Brain Network Models & Dynamics

A geometrically grounded neural field model coupling cortex and hippocampus reproduces seizure onset transitions in three patients with open code, but validation is limited by circular parameter fitting and isotropic coupling simplifications.

Simulations of a two-variable integrate-and-fire field show refractoriness forces coherent excitations onto lower-dimensional boundaries, though the claim rests on numerical evidence in small idealized grids without biological validation.

Path-integral theory indicates mean firing and response variability are governed by the same coupling statistics, but no supplied record directly tests the motif-causality claim, which remains a simulation-based prediction.

Single Neuron Biophysics

Recasting conductance as lognormal ion-transit densities recovers Nernst potentials from squid and lamprey spike data, yet the fits were guided by the same potentials being recovered, so independent pharmacological tests remain unrun.

An 11-dimensional conductance-based model predicts complex spikers lose regular firing with tiny BK-conductance reductions while spikers tolerate ≥35% loss, but no experimental validation or code release accompanies the prediction.

Human volume-EM–constrained modeling shows fewer but larger synapses raise firing up to 2.15-fold under sparse input, though results come from a single passive model cell without clinical validation.

Learning, Cognition & Consciousness Models

A 10,000-neuron Izhikevich network reports integrated information φ rising to ~12.5–15 bits during synchronized sensory–DMN bursts, but a single unreplicated 900 ms run leaves the decisive tonic-ablation control unrun and the "endogenous Self" claim untested.

A 2×2 factorial ablation (pacemaker on/off × synchrony on/off) can decisively separate the two hypotheses at near-zero compute cost, with existing evidence from one unreplicated preprint tilting toward synchrony as the φ driver.

A simulated cerebellum with plastic granule–Purkinje synapses cuts minimum reaching preparation time from ~400 ms to ~150 ms in simulation, but with no statistical reporting, no seed variance, and no empirical validation.

Depression patients show higher mean EEG entropy yet lower inter-individual dispersion, qualitatively matched by simulations with electrical coupling off, but the model-data match is qualitative and causality remains unproven.

Neural Encoding & Neuroimaging Methods

In two Natural Scenes Dataset subjects, a caption-narrative contrast in visual cortex reversed sign after word-count matching, showing encoding-model conclusions can depend on how text predictors are built rather than on brain differences.

A model using only raw stimulus statistics reaches ρ≈0.97–0.99 with behavior across 534 humans, 110 rats, and 2 monkeys, though the authors themselves flag missing top-down mechanisms and possible dataset curation bias.

Pixel-brightness graphs classify four Alzheimer's stages at 89.45% cross-validated accuracy with open code, but the topological indices may capture image texture rather than neural connectivity, and validation is confined to one dataset.