Rafko

A deep learning Framework extended with per Neuron memory capabilities with focus on architecture search through training. The Framework is of a server-client architecture, in which a deep learning server provides calculation slots for different clients. The calculations are neural network related, such as solving a network, calculating gradients and updating a network, etc..

Structure Overview:

• src/main/cxx: Contains the Source code for the Neural Network and support libraries
  • rafko_gym: Building blocks of network Training
  • rafko_mainframe: A basic incomplete implementation if a deep learning service based on the library
  • rafko_net: Building blocks of a Neural Network
  • rafko_utilities: Various Utility implementations not strictly part of the library in topic. It is preferred that his module does not have any dependency, not even from the Rafko repository.
  • test: Test suites for checking consistency and corect behavior
• src/main/java: Let's not look in there yet..
• /res: miscellianeous resources

Build Dependencies:

The Library is built upon Googles protocol buffer library and the services are using GRPC; Tested with the Catch Framework.

• Building the Network Library only requires a working installation of protobuf and g++
  • clang not supported yet
• Building the mainframe requires a working installation of GRPC

Unix dependencies:

• CMake
• pkg-conf
• build-essential

Windows dependencies:

Probably much more... It might work, but it's not supported currently.