Splash is a C++ framework designed to allow anyone to create capable AI models with little set-up. Utilities which allow users to save networks easily and load datasets automatically help create a streamlined development workflow.
Whilst currently in an early development phase, at this point in time Splash is capable of producing:
- Multilayer networks with a variety of available activations
- Partial convolutional networks with max pooling
- SGD optimization with an MSE loss
- Network Saving/Loading functionality
- A dataset loader tool
In time I would like to really elaborate on this, adding support for diferent forms of pooling, more optimization algorithms, reinforcement learning and more.
- Add support for softmax actiavtion
- Add support for fully convvolutional networks
- Add support for graphical analysis of network performance
- Add new optimization algorithms and LR decay
- Support reinforcement learning, transformers, NERFs
Coming soon! (Soon being some arbitrary point in the future which may not actually be particularly soon at all)