Hugo Ninou

@hugoninou.bsky.social

I am a PhD student working at the intersection of neuroscience and machine learning. My work focuses on learning dynamics in biologically plausible neural networks. #NeuroAI

1/10 🚨 New preprint: Using Large Language Models to Estimate Belief Strength in Reasoning 🚨 When asked: "There are 995 politicians and 5 nurses. Person 'L' is kind. Is Person 'L' more likely to be a politician or a nurse?", most people will answer "nurse", neglecting the base-rate info. A 🧵👇

Abstract

Accurately quantifying belief strength in heuristics-and-biases tasks is crucial yet methodologically challenging. In this paper, we introduce an automated method leveraging large language models (LLMs) to systematically measure and manipulate belief strength. We specifically tested this method in the widely used “lawyer-engineer” base-rate neglect task, in which stereotypical descriptions (e.g., someone enjoying mathematical puzzles) conflict with normative base-rate information (e.g., engineers represent a very small percentage of the sample). Using this approach, we created an open-access database containing over 100,000 unique items systematically varying in stereotype-driven belief strength. Validation studies demonstrate that our LLM-derived belief strength measure correlates strongly with human typicality ratings and robustly predicts human choices in a base-rate neglect task. Additionally, our method revealed substantial and previously unnoticed variability in stereotype-driven belief strength in popular base-rate items from existing research, underlining the need to control for this in future studies. We further highlight methodological improvements achievable by refining the LLM prompt, as well as ways to enhance cross-cultural validity. The database presented here serves as a powerful resource for researchers, facilitating rigorous, replicable, and theoretically precise experimental designs, as well as enabling advancements in cognitive and computational modeling of reasoning. To support its use, we provide the R package baserater, which allows researchers to access the database to apply or adapt the method to their own research.

🚨New spotlight paper at Neurips 2025🚨 We show that in sign-diverse networks, inherent non-gradient “curl” terms arise, and can, depending on network architecture, destabilize gradient-descent solutions or paradoxically accelerate learning beyond pure gradient flow. 🧵⬇️ www.arxiv.org/abs/2510.02765

Curl Descent: Non-Gradient Learning Dynamics with Sign-Diverse Plasticity

Gradient-based algorithms are a cornerstone of artificial neural network training, yet it remains unclear whether biological neural networks use similar gradient-based strategies during learning. Expe...

arxiv.org

Big week for astrocyte research: 3 new Science papers link astrocytes to behavior. We're excited to add to the momentum with our new PNAS paper: a theory, grounded in biology, proposing astrocytes as key players in memory storage and recall. w/ JJ Slotine and @krotov.bsky.social (1/6)

Bild

Created a starter pack of neuroscience in/from Paris. Let me know if you want to be added (the 'from' can include those not in Paris anymore) or just tap in if you want to know what we're talking about! Regardless, please re-tweet! go.bsky.app/3Zs9w5w

Post nicht verfügbar.