Olivier Codol
2.4K posts
Olivier Codol
@OlivierCodol
Postdoc at @Mila_Quebec @UMontreal, studying motor learning and control.
Montreal, Quebec Katılım Ekim 2017
300 Takip Edilen437 Takipçiler
Now published in @NatureNeuro!
Animals can quickly learn, and reconfigure, global relations from partial info (learn A>B, B>C, C>D, immediately know A>D).
So far, no one found a neural model to explain all observed effects.
So we asked the machine to find one for us!
Nature Neuroscience@NatureNeuro
Fully neural mechanisms for structured generalization and rapid learning identified using a meta-learning (“learning-to-learn”) approach to model relational learning (e.g. if A > B and B > C, then A > C) @ThomasMiconi and Kenneth Kay nature.com/articles/s4159…
English
Olivier Codol retweetledi
Long time coming. A very cool project that showcases the advantages of single neuron adaptation in RNNs.
#PLOSCompBio: Neural networks with optimized single-neuron adaptation uncover biologically plausible regulari ... dx.plos.org/10.1371/journa…
Props to @vgeadah and co-authors
English
Olivier Codol retweetledi
Excited to share my latest work with @DiedrichsenJorn and @andpru. In this work, we ask whether motor sequence learning is motoric at all! Check out the 🧵version of the abstract: biorxiv.org/content/10.110… 1/n
English
Olivier Codol retweetledi
Fantastic work by @mehrdadkashefi -- a deep investigation into the nature of learning motor sequences.
Mehrdad Kashefi@mehrdadkashefi
Excited to share my latest work with @DiedrichsenJorn and @andpru. In this work, we ask whether motor sequence learning is motoric at all! Check out the 🧵version of the abstract: biorxiv.org/content/10.110… 1/n
English
Olivier Codol retweetledi
📢📢📢 Big paper out from the lab today! We show how motor circuits across cortex and thalamus do sensory planning and how this improves reaching.
English
Olivier Codol retweetledi
I’ll be attending #NeurIPS2024 in Vancouver this week. Excited to meet new people and chat about comp neuro, NeuroAI, and foundation models for neuroscience. Also keen to attend the NeuroAI, @unireps and @neur_reps workshops!
English
@harrison_ritz @jpillowtime @SebastianSeung It absolutely is. A network is merely one way to implement a computation. In fact many networks would have several connection patterns that implement the same computation.
English
@jpillowtime @SebastianSeung But then you analyze the network — isn’t something like fixed point analysis zooming back out to algorithmic-level?
English
Olivier Codol retweetledi
It’s been a long time coming, but I’m thrilled to share my first research paper with @arna_ghosh , @ckaplanis1, @tyrell_turing, and Doina! Just accepted to NeurIPS 2024 (see u in Vancouver! 🇨🇦). This will be a longer thread—thanks for following along! arxiv.org/abs/2410.22133
1/11
English
Check out Jonny's latest work on a bio-plausible synaptic learning rule that yield interesting properties in artificial neural networks! It was very exciting and fun to contribute to this work.
Jonathan@JHCornford
Why does #compneuro need new learning methods? ANN models are usually trained with Gradient Descent (GD), which violates biological realities like Dale’s law and log-normal weights. Here we describe a superior learning algorithm for comp neuro: Exponentiated Gradients (EG)! 1/12
English
Olivier Codol retweetledi
For real, everyone in neuroscience and AI:
Get off this site. Elon is now using this platform to mess with American democracy. No one should be here anymore.
NeuroAI Bluesky is getting livelier every day. Please come join us.
English
The other unknown is whether this happens with different learning schemes, which could inject bias in the gradient. It wouldn't be surprising that this is more potent in forcing diff solutions as it directly alters the loss landscape, and the path taken on it during optimization
English
Interesting thread on convergence (or lack thereof) of neural solutions, with cool refs in the replies.
It seems a lot of studies focus on classification rather than control tasks, and oft simple input dynamics. I'm looking forward to seeing this further explored in the future.
David Sussillo@SussilloDavid
Is there a name for the hypothesis that task-trained neural networks and biological brains develop similar representations when they perform well, despite differences in architecture and learning methods? I.e. that optimization (whether backprop or evolution) leads to a similar solution? If not, I think we should call it The Tolstoy Hypothesis: "All well-functioning networks are alike; each poorly functioning network is poor in its own way." 🤓💅
English
Thank you to the @wusmsl for providing a great platform to exchange ideas and discuss science in-depth! Always a pleasure to interact with the folks down there, I warmly recommend
Andrew Pruszynski@andpru
No video this time around but @wusmsl was lucky to have lab alum @OlivierCodol present his recent work: "Brain-like neural dynamics for behavioral control develop through reinforcement learning". Check it out, definitely some food for thought! doi.org/10.1101/2024.1…
English
@djoshea @_mshahbazi @JonAMichaels Correct. I did that myself and you just reminded me I need to do that for the benefit of some co-authors of mine as well!
English
@_mshahbazi @JonAMichaels I think you can edit the way that individual pubs are formatted in your google scholar profile actually and add the asterisks yourself.
English
Olivier Codol retweetledi
To me, these results emphasize that scores are tools that need to be used along with domain expertise. In that view, model analyses - score-based or not- uncover to-be-tested hypotheses about how a model may capture an important brain-like mechanism. This is not a 1-d assessment.
English
@djoshea @KanakaRajanPhD @harvardmed @KempnerInst @stellatecomms And the Comics page! Loving a good neuro comic strip
English
@KanakaRajanPhD @harvardmed @KempnerInst @stellatecomms Love the comp. neuro faq and people pages. Looks amazing!
English
The new Rajan lab website is now LIVE 🚀
I’m excited to officially (re)introduce our lab as part of @harvardmed & @KempnerInst.
Special thanks to the web team @StellateComms for creating a website that showcases our discoveries & scientific vision ✨ rajanlab.com
English
@gershbrain Possibly event cameras? Which capture images using a retinomorphic approach
en.m.wikipedia.org/wiki/Event_cam…
English
I'd like to teach a paper which shows how a fact about the brain materially improved an AI system in a way that is unlikely to have been figured out by engineering alone. I haven't been able to find a single example of this. Suggestions welcome.
English
Olivier Codol retweetledi
I’ll be recruiting 1-2 graduate students for Fall 2025 to work on visual learning generalization in humans and artificial neural networks.
If you’re interested, apply!
Check out our lab website (snailab.ca) and reach out if you need more info.
RT please.
Mila - Institut québécois d'IA@Mila_Quebec
📷 Meet our student community! Interested in joining Mila? Our annual supervision request process for admission in the fall of 2025 is starting on October 15, 2024. More information here mila.quebec/en/prospective…
English
Olivier Codol retweetledi
Investigating the experimentally-verifiable impact of different credit assignment mechanisms for learning in the brain is a crucial endeavor for computational neuroscience. Here: our take for motor learnning and the RL/SL question when looking at neural representations in cortex.
Matt Perich@mattperich
Check out our new paper led by @OlivierCodol ! We use RNNs to explore possible learning rules that lead to the dynamics we see in brains during behavior. biorxiv.org/content/10.110…
English