INTERACTIVE TRAINING RUN

Teach a model to see the pattern.

Adjust the learning recipe, run the simulation, and watch the decision boundary learn in real time.

CURRENT STATUS Ready to run Weights initialized

Dataset

Model

Learning rate0.08

Batch size16

Step Run training

EPOCH0Not started

LOSS—Awaiting first pass

ACCURACY—Awaiting first pass

CONFIDENCE—Model signal strength

LIVE MODEL VIEW

Decision boundary

Class AClass B

HOVER / TAP TO INSPECTMove over the mapx 0.00 · y 0.00

Press Run to begin learning

Boundary updates after every batch32 samples

TRAINING SIGNAL

Loss over time

—

startepochnow

↘

What to watch

Run a few epochs to see the model separate the classes.

MODEL ANATOMY

Network activity

2 layers

low activationhigh activationpositive weightnegative weight

RUN CONSOLE

Training log

Clear

AGENT LEARNING LAB

Teach an agent to reach the beacon.

Watch a small policy learn from reward, explore obstacles, and complete a route one decision at a time.

MISSION STATUS

Task

Exploration24%

Run speed

Step agent Reset mission

EPISODE—current learning run

REWARD—step return

STEPS—path length

SUCCESS RATE0%

STATE SPACE + POLICY

Learned navigation field

Q-POLICY

STATE INSPECTOR

learned policy agent trail obstacle goal beacon

Arrows show the highest-value action for each state. Hover or tap a cell to inspect its Q-values.

REINFORCEMENT SIGNAL

Reward over episodes

—

startepisodenow

↗

What to watch

Q-VALUE INSPECTOR

Selected state

↑Up0.00

→Right0.00

↓Down0.00

←Left0.00

A Q-value estimates the future return of taking an action from this state.