🔬

Neuroscience Engineering

6
Open Unknowns
0
Cross-Domain Bridges
6
Active Hypotheses

Open Unknowns (6)

Unknown Can computational phenotypes (learning rates, prior precision parameters) derived from brief computerised behavioural tasks predict individual antidepressant or antipsychotic treatment response before drug administration, enabling precision psychiatry prescribing? u-computational-psychiatry-treatment-response-prediction
Unknown What learning rule does the cerebellum use to update its forward model predictions, and can the supervised climbing fiber error signal account for the full range of motor adaptation timescales from milliseconds to months? u-forward-model-cerebellum-learning-rule
Unknown Under what conditions do cortical circuits implement variance-weighted prediction–correction equivalent to a Kalman gain, and when do they require nonlinear filtering? u-kalman-filter-neural-implementation-limits
Unknown Under correlated synaptic input and active dendrites, when do membrane τ estimates from somatic recordings remain identifiable with simple RC reductions? u-lif-parameter-identifiability-noisy-synapses
Unknown Whether the brain primarily uses rate coding, temporal coding, or a multiplexed combination for information representation, and what the relative information capacity of each scheme is in identified neural circuits u-neural-spike-coding-rate-vs-temporal
Unknown How does the central nervous system resolve the redundancy problem in muscle coordination — the fact that any joint torque can be produced by infinitely many combinations of muscle activation levels — and does the nervous system optimise a single cost function or use task-dependent switching? u-neuromuscular-control-redundancy-resolution

Active Hypotheses

Hypothesis The cerebellum implements a forward model that predicts sensory consequences of motor commands via a biologically plausible approximation to the Kalman filter: Purkinje cells encode prediction of sensory state given efference copy, granule cells provide the basis for state representation, and climbing fiber error drives gradient descent on prediction error, implementing a neural linear quadratic regulator for motor control. high
Hypothesis Antipsychotic drugs (D2 antagonists) reduce psychotic symptoms by lowering the dopaminergic precision-weighting signal — reducible to a single parameter ω_DA in the hierarchical Bayesian model — and their therapeutic efficacy across patients is quantitatively predicted by the degree to which they normalise precision-weighted prediction error updating in computational task assays. high
Hypothesis Joint intracellular noise injections spanning three decades of bandwidth will yield τ estimates whose RC-fit residuals correlate with multicompartment model mismatch scores — falsifying universal single τ under active conductances above threshold noise variance. medium
Hypothesis Neuromorphic chips implementing spiking neural networks will achieve 100–1000× lower energy per inference than GPU-based neural networks for always-on edge AI tasks (keyword spotting, gesture recognition, anomaly detection) while maintaining competitive accuracy, making them the dominant architecture for IoT sensing applications by 2030 high
Hypothesis Henneman's size principle (slow-twitch S motor units recruited before FR before FF) is the optimal recruitment strategy that minimises total metabolic energy expenditure per unit force-time integral across all submaximal contractions — and violations of orderly recruitment observed in some tasks (reverse recruitment) are predicted by the optimisation when metabolic cost includes task-specific penalty terms. medium
Hypothesis Manipulating cue reliability in a multisensory integration task will shift population-level sensory weights in proportion to relative variances consistent with a Kalman gain within ~20% error. medium

Know something about Neuroscience Engineering? Contribute an unknown or hypothesis →

Generated 2026-05-10 · USDR Dashboard