Friedemann Zenke

@fzenke.bsky.social

Computational neuroscientist at the FMI. www.zenkelab.org

Can we match self-supervised backpropagation using local learning rules? We show it is possible in our new paper accepted by ICML. We achieve: 1. theoretical equivalence to BP in a controlled setup 2. new SOTA for local learning across image datasets 3. same performance as BP on multiple datasets

Bild

Our latest publication grapples with how the brain could implement gradient descent by sending learning targets top-down, gating plasticity with dendritic inhibition, and updating synaptic weights with biologically observed learning rules like BTSP. www.cell.com/cell-reports...

Cellular and subcellular specialization enables biology-constrained deep learning

Galloni et al. introduce “dendritic target propagation”: a Dale’s law-compliant learning algorithm for cortical microcircuits with soma- and dendrite-targeting inhibition and realistic connectivity co...

cell.com

Aaron Milstein@neurosutras.bsky.social · last yr.

New #NeuroAI #compneurosky preprint! To better understand how target-directed learning works in the brain, we sought to engineer an artificial neural network capable of solving complex image classification tasks that comprises only experimentally-supported biological building blocks. (1/15)

First preprint from the lab! Using intracellular recordings & analysis of 2-photon imaging data, we show that spiking & neuromodulatory input during experience drive a reorganization of visuomotor inputs in V1 layer 2/3 neurons, consistent with enhanced visuomotor cancellation - bioRxiv link below.

Bild

Congrats to Fabian Mikulash, a postdoc in the @fzenke.bsky.social lab, for being awarded a Marie Skłodowska-Curie Actions fellowship! His project aims to develop a new theory—tested with real brain data—explaining how neurons decide when to trust what we see versus what we expect 🧠

Bild

Our work with @georgkeller.bsky.social on testing predictive processing (PP) models in cortex is out on biorvix now! www.biorxiv.org/content/10.6... A short thread on our findings and thoughts on where we should move on from PP below.

A functional influence based circuit motif that constrains the set of plausible algorithms of cortical function

There are several plausible algorithms for cortical function that are specific enough to make testable predictions of the interactions between functionally identified cell types. Many of these algorithms are based on some variant of predictive processing. Here we set out to experimentally distinguish between two such predictive processing variants. A central point of variability between them lies in the proposed vertical communication between layer 2/3 and layer 5, which stems from the diverging assumptions about the computational role of layer 5. One assumes a hierarchically organized architecture and proposes that, within a given node of the network, layer 5 conveys unexplained bottom-up input to prediction error neurons of layer 2/3. The other proposes a non-hierarchical architecture in which internal representation neurons of layer 5 provide predictions for the local prediction error neurons of layer 2/3. We show that the functional influence of layer 2/3 cell types on layer 5 is incompatible with the hierarchical variant, while the functional influence of layer 5 cell types on prediction error neurons of layer 2/3 is incompatible with the non-hierarchical variant. Given these data, we can constrain the space of plausible algorithms of cortical function. We propose a model for cortical function based on a combination of a joint embedding predictive architecture (JEPA) and predictive processing that makes experimentally testable predictions. ### Competing Interest Statement The authors have declared no competing interest. Swiss National Science Foundation, https://ror.org/00yjd3n13 Novartis Foundation, https://ror.org/04f9t1x17 European Research Council, https://ror.org/0472cxd90, 865617

biorxiv.org

The hippocampal map has its own attentional control signal! Our new study reveals that theta #sweeps can be instantly biased towards behaviourally relevant locations. See 📹 in post 4/6 and preprint here 👉 www.biorxiv.org/content/10.6... 🧵(1/6)

Attention-like regulation of theta sweeps in the brain's spatial navigation circuit

Spatial attention supports navigation by prioritizing information from selected locations. A candidate neural mechanism is provided by theta-paced sweeps in grid- and place-cell population activity, which sample nearby space in a left-right-alternating pattern coordinated by parasubicular direction signals. During exploration, this alternation promotes uniform spatial coverage, but whether sweeps can be flexibly tuned to locations of particular interest remains unclear. Using large-scale Neuropixels recordings in freely-behaving rats, we show that sweeps and direction signals are rapidly and dynamically modulated: they track moving targets during pursuit, precede orienting responses during immobility, and reverse during backward locomotion — without prior spatial learning. Similar modulation occurs during REM sleep. Canonical head-direction signals remain head-aligned. These findings identify sweeps as a flexible, attention-like mechanism for selectively sampling allocentric cognitive maps. ### Competing Interest Statement The authors have declared no competing interest. European Research Council, Synergy Grant 951319 (EIM) The Research Council of Norway, Centre of Neural Computation 223262 (EIM, MBM), Centre for Algorithms in the Cortex 332640 (EIM, MBM), National Infrastructure grant (NORBRAIN, 295721 and 350201) The Kavli Foundation, https://ror.org/00kztt736 Ministry of Science and Education, Norway (EIM, MBM) Faculty of Medicine and Health Sciences; NTNU, Norway (AZV)

biorxiv.org

I’m very grateful to the FMI, the tenure committee, inspiring colleagues, and all the hidden supporters who made this possible. Huge thanks to past and present group members for their curiosity and creativity. Excited for the next chapter.

FMI science@fmiscience.bsky.social · 8mo ago

Huge congratulations to @fzenke.bsky.social on his promotion to Senior Group Leader! His team uses AI-powered models of neural networks to uncover how dynamic brain connections enable new memories. Learn more about his research in this video👇 www.youtube.com/watch?v=8xDu...

Finally got the job ad—looking for 2 PhD students to start spring next year: www.gao-unit.com/join-us/ If comp neuro, ML, and AI4Neuro is your thing, or you just nerd out over brain recordings, apply! I'm at neurips. DM me here / on the conference app or email if you want to meet 🏖️🌮

Bild

Excited to see the paper fully published. It's an important milestone for training SNNs with exact gradients, replacing our earlier tricks of a "delay line augmentation" to capture temporal relationships. Delays can now be learnt alongside weights naturally. Amazing work @mbalazs98.bsky.social !

Balázs@mbalazs98.bsky.social · 8mo ago

Our paper on event-based delay learning is now published! @neworderofjamie.bsky.social @drtnowotny.bsky.social TL;DR: It’s now possible to train synaptic delays in large-scale spiking neural networks with high temporal precision—even in recurrent connections. www.nature.com/articles/s41...