Pithy Neural Kernel V2.2.1

LEARN. REBIRTH. LISTEN.

This is a 40 × 40 Self-Organizing Map. Each neuron holds an RGB vector. Normal training picks an input, finds the best-matching neuron, then nudges that neuron and nearby neurons toward the input. No labels, answer key, Python, or server required.

LEARN samples uniform random RGB colors with a decaying schedule. REBIRTH samples an evolving palette, snapshots each cycle boundary, changes the palette, optionally mutates or resets weights, and restarts the schedule. LISTEN uses local audio-band features as input vectors. It learns a color representation of those features, not musical meaning.

Painting directly edits a grid neighborhood. It is not counted as SOM training, and PAINT STEPS counts edit operations, not individual neurons. Total training counts actual BMU/update iterations; cycle steps advance only in REBIRTH; cooling advances only during SOM training and means iterations since the current learning schedule began. Rendering does not advance any of them.

The U-Matrix-style view shows average RGB differences between four neighbors. Bright ridges indicate discontinuities, not guaranteed clusters. Live QE is a smoothed input-to-winner distance. Because modes and palettes use different input distributions, their QE readings are not controlled cross-generation benchmarks.

THEN / NOW renders both weight buffers with the same current display settings and topology interpretation. A snapshot is actual restorable state, not a screenshot. Restoring pauses updates and stops audio; microphone permissions and audio playback are not restored. Saving or restoring never changes THEN.

Audio stays in this browser. No recording, uploads, trackers, remote fonts, libraries, or automatic network calls are included. External author links require a click. Pausing or opening this dialog suspends learning but does not stop an active audio source: the HUD says when audio is still captured. Use STOP AUDIO to release it. If your machine cannot keep up, training load-sheds; the HUD shows the measured rate.

Constant motion does not mean better learning. LEARN teaches stabilization. REBIRTH deliberately disrupts it. LISTEN changes the signal.

Built by Pithy Cyborg · AI News Made Simple