BackpropNet provides basic neural network capabilities for .NET to train feed-forward models. It is meant to fuel lightweight decision making tasks through reinforcement learning.
- Dense
- ReLU
- Softmax
- Mean Squared Error
- Crossentropy (Full / Sparse)
- Gradient Descent (Vanilla)
- Adam
- Regression: sin(x)
- Classification: MNIST
BackpropNet is available under the terms of the GPLv3 license.