F2FNeural dynamics

Function-to-function emulation

Given the activity of a cultured microisland up to a cutoff in time, forecast the spikes of each neuron in the interval that follows.

The task

When will each neuron spike next?

Input
A time-ordered history of spike events, with each event assigned to a neuron.
Output
A set of predicted spike times for each neuron in the held-out interval.
Spike rasterTask schematic
55%
Each row represents a neuron. Move the cutoff to change how much of the schematic is visible to the model.

Event matching

Pair a predicted spike with at most one reference spike from the same neuron, within the stated timing tolerance. Extra predictions and missed spikes both count.

Precision and recall

Precision is the fraction of predicted spikes that match. Recall is the fraction of reference spikes recovered. Report both alongside their harmonic mean, F1.

Evaluation split

Report within-recording forecasts separately from tests on entirely held-out microislands. State the prediction horizon and timing tolerance with every result.

Resources

Functional recordings

Neuronal event histories and evaluation splits.

Evaluation code

Event matching and scoring for held-out spike forecasts.