Hannes Mehrer

Do grant proposal texts matter for funding decisions? A field experiment - Scientometrics Scientists and funding agencies invest considerable resources in writing and evaluating grant proposals. But do grant proposal texts noticeably change panel decisions in single blind review? We report...

Arguably this is 2 pages too many, given that grant decisions are mostly based on CV + abstract:

"We find that withholding proposal texts from panelists did not
detectibly impact their proposal rankings."

link.springer.com/article/10.1...

🚀 Excited to share a major update to our “Mixture of Cognitive Reasoners” (MiCRo) paper!

We ask: What benefits can we unlock by designing language models whose inner structure mirrors the brain’s functional specialization?

More below 🧠👇
cognitive-reasoners.epfl.ch

But to come back to focal neural effects of stimulation: without topography in the model I find it hard to see how to explicitly model the interaction of stimulation at multiple sites that might allow to obtain stronger behavioral effects than we observed using single-site stimulation.

Defining face-selective units in a non-topo model, increasing their activation level, and projecting that to the latents of a GAN used for image generation probably also results in face-related changes in model percepts. Is that more what you are asking?

This allows eg our visualizations, where we stimulate face-selective regions - defined using a standard face-localizer - which then lead to face-related changes to model percepts (Figs 5, 9-15).

Thanks, Adrian! The main purpose of using topo models is to allow the implementation of neural activation changes and their propagation across the cortical sheet using the topo model's in-silico equivalent of the cortex.

Absolutely agree that neural responses to the perturbations would have been very useful. While that type of data was not recorded in our experiments, the groups of Michael Beyeler @mbeyeler.bsky.social and Eduardo Fernandez just presented some great work in this direction: bsky.app/profile/mbey...

Thanks for the question, Konrad! Do you mean what percentage of variance of the neural effects of perturbations our topo models can explain? Would be great if we could investigate that, but we haven't recorded neural data during perturbation trials. Or what did you mean by observational data?

@neuroxepfl.bsky.social @icepfl.bsky.social @epfl-ai-center.bsky.social

Special thanks go out to Paolo Papale and Anna Mitola who initiated this collaboration and performed the in-vivo experiments. And to the rest of a great team: Ben Lonnqvist @benlonnqvist.bsky.social, Abdulkadir Gokce @akgokce.bsky.social, Martin Schrimpf @mschrimpf.bsky.social

Take-home-message

Proof-of-principle that topographic models can guide stimulation of high-level cortex to bias object-level behavioral choices. A step toward next-generation visual prosthetics allowing more complex visual experience.

Model perceptual changes via simulated microstimulation

We visualized perceptual changes from simulated stimulations in model face-selective regions. This results in face-related changes: additional faces appear (#1), face becomes larger (#1161), or specific face-features get enhanced (#533).

Experiment 2

With a slightly different site-selection criterion, stimulation shifted behavior above baseline in monkey 1 (Cohen’s d=0.67), though our model was not able to accurately predict monkey behavior anymore.

Experiment 1

Model-predicted behavioral shifts correlated with stimulation-evoked behavioral shifts in both monkeys. While predicted model responses were strong, monkey behavior was not shifted above baseline.

Visual stimuli via GANs

We generate image sequences that smoothly modulate neural activity along a stimulation site’s tuning dimension. This links visual input to the direction of activation changes resulting from microstimulation (Papale et al. 2024: www.biorxiv.org/content/10.1...)

How it works

1. Map the in-silico cortical sheet of a topographic model to the monkey cortex.
2. Optimize stimulation parameters by prototyping experiments in the model.
3. Only test those parameters in-vivo that are predicted to yield the largest behavioral effects.

Stimulate high-level vs early visual cortex

Visual prosthetics in early visual areas can evoke simple percepts (letters), but they are limited by 1. electrode count and 2. low-level features. We target high-level cortex to elicit percepts of more complex objects.

🧠 New preprint: we show that model-guided microstimulation can steer monkey visual behavior.

Paper: arxiv.org/abs/2510.03684

🧵

Take-home-message

Proof-of-principle that topographic models can guide stimulation of high-level cortex to bias object-level behavioral choices. A step toward next-generation visual prosthetics allowing ore complex visual experience.

Model perceptual changes via simulated microstimulation

We visualized perceptual changes from simulated stimulations in model face-selective regions. This results in face-related changes: additional faces appear (#1), face becomes larger (#1161), or specific face-features get enhanced (#533).

Experiment 2

With a slightly different site-selection criterion, stimulation shifted behavior above baseline in monkey 1 (Cohen’s d=0.67), though our model was not able to accurately predict monkey behavior anymore.

Experiment 1

Model-predicted behavioral shifts correlated with stimulation-evoked behavioral shifts in both monkeys. While predicted model responses were strong, monkey behavior was not shifted above baseline.

Visual stimuli via GANs

We generate image sequences that smoothly modulate neural activity along a stimulation site’s tuning dimension. This links visual input to the direction of activation changes resulting from microstimulation (Papale et al. 2024: www.biorxiv.org/content/10.1...)

How it works

1. Map the in-silico cortical sheet of a topographic model to the monkey cortex.
2. Optimize stimulation parameters by prototyping experiments in the model.
3. Only test those parameters in-vivo that are predicted to yield the largest behavioral effects.

Stimulate high-level vs early visual cortex

Visual prosthetics in early visual areas can evoke simple percepts (letters), but they are limited by 1. electrode count and 2. low-level features. We target high-level cortex to elicit pecepts of more complex objects.

Very happy to be part of this project: Melika Honarmand has done a great job of using vision-language-models to predict the behavior of people with dyslexia. A first step toward modeling various disease states using artificial neural networks.

My department at Emory is a hiring a tenure-track neuroscientist!

Anyone who's talked to me in the last 4 years knows I cannot say enough good things about my dept and the neuroscience community here. My colleagues are so wonderfully supportive. Postdocs, please apply!

apply.interfolio.com/174371

Are you a graduate student in #Ukraine interested in machine learning and neuroscience? My research lab at #UofT is now accepting applications for remote thesis supervision.

(1/3) #neuroAI #compneuro @vectorinstitute.ai @uoft.bsky.social @uoftcompsci.bsky.social @uhn.ca

Here is our best thinking about how to make world models. I would apologize for it being a massive 40-page behemoth, but it's worth reading. arxiv.org/pdf/2509.09737

Come join us at University of Toronto. We're hiring a Professor of computational cognitive neuroscience.

#neuroAI #compneuro jobs.utoronto.ca/job/Toronto-...