In this repository are various implementations of the ideas described in Miconi's "Neural Networks with Differentiable Structure". Currently, we have extended those ideas to multilayer perceptrons with one hidden layer. Implementations are available in NumPy, PyTorch, and TensorFlow. Based on our experience implementing these techniques, we believe they require a lot more work before they become a viable means of model selection for MLPs, and for neural networks in general. Feel free to use what you wish, however. Thanks for visiting.
rkingery/self_selecting_neural_nets
Folders and files
| Name | Name | Last commit date | ||
|---|---|---|---|---|