webnn

Browser based Deep Neural Network training and inference powered by weblas.

Inference is currently implemented for the following layer types:

• Dense (fully connected/inner product)
• Convolution
• Pooling (max)
• ReLU

Support for more layers is planned, including:

• Pooling (avg)
• Normalization (local and batch)

Support for training (via backpropagation) is also planned.

Donations

Want to see more happen here? Contribute on