Arbiter is a library for hyperparameter optimization of neural networks. Hyperparameter optimization refers to the process of automating the selection of network hyperparameters (learning rate, number of layers, etc) in order to obtain good performance. It is part of the Deeplearning4j suite of software.
Automated hyperparameter search is necessary when the optimization search is too large for a human to manually tune. Arbiter helps by doing this through multiple methods, including random search, grid search, and Bayesian methods.