Swift port of ConvnetJS - a library for training Deep Learning models (mainly Neural Networks).
Updated to Swift 3.
Available NN layer types:
- Fully Connected
- Convolutional
- Dropout
- ReLU
- Sigmoid
- Maxout
- Hyperbolic Tangent
- Local Response Normalization
- Max Pooling
- Output: Softmax, Regression, SVM
Optimization methods ('trainers'/'solvers'):
- Stochastic Gradient Descent
- Adam
- Adaptive Gradient
- WindowGrad
- AdaDelta
- Nesterov’s Accelerated Gradient
Reinforcement learning:
- Q-learning
Acceleration:
- Accelerate framework
- Metal
This is not a production-ready library. It is a personal research project and work in progress.