deeplearning-go

The framework has been designed with a modular structure in mind, allowing for the creation of dynamic neural network architectures and experimentation. It's a work in progress while venturing deeper into the field. Please submit a pull request if you'd like to contribute. 🙂

There is an accompanying guide written on gitbook, primarily for software engineers who would like an overview of this fascinating field of computation. It covers the core components of neural networks and how they function.

Core Components

• Neural Network
• Tensor
• Layer Interface with implementations for
  • Fully Connected (Dense) Layer
  • Long-Short-Term-Memory (LSTM) Layer
  • Gated Recurrent Unit (GRU) Layer
  • Convolutional (Conv2D) Layer
  • Embedding Layer
  • Dropout Regularization Layer (Interlayer Dropout)
  • Average and Maximum Pooling Layers
  • Flatten and Reshape Layers
• Activation Functions
  • ReLU and Leaky ReLU
  • Sigmoid
  • Tanh
  • Softmax
• Loss Functions
  • Cross Entropy (Binary and Categorical)
  • Mean Squared Error (MSE)
  • Cosine Proximity
• Regularisation
  • Lasso (L1)
  • Ridge (L2)
  • Elastic Net
• Optimisation Algorithms
  • Stochastic Gradient Descent (SGD)
  • SGD with momentum
  • Root Mean Square Propagation (RMSProp)
  • Adaptive momentum (ADAM)

Examples

The examples are currently showing how to set up a neural network with layers etc. The ThreeLayer example is running from main() at the moment. The MNIST dataset has been stored in the /data directory for testing purposes.

Future Work

• GPU processing using Metal on MacOS, using C via Go.
• Potentially implement a computation graph. (Not sure if I would like to remove the back propagation mechanism)
• Better Examples
• New Architectures
• New custom Layer types
• The model saving/loading system needs to be looked at further, to enable pipelines.
• Implement inference mechanisms with API deployment, model loading etc.

Contributing

Please submit a pull request if you'd like to contribute. 🙂