nature Communications nature.com/articles/s41467-0…
"Prefrontal cortex and striatum dissociate modality and learning order in cross-modal sensorimotor learning" with: @da_song0808 bioRxiv: biorxiv.org/content/10.1101/… (and at SfN, poster 451.06)

We @bold_lab_ai are hiring a large cohort of home and overseas students for Fall 2027 entry. If you’re excited about discovering the next paradigm of open-source AI and want to work on ambitious, exploratory, but scalable ML, please apply. More application details and exact timeline coming soon, in the meantime please register your interest here bold-lab.ai/opportunities/ 🦋🦋🦋

CCN 2026 paper: arxiv.org/abs/2606.15989 Code: github.com/pangelu9/MEDVAE_N… CCN talk: youtu.be/7csqgzhIXNY?si=1K79… Together with @achterbrain @ten_photos 6/6

Adding a new subject is cheap: we freeze the shared space and train only a new subject encoder/decoder on their own images. With 20% of the new subject's data, we recover ~98% of the alignment reached with all of it, and 90% of the noise ceiling in self-reconstruction. 5/6

The shared space clusters by image category, inherited from the ResNet scaffold. This is not a property of the RN-50 architecture alone: Untrained ResNet-50: retrieval drops from 46.2% to 6.6%. Shuffled neural data–model feature 1:1 correspondence: 1.0% (chance 0.8%). 4/6

Our method outperforms previous approaches: Top-1 retrieval (another person's closest latent is the same image): 46.2% vs 21.8% (SRM), 18.4% (Procrustes) Representational similarity analysis (RSA) across subjects' latent representations: 0.63 vs 0.41 3/6

In our MED-VAE architecture a common space arises by using a pretrained ResNet-50 as a scaffold onto which subjects brain activity can align to. Our method does not require common inputs across individuals and after training, the Resnet is no longer needed. 2/6

[Hugging Face release] We’re releasing the dataset behind Reason to Play! I’m incredibly proud of our team for reaching this milestone. @sreejan_kumar Austin Andrews @LHuntNeuro @summerfieldlab Josh Tenenbaum @somnirons @marcelomattar @momchiltomov

TL;DR: This is one of the most important and exciting opportunities in AI on the planet - please read on. The British Open-ended Learning & Discovery Lab is creating the perfect place for paradigm breaking AI research in the name of open-source and open-science. We have agency, we funding, we have unprecedented amounts of compute*, but WE NEED YOU! ..and we have created the dream job for you: The BOLD Fellow. This job combines a fast-moving, high agency, collaborative environment with full academic freedom and a salary that pays the bills. Apply by noon UK time on the 15th of September for this once in a lifetime opportunity to shape the history of our field and of our planet: my.corehr.com/pls/uoxrecruit… *by academic standards

Two major announcements for the #DigitalBrainProject: 🥇9 teams were selected in the first round 🚀we’re scaling the funding to $15M More details below 👇: digitalbrainproject.org

This recent paper posted by @MillerLabMIT got my attention. It shows how abstraction could emerge in PFC. If you are in CS/AI, it's a fun one. The paper is hard, but the main idea can be reduced to a low-d space, so I made a little animation. Here is my toy model and notes:🧵

1/9 How does the prefrontal cortex learn to think abstractly? A new Nature Neuroscience study (Wójcik et al., 2026) tracked macaque PFC during rule learning nature.com/articles/s41593-0… 🧵

Work brilliantly led by @WillGreedy ! Code available here: github.com/neuralml/BurstCCN…

9/9 Note: This work brings together the best of the multiplexing ideas by @NeuroNaud @tyrell_turing w/ our interneurons-for-backprop work (Sacramento 2018) to achieve: scalable (multiplexing) and highly plausible (interneurons + many other features) deep cortical learning.

8/9 Overall, our work shows how cell-type-specific cortical feedback enables AI-like hierarchical credit assignment, bridging synaptic plasticity, dendritic-circuit computations and behaviour. It opens many exciting avenues!