Caffe is a framework for deep learning. In a deep learning net it is quite hard to find good parameters (learning rate, dropout, size of convolutional filters, etc). Spearmint is a tool to perform Bayesian optimization over multiple variables given an objective function. The method was successfully applied to deep learning to help with the parameter search.
Caffe with Spearmint (CWSM) is my attempt to make a user-friendly combination of the two: https://github.com/kuz/caffe-with-spearmint. Primary audience: Caffe users. With the default Caffe MNIST example it managed to squeeze accuracy from 0.99 to 0.9931.
Thanks to Anna for the logo!