Authors: Christof Naumzik & Stefan Feuerriegel
Caffe (http://caffe.berkeleyvision.org) provides a powerful framework for deep learning. It is developed and maintained by the Berkeley Vision and Learning Center (BVLC) and has received a great deal of traction lately.
Caffe enables users to define and train custom-made neural networks without hard-coding. Furthermore, it allows users to execute all computations on CPUs as well as GPUs. Recent research has created a vast zoo of models. This rich prevalence of existing models makes it easy for users to leverage pre-trained neural networks that are known to perform well in various machine learning tasks.
While caffe already offers Matlab and Python interfaces, R is not currently supported. Our package caffeR aims at providing wrapper functions that allow its users to run caffe from R. These include data preprocessing and setup of networks, as well as monitoring and evaluation of training processes. For this purpose, caffeR prepares the correct configuration files and then passes routine calls directly to caffe.
Download of caffeR via GitHub: https://github.com/cnaumzik/caffeR