Guillaume Bellec
LightNews LightNews
banner
bellecguill.bsky.social
Guillaume Bellec
@bellecguill.bsky.social
1.1K followers 360 following 94 posts
AI, Neuroscience and Music
Posts Replies Media Videos
Pinned
bellecguill.bsky.social
Guillaume Bellec @bellecguill.bsky.social · Jan 8
Pre-print 🧠🧪
Is mechanism modeling dead in the AI era?

ML models trained to predict neural activity fail to generalize to unseen opto perturbations. But mechanism modeling can solve that.

We say "perturbation testing" is the right way to evaluate mechanisms in data-constrained models

1/8
Reposted by Guillaume Bellec
mackelab.bsky.social
Our work on training biophysical models with Jaxley is now out in @natmethods.nature.com. Led by @deismic.bsky.social, with @philipp.hertie.ai, @ppjgoncalves.bsky.social & @jakhmack.bsky.social et al.

Paper: www.nature.com/articles/s41...
Jaxley: differentiable simulation enables large-scale training of detailed biophysical models of neural dynamics - Nature Methods
Jaxley is a versatile platform for biophysical modeling in neuroscience. It allows efficiently simulating large-scale biophysical models on CPUs, GPUs and TPUs. Model parameters can be optimized with ...
www.nature.com
November 13, 2025 at 1:08 PM
bellecguill.bsky.social
Whoopsy
nthngdy.bsky.social Nathan Godey @nthngdy.bsky.social · 6d
We used the great Infinigram from Jiacheng Liu and found numerous hints of test set leakage in DCLM, which is used in OLMo-2

For instance, the fraction of MMLU questions that are leaked in pretraining had gone from ~1% to 24% between OLMo-1 and 2 😬
November 9, 2025 at 1:09 PM
Reposted by Guillaume Bellec
shahabbakht.bsky.social
I’m looking for interns to join our lab for a project on foundation models in neuroscience.

Funded by @ivado.bsky.social and in collaboration with the IVADO regroupement 1 (AI and Neuroscience: ivado.ca/en/regroupem...).

Interested? See the details in the comments. (1/3)

🧠🤖
AI and Neuroscience | IVADO
ivado.ca
November 7, 2025 at 1:52 PM
Reposted by Guillaume Bellec
initself.bsky.social
Introducing CorText: a framework that fuses brain data directly into a large language model, allowing for interactive neural readout using natural language.

tl;dr: you can now chat with a brain scan 🧠💬

1/n
November 3, 2025 at 3:17 PM
Reposted by Guillaume Bellec
neural-reckoning.org
Reminder this is happening this Wed/Thu. Free spiking neural network conference - registration required (see below).
neural-reckoning.org Dan Goodman @neural-reckoning.org · 21d
Spiking NN fans - the #SNUFA workshop (Nov 5-6) agenda is finalised and online now. Make sure to register (free) soon. (Note you can register for either day and come to both.)

Agenda: snufa.net/2025/
Registration: www.eventbrite.co.uk/e/snufa-2025...

Thanks to all who voted on abstracts!

🤖🧠🧪
SNUFA 2025
Spiking Neural networks as Universal Function Approximators
snufa.net
November 3, 2025 at 3:28 PM
Reposted by Guillaume Bellec
standehaene.bsky.social
This was a fabulous, once in a lifetime colloquium -- and now the videos are available in high quality on the College de France web site @college-de-france.fr
www.college-de-france.fr/fr/agenda/co...
With talks by Edvard Moser, Nancy Kanwisher, Liz Spelke, Manuela Piazza, Luca Bonatti and more!
November 2, 2025 at 3:08 PM
Reposted by Guillaume Bellec
arnaghosh.bsky.social
LLMs are trained to compress data by mapping sequences to high-dim representations!
How does the complexity of this mapping change across LLM training? How does it relate to the model’s capabilities? 🤔
Announcing our #NeurIPS2025 📄 that dives into this.

🧵below
#AIResearch #MachineLearning #LLM
New paper titled "Tracing the Representation Geometry of Language Models from Pretraining to Post-training" by Melody Z Li, Kumar K Agrawal, Arna Ghosh, Komal K Teru, Adam Santoro, Guillaume Lajoie, Blake A Richards.
October 31, 2025 at 4:19 PM
Reposted by Guillaume Bellec
trackingactions.bsky.social
#EPFL has an open faculty position - tenure track or tenured!! - in AI driven biophysics. Deadline is Nov 15th. Please spread the word!

www.epfl.ch/about/workin...
Assistant/associate Professor of AI-assisted Biophysics at the Ecole polytechnique fédérale de Lausanne (EPFL) and Group Leader at the Paul Scherrer Institute (PSI)
The School of Basic Sciences (Physics, Chemistry and Mathematics) at EPFL and the Paul Scherrer Institute (PSI) jointly seek to appoint a Tenure Track Assistant or tenured Associate Professor in AI-As...
www.epfl.ch
October 28, 2025 at 2:56 AM
Reposted by Guillaume Bellec
ktran.de
New #NeuroAI preprint on #ContinualLearning!

Continual learning methods struggle in mostly unsupervised environments with sparse labels (e.g. parents telling their child the object is an 'apple').
We propose that in the cortex, predictive coding of high-level top-down modulations solves this! (1/6)
June 10, 2025 at 1:17 PM
Reposted by Guillaume Bellec
trackingskills.bsky.social
The team just picked up the latest data from the Swiss National Parc. Full of goodies - find out more about the first paper from the project: eceo-epfl.github.io/MammAlps/
October 17, 2025 at 8:58 AM
Reposted by Guillaume Bellec
mschrimpf.bsky.social
Applications open for postdoc fellowships at the @epfl-ai-center.bsky.social, deadline Nov 3. www.epfl.ch/research/fun.... Great opportunity to work with our world class faculty; the fellowship is set up for collaboration between multiple labs, including my #NeuroAI group 🤖🧠🧪
EPFL AI Center and Swiss AI Initiative Postdoctoral Fellowships
The 2nd call is now open with a deadline for submissions of 3 November (17.00 CET)!Applications are encouraged from researchers at the postdoctoral level with a keen interest in collaborative, interdi...
www.epfl.ch
October 14, 2025 at 7:27 AM
bellecguill.bsky.social
Neuralink is making viral posts for bringing voice synthesis to non-verbal ALS patients. Great

And let's non forget the academic groups that have shown the way (Willett et al. 2023). And continue to lead the innovation in this field:

www.nature.com/articles/s41...
October 12, 2025 at 10:05 AM
Reposted by Guillaume Bellec
seeingwithsound.bsky.social
(YouTube) From LSD to AI with Switzerland's brain explorers www.youtube.com/watch?v=HqnA... via @trackingactions.bsky.social
From LSD to AI with Switzerland's Brain Explorers | Hello World with Ashlee Vance
YouTube video by Bloomberg Originals
www.youtube.com
October 9, 2025 at 6:06 PM
Reposted by Guillaume Bellec
leaduncker.bsky.social
Travel awards are available for undergraduate students looking to attend #Cosyne25! The application is short, and the deadline is Nov 12. @cosynemeeting.bsky.social

Application: shorturl.at/6NEyk

More info: www.cosyne.org/travel-grants

(1/2)
Travel Grants — COSYNE
Apply for COSYNE 2026 Travel Grants to support your participation in Lisbon and Cascais, Portugal. Grants are available for students, postdocs, and PIs, including programs for Childcare, Presenters, N...
www.cosyne.org
October 3, 2025 at 3:10 PM
Reposted by Guillaume Bellec
bio-emergent.bsky.social
🎉 "High-dimensional neuronal activity from low-dimensional latent dynamics: a solvable model" will be presented as an oral at #NeurIPS2025 🎉

Feeling very grateful that reviewers and chairs appreciated concise mathematical explanations, in this age of big models.

www.biorxiv.org/content/10.1...
1/2
September 19, 2025 at 8:01 AM
Reposted by Guillaume Bellec
kathyrastle.bsky.social
** New ** Seeking Postdoctoral Research Fellow (3 yrs) for ESRC project on skilled reading in Arabic. Based in my lab at @rhulpsychology.bsky.social and collaborating with @denisdrieghe.bsky.social (Soton) and Sami Boudelaa (UAE). Please share! jobs.royalholloway.ac.uk/vacancy.aspx...
Job Opportunity at Royal Holloway University of London: Postdoctoral Research Fellow
Applications are invited for a 3-year, full-time Postdoctoral Research Fellow position in the Department of Psychology at Royal Holloway, University of London.This position is funded by an ESRC resear...
jobs.royalholloway.ac.uk
October 1, 2025 at 7:07 AM
Reposted by Guillaume Bellec
modirshanechi.bsky.social
New in @pnas.org: doi.org/10.1073/pnas...

We study how humans explore a 61-state environment with a stochastic region that mimics a “noisy-TV.”

Results: Participants keep exploring the stochastic part even when it’s unhelpful, and novelty-seeking best explains this behavior.

#cogsci #neuroskyence
September 28, 2025 at 11:07 AM
Reposted by Guillaume Bellec
neural-reckoning.org
Today is the last day for submissions to the free online SNUFA spiking neural network workshop! This is a great opportunity to present your work (hundreds of participants and sometimes thousands of YouTube views per talk), and abstract submission is easy (300 words). Do it now! 🤖🧠🧪

snufa.net/2025/
SNUFA 2025
Spiking Neural networks as Universal Function Approximators
snufa.net
September 26, 2025 at 12:41 PM
Reposted by Guillaume Bellec
mjaggi.bsky.social
We're hiring again for AI research engineering roles: Join the team behind the Apertus LLM, if you share our passion to work on impactful AI that's truly open.

careers.epfl.ch/job/Lausanne...
AI Research Engineers - Swiss AI Initiative
AI Research Engineers - Swiss AI Initiative
careers.epfl.ch
September 25, 2025 at 9:08 PM
Reposted by Guillaume Bellec
fzenke.bsky.social
Truly honored (and a little overwhelmed) to see our work featured in The Transmitter's "This Paper Changed My Life." Huge thanks to @neural-reckoning.org for the kind words - and to our amazing community that keeps pushing spiking neural network research forward 🙏
September 17, 2025 at 2:50 PM
Reposted by Guillaume Bellec
neural-reckoning.org
Submissions (short!) due for SNUFA spiking neural networks conference in <2 weeks! 🤖🧠🧪

forms.cloud.microsoft/e/XkZLavhaJe

More info at snufa.net/2025/

Note that we normally get around 700 participants and recordings go on YouTube and get 100s-1000s views.

Please repost.
SNUFA 2025
Spiking Neural networks as Universal Function Approximators
snufa.net
September 16, 2025 at 9:33 AM
Reposted by Guillaume Bellec
mace-lab.bsky.social
Thrilled to share that our work is now published in Science! ✨

We found a preference for visual objects in the mouse spatial navigation system where they dynamically refine head-direction coding. In short, objects boost our inner compass! 🧭

www.science.org/doi/10.1126/...

🧵1/
September 11, 2025 at 8:14 PM
Reposted by Guillaume Bellec
meredithmeredith.bsky.social
📣 NEW -- In The Economist, discussing the privacy perils of AI agents and what AI companies and operating systems need to do--NOW--to protect Signal and much else!

www.economist.com/by-invitatio...
An illustration of me, and the headline: "AI agents are coming for your privacy, warns Meredith Whittaker The Signal Foundation’s president worries they will also blunt competition and undermine cyber-security" To put it bluntly, the path currently being taken towards agentic AI leads to an elimination of privacy and security at the application layer. It will not be possible for apps like Signal—the messaging app whose foundation I run—to continue to provide strong privacy guarantees, built on robust and openly validated encryption, if device-makers and OS developers insist on puncturing the metaphoric blood-brain barrier between apps and the OS. Feeding your sensitive Signal messages into an undifferentiated data slurry connected to cloud servers in service of their AI-agent aspirations is a dangerous abdication of responsibility. Happily, it’s not too late. There is much that can still be done, particularly when it comes to protecting the sanctity of private data. What’s needed is a fundamental shift in how we approach the development and deployment of AI agents. First, privacy must be the default, and control must remain in the hands of application developers exercising agency on behalf of their users. Developers need the ability to designate applications as “sensitive” and mark them as off-limits to agents, at the OS level and otherwise. This cannot be a convoluted workaround buried in settings; it must be a straightforward, well-documented mechanism (similar to Global Privacy Control) that blocks an agent from accessing our data or taking actions within an app. Second, radical transparency must be the norm. Vague assurances and marketing-speak are no longer acceptable. OS vendors have an obligation to be clear and precise about their architecture and what data their AI agents are accessing, how it is being used and the measures in place to protect it.
September 9, 2025 at 11:44 AM
Reposted by Guillaume Bellec
hhmijanelia.bsky.social
Become a Theory Fellow at Janelia.

🔍 Use computation & theory to tackle biological problems.
🖥️ Leverage state-of-the-art computational infrastructure
🤝 Collaborate across scientific disciplines

Apply by Nov. 3 @ https://janelia.link/theoryfellowprogram
September 9, 2025 at 3:00 PM
Reposted by Guillaume Bellec
mmitchell.bsky.social
🤖📰 Effective YESTERDAY: China has mandated a digital watermark for all AI-generated content.
www.cac.gov.cn/2025-03/14/c...
Translating in 🧵.
Chinese writing, with last phrase highlight. Something like, 第七条 互联网应用程序分发平台在应用程序上架或者上线审核时,应当要求互联网应用程序服务提供者说明是否提供人工智能生 务。互联网应用程序服务提供者提供人工智能生成合成服务的,互联网应用程序分发平台应当核验其生成合成内容标识相关材料。 第八条 服务提供者应当在用户服务协议中明确说明生成合成内容标识的方法、样式等规范内容,并提示用户仔细阅读并理解相关 理要求。 第九条 用户申请服务提供者提供没有添加显式标识的生成合成内容的,服务提供者可以在通过用户协议明确用户的标识义务和 后,提供不含显式标识的生成合成内容,并依法留存提供对象信息等相关日志不少于六个月。 第十条 用户使用网络信息内容传播服务发布生成合成内容的,应当主动声明并使用服务提供者提供的标识功能进行标识。 任何组织和个人不得恶意删除、篡改、伪造、隐匿本办法规定的生成合成内容标识,不得为他人实施上述恶意行为提供工具或者 得通过不正当标识手段损害他人合法权益。 第十一条 服务提供者开展标识活动的,还应当符合相关法律、行政法规、部门规章和强制性国家标准的要求。 第十二条 服务提供者在履行算法备案、安全评估等手续时,应当按照本办法提供生成合成内容标识相关材料,并加强标识信息共 范打击相关违法犯罪活动提供支持和帮助。 第十三条 违反本办法规定的,由网信、电信、公安和广播电视等有关主管部门依据职责,按照有关法律、行政法规、部门规章的 处理。 第十四条 本办法自2025年9月1日起施行。
September 2, 2025 at 6:48 PM