The use of machine learning algorithms frequently involves careful tuning of learning parameters and model hyperparameters. Unfortunately, this tuning is often a "black art" requiring expert experience, rules of thumb, or sometimes brute-force search. There is therefore great appeal for automatic approaches that can optimize the performance of any given learning algorithm to the problem at hand. In this work, we consider this problem through the framework of Bayesian optimization, in which a learning algorithm's generalization performance is modeled as a sample from a Gaussian process (GP). We show that certain choices for the nature of the GP, such as the type of kernel and the treatment of its hyperparameters, can play a crucial role in obtaining a good optimizer that can achieve expertlevel performance. We describe new algorithms that take into account the variable cost (duration) of learning algorithm experiments and that can leverage the presence of multiple cores for parallel experimentation. We show that these proposed algorithms improve on previous automatic procedures and can reach or surpass human expert-level optimization for many algorithms including latent Dirichlet allocation, structured SVMs and convolutional neural networks.

/pdf/practical-bayesian-optimization-of-machine-learning-3z1au6zrix.pdf

Practical Bayesian Optimization of Machine Learning Algorithms

Big Data applications are typically associated with systems involving large numbers of users, massive complex software systems, and large-scale heterogeneous computing and storage architectures. The construction of such systems involves many distributed design choices. The end products (e.g., recommendation systems, medical analysis tools, real-time game engines, speech recognizers) thus involve many tunable configuration parameters. These parameters are often specified and hard-coded into the software by various developers or teams. If optimized jointly, these parameters can result in significant improvements. Bayesian optimization is a powerful tool for the joint optimization of design choices that is gaining great popularity in recent years. It promises greater automation so as to increase both product quality and human productivity. This review paper introduces Bayesian optimization, highlights some of its methodological aspects, and showcases a wide range of applications.

/pdf/taking-the-human-out-of-the-loop-a-review-of-bayesian-4olcznoudw.pdf

Taking the Human Out of the Loop: A Review of Bayesian Optimization

Preface to the revised edition Remarks on notation 1. Basic examples 2. Characterization and existence 3. Stable processes and their extensions 4. The Levy-Ito decomposition of sample functions 5. Distributional properties of Levy processes 6. Subordination and density transformation 7. Recurrence and transience 8. Potential theory for Levy processes 9. Wiener-Hopf factorizations 10. More distributional properties Supplement Solutions to exercises References and author index Subject index.

Lévy processes and infinitely divisible distributions

Bayesian optimization is an approach to optimizing objective functions that take a long time (minutes or hours) to evaluate. It is best-suited for optimization over continuous domains of less than 20 dimensions, and tolerates stochastic noise in function evaluations. It builds a surrogate for the objective and quantifies the uncertainty in that surrogate using a Bayesian machine learning technique, Gaussian process regression, and then uses an acquisition function defined from this surrogate to decide where to sample. In this tutorial, we describe how Bayesian optimization works, including Gaussian process regression and three common acquisition functions: expected improvement, entropy search, and knowledge gradient. We then discuss more advanced techniques, including running multiple function evaluations in parallel, multi-fidelity and multi-information source optimization, expensive-to-evaluate constraints, random environmental conditions, multi-task Bayesian optimization, and the inclusion of derivative information. We conclude with a discussion of Bayesian optimization software and future research directions in the field. Within our tutorial material we provide a generalization of expected improvement to noisy evaluations, beyond the noise-free setting where it is more commonly applied. This generalization is justified by a formal decision-theoretic argument, standing in contrast to previous ad hoc modifications.

A Tutorial on Bayesian Optimization

In this work, we propose a method that combines a single hand-held camera and a set of Inertial Measurement Units (IMUs) attached at the body limbs to estimate accurate 3D poses in the wild. This poses many new challenges: the moving camera, heading drift, cluttered background, occlusions and many people visible in the video. We associate 2D pose detections in each image to the corresponding IMU-equipped persons by solving a novel graph based optimization problem that forces 3D to 2D coherency within a frame and across long range frames. Given associations, we jointly optimize the pose of a statistical body model, the camera pose and heading drift using a continuous optimization framework. We validated our method on the TotalCapture dataset, which provides video and IMU synchronized with ground truth. We obtain an accuracy of 26 mm, which makes it accurate enough to serve as a benchmark for image-based 3D pose estimation in the wild. Using our method, we recorded 3D Poses in the Wild (3DPW), a new dataset consisting of more than 51, 000 frames with accurate 3D pose in challenging sequences, including walking in the city, going up-stairs, having coffee or taking the bus. We make the reconstructed 3D poses, video, IMU and 3D models available for research purposes at http://virtualhumans.mpi-inf.mpg.de/3DPW.

/pdf/recovering-accurate-3d-human-pose-in-the-wild-using-imus-and-1k3jfjjean.pdf

Recovering Accurate {3D} Human Pose in the Wild Using {IMUs} and a Moving Camera

In the efficient global optimization problem, we minimize an unknown function f, using as few observations f(x) as possible. It can be considered a continuum-armed-bandit problem, with noiseless data, and simple regret. Expected-improvement algorithms are perhaps the most popular methods for solving the problem; in this paper, we provide theoretical results on their asymptotic behaviour. Implementing these algorithms requires a choice of Gaussian-process prior, which determines an associated space of functions, its reproducing-kernel Hilbert space (RKHS). When the prior is fixed, expected improvement is known to converge on the minimum of any function in its RKHS. We provide convergence rates for this procedure, optimal for functions of low smoothness, and describe a modified algorithm attaining optimal rates for smoother functions. In practice, however, priors are typically estimated sequentially from the data. For standard estimators, we show this procedure may never find the minimum of f. We then propose alternative estimators, chosen to minimize the constants in the rate of convergence, and show these estimators retain the convergence rates of a fixed prior.

/pdf/convergence-rates-of-efficient-global-optimization-430mztct1c.pdf

Convergence Rates of Efficient Global Optimization Algorithms

Efficient global optimization is the problem of minimizing an unknown function f, using as few evaluations f(x) as possible. It can be considered as a continuum-armed bandit problem, with noiseless data and simple regret. Expected improvement is perhaps the most popular method for solving this problem; the algorithm performs well in experiments, but little is known about its theoretical properties. Implementing expected improvement requires a choice of Gaussian process prior, which determines an associated space of functions, its reproducing-kernel Hilbert space (RKHS). When the prior is fixed, expected improvement is known to converge on the minimum of any function in the RKHS. We begin by providing convergence rates for this procedure. The rates are optimal for functions of low smoothness, and we modify the algorithm to attain optimal rates for smoother functions. For practitioners, however, these results are somewhat misleading. Priors are typically not held fixed, but depend on parameters estimated from the data. For standard estimators, we show this procedure may never discover the minimum of f. We then propose alternative estimators, chosen to minimize the constants in the rate of convergence, and show these estimators retain the convergence rates of a fixed prior.

/pdf/convergence-rates-of-efficient-global-optimization-5yg57httqx.pdf

Convergence rates of efficient global optimization algorithms

Confidence bands are confidence sets for an unknown function f, containing all functions within some sup-norm distance of an estima- tor. In the density estimation, regression, and white noise models, we consider the problem of constructing adaptive confidence bands, whose width contracts at an optimal rate over a range of Holder classes. While adaptive estimators exist, in general adaptive confidence bands do not, and to proceed we must place further conditions on f. We discuss previous approaches to this issue, and show it is necessary to restrict f to fundamentally smaller classes of functions. We then consider the self-similar functions, whose Holder norm is similar at large and small scales. We show that such functions may be considered typical functions of a given Holder class, and that the assumption of self-similarity is both necessary and sufficient for the construction of adaptive bands. Finally, we show that this assumption allows us to resolve the problem of undersmoothing, creating bands which are honest simultaneously for functions of any Holder norm.

/pdf/honest-adaptive-confidence-bands-and-self-similar-functions-mu5vdb7711.pdf

Honest adaptive confidence bands and self-similar functions

The problem of constructing confidence sets that are adaptive in L^2-loss over a continuous scale of Sobolev classes of probability densities is considered. Adaptation holds, where possible, with respect to both the radius of the Sobolev ball and its smoothness degree, and over maximal parameter spaces for which adaptation is possible. Two key regimes of parameter constellations are identified: one where full adaptation is possible, and one where adaptation requires critical regions be removed. Techniques used to derive these results include a general nonparametric minimax test for infinite-dimensional null- and alternative hypotheses, and new lower bounds for L^2-adaptive confidence sets.

Adaptive confidence sets in L^2

The problem of constructing confidence sets that are adaptive in L 2 -loss over a continuous scale of Sobolev classes of probability densities is considered. Adaptation holds, where possible, with respect to both the radius of the Sobolev ball and its smoothness degree, and over maximal parameter spaces for which adaptation is possible. Two key regimes of parameter constellations are identified: one where full adaptation is possible, and one where adaptation requires critical regions be removed. Techniques used to derive these results include a general nonparametric minimax test for infinite-dimensional null- and alternative hypotheses, and new lower bounds for L 2 -adaptive confidence sets.

/pdf/adaptive-confidence-sets-in-l-2-zscn13h7mn.pdf

Adam D. Bull

Papers

Convergence Rates of Efficient Global Optimization Algorithms

Convergence rates of efficient global optimization algorithms

Honest adaptive confidence bands and self-similar functions

Adaptive confidence sets in L^2

Adaptive confidence sets in L^2