https://github.com/pluskid/Mocha.jl
- Infrastructure
- Backward compatibility with Julia v0.3, and Julia v0.4 compatability
- Solver refactoring (@benmoran, @CarloLucibello)
- Adam Solver (@benmoran)
- Improved compatibility on Windows (@droidicus)
- Network
- Leaky ReLU (@stjanovitz)
- Shuffling support for memory-data layer
- DecayOnValidation now support square-loss like criterions
- Variational Auto-Encoder and related layers: RandomNormal,
BinaryCrossEntropyLossLayer, GaussianKLLoss, and Exponential neuron
type (@benmoran)
- SVM related layers: HingeLossLayer and BinaryAccuracyLayer
(@nstiurca)
- OrthogonalInitializer (@benmoran)
- Documentation
- Various typos and improvements from the community
- MNIST Variational Auto-Encoder example (@benmoran)
