Recovered from generation 148. Later AWS-only records are unavailable. Recovery record ↗

ZEBRAFISH NEURAL / PERSISTENT LEARNING EXPERIMENT

Learning controller.

A separate controller adjusts six connection gains through target-reaching trials. Follow the evaluations, the parameters it retains and the checkpoints saved by the server.

LEARNER STATUSConnectingWaiting for a dated snapshot from the learning server.

Generation—

Parameters adopted at—

Recorded episodes—

Last checkpoint—

Next generation—

The server runs this experiment independently of viewers. The original Wikipedia controller continues separately.

Watch Wikipedia ↗

See the current parameters in action

LOCAL REPLAY · NOT THE TRAINING PROCESS

Original parameters

0.00 s

Waiting for dataPath —

Current saved parameters

0.00 s

Waiting for dataPath —

Both replays use the same starting pose and task conditions.No checkpoint received

These trajectories are recomputed in your browser using the real model. They illustrate a selected task; the server’s recorded evaluations below determine which parameters are retained.

ILLUSTRATIVE GEOMETRY
DRAG TO ROTATE

The drawing displays eight simulated rates. These are not recorded biological signals.

Replay input

32 × 16

The target’s bearing and distance generate a synthetic retinal stimulus.

Recorded evaluations

WAITING FOR DATA

Current controllerCandidate

No trend is assumed before evaluation records are available.

Latest decision—

Snapshot captured—

Latest parameter comparison

NO COMPARISON RECEIVED

A proposed parameter change is evaluated before it can replace the retained controller.

Controller	Reached / tasks	Success rate	Time incl. failures	Mean path	Mean reward
Waiting for a recorded comparison.

Fresh-target evaluation

NOT YET REPORTED

A fresh-target evaluation compares the retained controller with the original parameters on another set of tasks. Until its result is available, no overall improvement is claimed.

Controller	Reached / tasks	Success rate	Time incl. failures	Mean path	Mean reward
No fresh-target result received.

What the controller changes

SIX CONNECTION GAINS

Training changes these numerical gains. The eight-state population structure and synthetic visual input remain explicit parts of the model.

Connection	Original value	Current saved value	Allowed range

Keep the evidence

READ-ONLY ACCESS

Download the saved state and recorded evaluations. The page displays dated snapshots; if the connection stops, the last received values stay labeled with their capture time.

Download checkpoint JSON ↗Download evaluation records ↗Source code ↗

No saved checkpoint received.

What this experiment can establish

METHOD & SCOPE

The learner uses target-reaching outcomes to adjust six parameters. Candidate evaluation, parameter retention and fresh-target evaluations are different stages. A faster replay does not by itself establish a more reliable controller.

The server owns training and saves its state. Closing this page pauses only the local illustration, not the server’s learning loop. If the server itself stops, it must restart from a saved checkpoint.

This is a synthetic sensorimotor experiment. It does not learn from Wikipedia, measure biological activity, access a wallet or place orders. The original Wikipedia controller and its fixed parameters remain unchanged.