003 — PING Stimulus Overdrive Video

Abstract

PING — Pyramidal-Interneuron Network Gamma — emerges when excitatory (E) and inhibitory (I) populations drive each other: E fires, recruits I, I silences E, I decays, E fires again. The cycle repeats at \sim30–80 Hz, but only when E gets enough external drive; below threshold the network sits in background firing. This notebook is the basic drive-threshold scan — companion sweeps live in nb005 (integration step) and nb006 (E→I coupling strength). See The Notebooks for how this entry’s runner/artifact/figure triple is wired up.

Methods

A scan over the oscilloscope video subcommand: each frame is a fresh 600 ms simulation on MNIST digit 0, sample 0. The stim window (200–300 ms) multiplies the Poisson input rate by the overdrive factor shown on the frame; the sweep walks overdrive 1× → 10× over 300 frames. All knobs are hardcoded literals in src/pinglab/notebooks/nb003.py per the runner contract.

Parameter Value
Setup
Model ping
Input mnist d0 s0 @ 50 Hz max
Seed 42
Architecture
N_E / N_I 512 / 128
dt / T 0.1 / 600 ms
Stimulus & scan
Stim window 200–300 ms
Overdrive scan 1×–10×
Frames / FPS 300 / 30
Provenance
Tier large
Elapsed 37m 43s
Run ID r013
Git SHA ?

Results

Figure 1. Stim-overdrive sweep (scan_overdrive.mp4)

Each frame is a fresh 600 ms sim on MNIST digit 0, sample 0. The stim window (200–300 ms) multiplies the Poisson input rate by the overdrive factor shown on the frame; the sweep walks overdrive 1× → 10× over 300 frames.

In Figure 1 three regimes emerge: async baseline (1–1.2×), unstable PING onset (1.2–6×), stable PING (6×–onwards).

Population rates

WindowE (Hz)I (Hz)
Pre-stim6.80.0
In-stim15.535.5
Post-stim5.60.0

Discussion

TODO: discussion paragraph — write what the results above mean for the project.

Next steps

The threshold around overdrive ≈ 1.2× is sharp on this MNIST sample but the structure depends on input statistics; a sample-averaged version (mean rate envelope across many digits) would harden the regime boundaries. With the threshold map in hand, nb005 and nb006 hold drive fixed in stable-PING territory and sweep the other two knobs (dt, ei-strength) — together the three videos pin the working-point box for every training entry downstream.

Appendix

Reproduction

uv run src/pinglab/notebooks/nb003.py