In deterministic optimization, line searches are a standard tool ensuring stability and efficiency. Where only stochastic gradients are available, no direct equivalent has so far been formulated, because uncertain gradients do not allow for a strict sequence of decisions collapsing the search space. We construct a probabilistic line search by combining the structure of existing deterministic methods with notions from Bayesian optimization. Our method retains a Gaussian process surrogate of the univariate optimization objective, and uses a probabilistic belief over the Wolfe conditions to monitor the descent. The algorithm has very low computational cost, and no user-controlled parameters. Experiments show that it effectively removes the need to define a learning rate for stochastic gradient descent.
Maren Mahsereci (MPI for Intelligent Systems Tübingen)
Philipp Hennig (MPI for Intelligent Systems, Tübingen)
More from the Same Authors
2016 Workshop: Optimizing the Optimizers »
Maren Mahsereci · Alex Davies · Philipp Hennig
2015 Poster: Probabilistic Line Searches for Stochastic Optimization »
Maren Mahsereci · Philipp Hennig