Saurabh Bedi

📢 New preprint! 📢

Very excited to be a part of the project led by
@saurabhbedi.bsky.social on how the brain learns from multimodal inputs (e.g. audiovisual):

Separable neurocomputational mechanisms underlying multisensory learning
biorxiv.org/content/10.1...

When we induce new boundaries, the distortions are largely range-independent, pointing to the decoding stage as the main source.
These principles are likely to generalize to other contexts with bounded quantities 🔄

This work is with @GillesdeH and Christian Ruff 🙂

Efficient encoding & Bayesian decoding origins of distortions and variance (via cognitive boundary repulsions) come with dissociable signatures:
➡️ Encoding → range-dependent distortions + variance patterns
➡️ Decoding → range-independent distortions + variance patterns

The mechanism is cognitive boundary repulsion: during inference, the posterior is pushed away from boundaries, generating distortions and variance patterns.
This arises independently under both efficient encoding and Bayesian decoding of bounded quantities like probabilities.

Probability weighting arises from boundary repulsions of cognitive noise In both risky choice and perception, people overweight small and underweight large probabilities. While prospect theory models this with a probability weighting function, and Bayesian noisy coding mod...

📢 Preprint out! biorxiv.org/content/10.1... What gives rise to probability weighting, a cornerstone of Prospect Theory?
We show it comes from the natural boundedness of probabilities + cognitive noise. Adding boundaries adds multiple distortions, across risky choice & perception.