NeuroSim / Experiments
Perturb the circuit. Observe the consequence.
Each protocol holds everything constant except one manipulation, runs the circuit on this device, and reports the measured difference against a control.
Loading laboratory
Catalogue
Six protocols.
- 01
Baseline circuit activity
What does the circuit do when it is driven by a stationary stimulus and nothing is perturbed?
- Intervention
- None. This condition defines the reference.
- Measure
- Population rate, active fraction, spike irregularity, readout latency.
- 02
Neuron ablation
Which populations does the circuit's output actually depend on, and by how much?
- Intervention
- A chosen fraction of a chosen population is silenced. Everything else is held.
- Measure
- Readout rate, active fraction, latency, spike statistics, against control.
- 03
Synaptic perturbation
How sensitive is circuit output to the strength of its synapses?
- Intervention
- A global multiplier is applied to every synaptic weight in the circuit.
- Measure
- Rate and irregularity as a function of the multiplier.
- 04
Input noise tolerance
How much membrane noise can the circuit absorb before its response degrades?
- Intervention
- Gaussian membrane noise is raised across conditions with the stimulus unchanged.
- Measure
- Background rate against readout rate as noise increases.
- 05
Sensory stimulus response
How does the circuit's output scale with stimulus intensity, and how quickly does it respond?
- Intervention
- Poisson input rate is stepped upward across conditions.
- Measure
- Readout rate and first-spike latency against input rate.
- 06
Embodied obstacle navigation
Can the spiking circuit drive a body through an environment, and what happens to behaviour when the circuit is damaged?
- Intervention
- A chosen part of the circuit is silenced. Arena, goal, body and seed are identical.
- Measure
- Collisions, path length, time to goal, and the circuit statistics behind them.
Method
Why the comparison holds.
An experiment on a simulation is only worth reading if the control and the condition differ in exactly one respect. That is enforced in the engine rather than by convention.
Circuit generation, membrane noise and input all draw from a single seeded generator, so the same seed reproduces the same numbers on any machine. Each condition is built from the control’s configuration with one field changed, and every condition runs through the same measurement code.
Ablation is applied as a mask rather than a graph edit, so the graph, the layout and every index stay identical between conditions.