neuro

Cognitive science publishes a great many models and very few you can actually poke. The code, when there is code, is a notebook in a footnote that stopped running three library versions ago. This wing takes models worth playing with, reimplements them from the paper in Rust, compiles them to WebAssembly, and puts them behind a URL — with the replication against the published result shown on the page, so you can see for yourself whether the rebuild earns the citation.

Residents

• homeostasis →

  Falandays, Nguyen & Spivey (2021) · Brain Research 1768:147578

  A hundred randomly wired spiking neurons, each doing nothing but holding its own activation near a private target. No teaching signal, no stored sequence, nothing anywhere that represents a prediction. Train it on toy sentences, cut the input off, and its fading activity rolls into the pattern for the word that would most likely have come next — with the strength tracking how often it actually did. An argument that prediction in brains might be much cheaper than the Bayesian story needs it to be.

  reservoir computing predictive processing spiking network table 2 replicated

How a model gets in here