OpenTrainDNN — watch a neural network learn in your browser

Same network, two input types. Learn from images or from sound — the pipeline is identical.

DNN Architecture — live weights · strength encoded as thickness + glow

—

weak 0.1 medium 0.7 strong 2.0 green = positive red = negative V = validation sample

Prediction

—

no model trained yet

Camera: capture 20–30 images per class. Keys 1–9 grab the live frame. Microphone: say a word 20–30 times per class.

Input source

Camera: spatial pattern of light. Microphone: 16×16 mel spectrogram of sound.

Network type

Hidden layers (Dense)

—

Neuron type

Input resolution

Readout

Train accuracy—

Val accuracy—

Best val—

Loss—

Steps0

Network—

Parameters—

Input—

Classes—

|w| range—

Overfitting detected. Validation accuracy has dropped below its peak of —. Stop training and click Reset, or add more training samples.