Paper: The network around a cortical core is a regulator, not an optimizer
maier_ak · x · 2026-10-21
Andreas Maier summarizes his team's new paper in Frontiers in Computational Neuroscience. Building on Song et al.'s 2005 measurements of cortical layer 5 (a few strong, non-random connections embedded in a background of weak ones), the authors constructed an "Embedded Core Model": 125 stochastic binary neurons — 10 densely and strongly connected core excitatory neurons, 90 sparsely weakly connected peripheral excitatory neurons, and 25 inhibitory interneurons, with connection density and lognormal synaptic weight distribution taken from the measurements.
Using information flux (mutual information between successive network states) as the metric, they find the peripheral network does not maximize information flowing through the core; it keeps the core out of trouble — regulating rather than optimizing. A local rule keeping each neuron active half the time reaches an information flux of 213.8. The authors note the measure is not one of computation and does not show causal information transfer; that test comes next.
More from Research
- Study: decision-model controllers raise agent cost 19% with zero success gain — Naman Vats · 2026-10-21
- Tencent's RiseHarness iteratively evolves agent harnesses without touching model weights — Tencent-Hunyuan · 2026-10-21
- More memory, worse agents: Anti-Persistence discards synthesized guidance to cut errors — IUB · 2026-10-21
- CFG lifts single-step discrete action success from 27.1% to 49.2% in robot policy training — YouJiacheng · 2026-10-21
- Hamel Husain: BenchGate is pervasive — look at failing traces in any public benchmark — HamelHusain · 2026-10-21
- Menlo Research open-sources Cyclotron, an RL training framework for its Asimov humanoid robot — freelerobot · 2026-10-21