Olivier Codol

Request for Proposals: Motor Learning & Neuromotor Ethics | Meta Quest Blog To foster innovation in this area, and to deepen our collaboration with academia, Reality Labs’ EMG Foundational Research team is pleased to invite faculty to submit their ideas to this request for re...

FUNDING OPPORTUNITY: Motor Learning and Neuromotor Ethics

Our Reality Labs EMG Foundational Research team is pleased to announce an RFP! The topic is on motor learning for EMG-based HCI and the related neuromotor ethics. Submissions are due September 1!

www.meta.com/blog/reality...

Open-sourced EMG data plus training/testing scripts to make contact with our results at 100 users for each three tasks (discrete gestures, 1D wrist navigation, and handwriting).

github.com/facebookrese...

Motor sequence learning involves better prediction of the next action and optimization of movement trajectories Learning new sequential movements is a fundamental skill for many animals. Motor sequence learning may arise from three distinct processes: (1) improved execution of individual movements independent o...

Careful dissection of motor sequence learning by @andpru.bsky.social and @mkashefi.bsky.social. In my humble opinion this should be an instant classic 😏.

www.jneurosci.org/content/earl...

Pleased to announce that our paper on "Identifying Neural Dynamics Using Interventional State Space Models" has been selected for a poster presentation in #ICML2025. Please check the thread for paper details (0/n).

Presentation info: icml.cc/virtual/2025....

Comment: The brain works at more than 10 bits per second

www.nature.com/articles/s41...

Intermittent rate coding and cue-specific ensembles support working memory - Nature By recording large populations of neurons in behaving monkeys using newly developed silicon probes, we show that intermittent periods of memorandum-specific spiking coexist with synaptic mechanis...

This paper had a big impact on me on this in admitting that maybe *some* amount of WM can be through synaptic plasticity not activity dynamics

www.nature.com/articles/s41...

1/N
How do neural dynamics in motor cortex interact with those in subcortical networks to flexibly control movement? I’m beyond thrilled to share our work on this problem, led by Eric Kirk @eric-kirk.bsky.social with help from Kangjia Cai!
www.biorxiv.org/content/10.1...

Excited for this new step in your journey! Congrats!

Very interesting abstract! Looking forward to reading more. I also wonder how much your observations would change as we move to more evolutionary complex movements than locomotion, like grasping or tying shoe lace.

All the sensorimotor processing to make the production of speech happen. Ultimately the language produced may be much lower bound but the process of creating it can be arbitrarily more complex. Note that this involves asking what is the info channel here, is it btw indivs or brain-body within indiv

Sorry I mean more

An auditor is receiving maybe 10b/s. The speaker isn’t processing 10b/s at all.

Which classical monkey reaching studies do you want replicated/investigated at mass-Neuropixel scale?

Deep RL Needs Deep Behavior Analysis: Exploring Implicit Planning by Model-Free Agents in Open-Ended Environments Understanding the behavior of deep reinforcement learning (DRL) agents -- particularly as task and agent sophistication increase -- requires more than simple comparison of reward curves, yet standard ...

Our new preprint from Rajan lab (Harvard):

"Deep RL Needs Deep Behavior Analysis: Exploring Implicit Planning by Model-Free Agents in Open-Ended Environments"

Sophisticated & sometimes insect-like planning, exploration, predator evasion, and foraging strategies by DRL.

arxiv.org/abs/2506.06981

Would have to be on temporal problems specifically (RNNs) since their purpose is to keep credit assignment temporally local. In RNNs vanishing gradients naturally achieve that. In RL I suspect gradients propagate for longer when used at all (more on that on our upcoming pre print update)

Learning Dynamics of RNNs in Closed-Loop Environments Recurrent neural networks (RNNs) trained on neuroscience-inspired tasks offer powerful models of brain computation. However, typical training paradigms rely on open-loop, supervised settings, whereas ...

All our motor control modelling efforts focus on closed-loop systems for this reason:

"...closed-loop and open-loop training produce fundamentally different learning dynamics, even when using identical architectures and converging to the same final solution."

arxiv.org/abs/2505.13567

A warning:

"Seldom do more than a few of nature’s secrets give way at one time. It will be all too easy for our somewhat artificial prosperity to collapse overnight when it is realized that the use of a few exciting words like information, entropy, redundancy, do not solve all our problems."

J'ai une question pour les francophones ici (ou n'importe qui si vous êtes bilingues):

C'est quoi la meilleure façon de dire « It was a humbling experience » ?

Il me semble qu'on n'a pas un mot qui est exactement équivalent à « humbling » en français...

Is cognition a distinct step from action? Equivalently, is sensation a distinct step as well? Those are rhetorical questions of course, and I agree with the general sentiment of complete mind-environment loops

Fred Crevecoeur
@fredericcrevec1
🚨preprint time by @harikalidindi.bsky.social
for our work on neural population dynamics: we show that features of neural population activity during reaching emerge from a simple linear body-network system 🧵👇
biorxiv.org/content/10.1...

What I like about the Turing test is that it doesn’t try to define intelligence objectively. It embraces it as a subjective experience. And indeed there are many real life situations where one would find something intelligent and not the next person. In that sense LLMs easily qualify as intelligent.

Congrats Caro, fantastic news!

GitHub - patrickmineault/bluesky-feed: An algorithmic bluesky feed that works based on my feed An algorithmic bluesky feed that works based on my feed - patrickmineault/bluesky-feed

foundses
github.com/patrickminea...

I believe @patrickmineault.bsky.social made something to that effect. I have recently been thinking about setting this up myself as well!

Big FOMO but couldn’t make it. I hope everyone has a blast!

Scaling models across multiple animals was a major step toward building neuro-foundation models; the next frontier is enabling multi-task decoding to expand the scope of training data we can leverage.

Excited to share our #ICLR2025 Spotlight paper introducing POYO+ 🧠

poyo-plus.github.io

🧵

Interested in motor control or cerebellar function? We have two openings for graduate students for Fall 25. Join the sensorimotor superlab - our interdisciplinary research group Paul Gribble and Andrew Pruszynski. Application instructions at diedrichsenlab.org. Please repost 🙏

Thanks for the shout-out!

This isn’t on the pre-print yet but will add soon. We indeed find that some networks can express edge-of-chaos dynamics. That leads them to maintain a memory of noise over time, which might increase dimensionality as noise grows as well.

Yup! As well as strong manipulability of the constraints on output by changing task goals or perturbing movements.

But generally I view holistic loss functions as an abstraction of what brains really do of course

A neural network that finds a naturalistic solution for the production of muscle activity - Nature Neuroscience How motor cortical activity relates to muscle movement is still unclear. Here the authors trained neural networks to reproduce muscle activity of reaching monkeys. The optimal solutions produced by th...

There’s a good deal of behavioral work in motor literature using optimal control models (which are conceptually very close since normative) asking this very question.

In DL there is also this classic of course

www.nature.com/articles/nn....