Tessera
Folgen

Karim Habashy

@krhab.bsky.social

Computational neuroscientist, machine learning, foundations of intelligence

100 Posts204 Follower82 Folgt
Marcus Ghosh@marcusghosh.bsky.social · 3w ago

Now published @commsaicomp.nature.com with @neural-reckoning.org We: 🤖 Define a taxonomy of neural network architectures (here 128 unique structures) 🦾 Train > 25,000 models on different navigation tasks #RL 🧠 Link each architecture's behaviour to its memory dynamics doi.org/10.1038/s444...

A taxonomy of recurrence. We defined a family of neural networks for exploring questions in neuroscience and machine learning. Here each network is represented by three circles joined by different connections (short, coloured lines). Networks are joined by longer, pale grey lines if they differ by one connection. Moving from left to right takes us from the simplest to the most complex network.
Marcus Ghosh@marcusghosh.bsky.social · last yr.

How does the structure of a neural circuit shape its function? @neuralreckoning.bsky.social & I explore this in our new preprint: doi.org/10.1101/2025... 🤖🧠🧪 🧵1/9

A diagram showing 128 neural network architectures.
Karim Habashy@krhab.bsky.social · 3w ago

New (and updated) preprint is now out!🔬🧠 We propose a hypothesis regarding how the genome, despite its limited information-carrying capacity, can initialize a brain with billions of neurons that comprise a diverse set of functions. Relevant to: #neuroscience #devleopmental-biology Details 👇

Factorization and spatial encodings: a hypothesis about the foundations of the genomic code

The genomic mechanisms that efficiently encode the initial architecture and synaptic connectivity of neural circuits remain poorly understood. We hypothesise that two primary mechanisms — spatial enco...

biorxiv.org

Karim Habashy@krhab.bsky.social · last mo.

A question I ask myself all the time, how complex should I make things be? "Oversimplified models can, of course, give misleading results, but excessively detailed models can obscure interesting results beneath inessential and unconstrained complexity." Dayan and Abbott (2001)

Karim Habashy@krhab.bsky.social · last mo.

In my opinion, the most sought for skills in future academic job posts might be: “Asking LLMs good questions” “Making sense of what the LLM did” “Fixing and cleaning the LLM output” 😂

Dan Goodman@neural-reckoning.org · 2mo ago

New preprint (well, very updated). 🤖🧠🧪 We find that an abstract model of neuromodulation lets spiking neural networks perform much better, particularly in challenging noisy environments, using less energy. Relevant to #neuroscience and #neuromorphic computing. 🧵👇 www.biorxiv.org/content/10.1...

Neuromodulation enhances the capability and efficiency of spiking neural networks

Spiking neurons underlie the brain’s extreme energy efficiency, and therefore have great potential in neuromorphic computing, although realising this efficiency in practice has proven challenging. We ...

biorxiv.org

Dan Goodman@neural-reckoning.org · 2mo ago

Is the brain a computer solved in 4 posts so that we never have to talk about this again (please please please). 1. It's not the same physically as a computer - it's made of meat. Hopefully this one at least is uncontroversial.

J
Jeffrey Bowers@jeffreybowers.bsky.social · 2mo ago

There is a great deal of interest in whether LLMs challenge Chomsky’s claim that human language acquisition requires some sort of innate inductive bias (e.g., some sort of universal grammar). aclanthology.org/2020.coling-... pemt.ru/wp-content/u... doi.org/10.1037/rev0...

Priorless Recurrent Networks Learn Curiously

Jeff Mitchell, Jeffrey Bowers. Proceedings of the 28th International Conference on Computational Linguistics. 2020.

aclanthology.org

Dan Goodman@neural-reckoning.org · 3mo ago

Interesting Transmitter article by Brian DePasquale on what will be the effect of the increasingly widespread use of LLMs to code up research ideas in neuroscience. Includes a somewhat sceptical commentary from me. 😉

The Transmitter @thetransmitter.bsky.social · 3mo ago

Agentic coding makes it possible to specify a neuroscience model in hours instead of months, writes @briandepasquale.bsky.social. The field risks becoming prolific but shallow—generating models faster than we can generate insights. #neuroskyence www.thetransmitter.org/the-big-pict...

Karim Habashy@krhab.bsky.social · 4mo ago

I see papers write "we use Euler" for integration in dynamical systems. Euler is inherently unstable for stable points that are center. It would be great to stress if one is using pure Euler or other modifications like Euler-Cromer or Verlet. I think this is an important detail in a papers methods.

megan peters 🧠@meganakpeters.bsky.social · 4mo ago

Neuromatch has been denied our Office of Foreign Assets Control "Iran license" renewal. We had renewed successfully before, but now our request is “inconsistent with current U.S. foreign policy.” I am absolutely gutted by this denial.

Neuromatch@neuromatch.bsky.social · 4mo ago

We have difficult news to share. Neuromatch's Office of Foreign Assets Control (OFAC) license renewal, which has allowed us to include participants residing in Iran since 2020, has been denied by the United States Government.

Dan Goodman@neural-reckoning.org · 5mo ago

Idea for grant proposal formative assessment. Very short proposal (1 page say) and reviewers are each allowed 3 questions which you get a very limited character count to respond to. Overall effort hugely reduced and lets the proposal evolve within a round rather than needing resubmission. Thoughts?

Karim Habashy@krhab.bsky.social · 5mo ago

Is it time to focus on Alignment? The rate of advancements of LLM is kind of scary. Though, I personally believe they might not lead us to human-like AGI, and I also don't advertise for them, still their rate of advancement in code databases and cybersecurity warrants caution.

Ben Williamson@benpatrickwill.bsky.social · 5mo ago

Will AI agents and automated editors be the first readers of your next academic journal submission? "The system makes preliminary evaluative judgments that humans then review. The human role shifts from doing the assessment to auditing the assessment." scholarlykitchen.sspnet.org/2026/04/08/a...

AI Rollout Is a People Problem: A Pulse on All Things AI, Part 2 - The Scholarly Kitchen

This post explores the human decisions needed in implementing AI at organizations.

scholarlykitchen.sspnet.org

Dan Goodman@neural-reckoning.org · 5mo ago

Science would be so much better if we did review (of grants and papers) constructively and collaboratively, instead of only using them to produce binary accept/reject decisions. To do that, we have to separate review processes from decisions. One idea for grants 👇

Dan Goodman@neural-reckoning.org · 5mo ago

Idea for a different way of doing project grants. Make decisions about which projects to fund based only on the questions asked, and then require successful grantees to go through a multi round post decision review process to sharpen up the methods and make sure they can answer the question.

Marcus Ghosh@marcusghosh.bsky.social · 6mo ago

Had great fun discussing some recent work and the value of toy models in neuroscience last night! Our next workshop will be: * 16.04 * 16-20:30 * At the Crick Sign up (for free) and come along! www.eventbrite.co.uk/e/understand...

Bild
Marcus Ghosh@marcusghosh.bsky.social · 6mo ago

How can we build better experiment-theory collaborations in neuroscience? We're hosting a series of 6 in-person, interactive workshops to discuss this. Come along! www.eventbrite.co.uk/e/understand...

Karim Habashy@krhab.bsky.social · 6mo ago

Hot take: I prefer if subfields of machine learning not use the key-query terminology (as a mental state) in every possible chance. I think, it biases thinking towards a lower-D subspace and limits our ability to draw more general conclusions and analogies. For example, see contrastive learning.

Karim Habashy@krhab.bsky.social · 6mo ago

On first thought, I have this feeling that Chaos and truly random events (like radioactive decay) will be provide predictive challenges to any artificial super intelligence.🤔