Interactive learning demos and no-fallback hardware checks for Synaptic.
Every page below compiles directly to WebGPU and must report
webgpu as its backend.
Interactive demo
Train a 2→3→1 network and watch all four truth-table predictions converge.
/packages/backend-webgpu/test/xor.html
Interactive demo
Train on MNIST, draw a digit, and inspect the network's ten output activations.
/packages/backend-webgpu/test/mnist.html
Interactive demo
Train direct and network MNIST rules, transfer both exactly, and compare them as Neural CAs.
/packages/backend-webgpu/test/mnist-automata.html
Interactive demo
Train growth, persistence, and repair through a 64–96-generation tape, then stream the artifact into Automata.
/packages/backend-webgpu/test/growing-neural-ca.html
Interactive demo
Train a four-cell LSTM, then animate random length-10 recall challenges.
/packages/backend-webgpu/test/dsr.html
Interactive demo
Continuously train a coordinate-to-RGB network to reconstruct a portrait.
/packages/backend-webgpu/test/learn-to-paint.html
Hardware check
Compile the WGSL module and report browser validation messages.
/packages/backend-webgpu/test/shader-smoke.html
Hardware check
Run a minimal recurrent train, forward, snapshot, and checkpoint lifecycle.
/packages/backend-webgpu/test/forward-smoke.html
Hardware check
Compare recurrent training, restoration, traces, and shared parameters with CPU.
/packages/backend-webgpu/test/training-parity.html
Hardware check
Verify Growing NCA loss descent and that Automata accepts the exported artifact.
/packages/backend-webgpu/test/growing-neural-ca-smoke.html
Automated regression
Learn XOR, MNIST, and DSR on hardware and report accuracy and elapsed time.
/packages/backend-webgpu/test/learning-workloads.html