Jascha Achterberg

Giacomo's commentary was in response to this great recent paper by Iqbal et al., also in PNAS:

"Biologically grounded neocortex computational primitives implemented on neuromorphic hardware improve vision transformer performance"
www.pnas.org/doi/10.1073/...

Enjoyed @giacomoi.bsky.social commentary in PNAS on how #NeuroAI and Neuromorphic Engineering should come together to allow brain circuit motifs to positively influence the design of computing systems:

Biological fidelity: The engine driving the neuromorphic renaissance
www.pnas.org/doi/full/10....

Let me know if you are in San Diego for #Neurips and wanted to chat about #compneuro / #neuroAI and neuroscience-inspired computing!

How do brain areas control each other? 🧠🎛️

✨In our NeurIPS 2025 Spotlight paper, we introduce a data-driven framework to answer this question using deep learning, nonlinear control, and differential geometry.🧵⬇️

A beautiful summary of our paper! Thank you @neurosock.bsky.social

One example of how multiplexing might be implemented in the brain was shown in the great work by Thomas Akam with Dmitri Kullmann, a decade ago.

The papers are cited well but the general multiplexing idea never really took the field by storm as much as it deserved

www.nature.com/articles/nrn...

Postdoc fellowship opportunity for ECRs (<3 yrs post-PhD). Note that if you want to apply to work with me as your mentor, our dept has an internal deadline of Dec 4th so please email me asap. Our internal process is shorter than the full application. 🤖🧠🧪

royalcommission1851.org/fellowships/...

Brain-Like Processing Pathways Form in Models With Heterogeneous Experts Examples of such pathways can be found in the interactions between cortical and subcortical networks during learning, or in sub-networks specializing for task characteristics such as difficulty or mod...

This new model opens a whole new world of analysing multi region interaction across trials and tasks! More analysis and findings can be found in our paper linked below. Work lead by Jack Cook, and with great help from @danakarca.bsky.social and @somnirons.bsky.social !

arxiv.org/abs/2506.02813

We also find that while complex regions are needed to learn complex tasks, these tasks are eventually moved toward simpler regions, similar to how you may struggle the first time when learning a new skill, but slowly get better with practice.

Furthermore, we find that these pathways mirror our expected behavior of pathways in the brain! We find that difficult tasks need to be learned in more complex regions, similar to how you need to think “harder” when learning how to solve a difficult math problem.

With these three features in place, we find that our third criterion of distinct pathways is also met. While baseline models exhibit largely random expert usage patterns, our models exhibit highly structured pathways between regions that reliably emerge during learning.

Our third contribution is expert dropout. Without this feature, we find models suffer large performance deficits when experts outside of the active pathway are disabled. However, we would want models to be primarily dependent on the experts that are most being used.

When put together, these two contributions resulted in remarkable pathway consistency in our model, which we measured by correlating the routing patterns across 10 different models trained on the same tasks.

We then identify three inductive biases that yield pathways that meet each of these criteria.

The first of these is a routing loss that penalizes the use of more complex experts during training, and the second scales this loss by the model’s performance on the task being solved.

We then set three criteria to determine whether pathways had formed:

(1) Consistency: Models trained on the same tasks should have similar pathways

(2) Self-sufficiency: Pathways should be primarily reliant on their own experts

(3) Distinctness: Many distinct pathways should be present

We first needed to create a model in which we could study pathway formation. We chose a Heterogeneous Mixture-of-Experts architecture, in which information can be dynamically routed to computational experts, or regions, of varying sizes.

We train model on 82 tasks of varying complexity (ModCog)!

Brains have many pathways / subnetworks but which principles underlie their formation?

In our #NeurIPS paper lead by Jack Cook we identify biologically relevant inductive biases that create pathways in brain-like Mixture-of-Experts models🧵

#neuroskyence #compneuro #neuroAI
arxiv.org/abs/2506.02813

All good Dan!

Check out this cool new work lead by @pengfei-sun.bsky.social !

With my great advisors and colleagues, @achterbrain.bsky.social @zhe @danakarca.bsky.social @neural-reckoning.org, we show that if heterogeneous axonal delays (imprecise) can capture the essential temporal structure of a task, spiking networks do not need precise synaptic weights to perform well.

Faculty Position in Neuroscience The School of Life Sciences at EPFL invites applications for a Tenure Track Assistant Professor position in Neuroscience. At EPFL researchers develop and apply innovative technologies to understand br...

Come be our colleague at EPFL! Several open calls for positions 🧪🧠🤖

* Neuroscience www.epfl.ch/about/workin... (deadline Oct 1)

* Life Science Engineering www.epfl.ch/about/workin...

* CS general call www.epfl.ch/about/workin...

* Learning Sciences www.epfl.ch/about/workin...

Statistical methods for dissecting interactions between brain areas The brain is composed of many functionally distinct areas. This organization supports distributed processing, and requires the coordination of signals…

Statistical methods for dissecting interactions between brain areas

www.sciencedirect.com/science/arti...

ATTN🚨: I will be looking for PhD students through NYU's Center for Data Science PhD program this year. Applicants should have an interest in either NeuroAI (specifically biological attention or AI interpretability) or ML for Remote Sensing. Visit my lab website for more info: lindsay-lab.github.io

The difficult constellation image is solved by generating candidate solutions with a GAN and refined using a genetic search conditioned on best fitting of the solution outline to the dots on the constellation image.

Could we understand vision as a type of problem-solving? In this new paper, we develop a computational model that iteratively refines the hypothesis about the visual input with evolutionary search.

www.biorxiv.org/content/10.1...

work led by @tarunkhajuria.bsky.social

#visionscience #neuroAI

This is really funny…

I find your point about probabilistic definition interesting -- never seen such a definition of it, but that could neatly link to my 'usefulness' framing, as for any sort of expected value computation you would need to take 'likelihood given context' into account.

Now the usefulness in program generation might sometimes align with policy compression, but that depends a lot on the given time horizon one assumes for the definition of 'usefulness'.

It also does not 100% align with my reading of it, but I found it an interesting angle. I think I find myself, naturally, being influenced by Alan Newell's take on it (which is the one John Duncan tends to reference), which is aimed at usefulness in program generation.

Can large language models play simple arcade games? Kind of. Sometimes. Slowly, and not as well as a simple search algorithm. And only if you format the input right. Of course, we made a benchmark to investigate this in more detail, because that's what we do. Paper here:
arxiv.org/html/2508.08...