More recent works introduced advanced multi-criteria optimization models [12, 21, 27] that can express more accurately several established cartographic principles, but still with the aim of a full automation of the map labeling process. While progress is made by incorporating more comprehensive cartographic rules for label placement, none of the above approaches includes decisions made by human experts - other than setting preferences, parameters, and priorities in the different scoring functions that control a single optimization run of the respective algorithm. A notable exception is the UserHints framework [7], where human interaction was integrated into solving the label number maximization problem in a fixed-position point labeling setting. In that system, two heuristic methods were implemented as labeling algorithms, and hence the evaluation could not assess the deviation from optimal solutions with respect to the objective function. Moreover, the authors did not consider the stability of the labeling under user interaction. Beyond the label placement problem, interactive optimization [22] and human-guided search [16] are of course techniques that are of general interest and more broadly applicable. 2 Popular GIS software like Mapbox 1, ArcGIS Pro 2, or QGIS 3 also provide labeling algorithms. Mapbox allows customized label modifications with data conditions, but no manual selection or drag-and-drop placement. The ArcGIS Pro documentation 4 states "Label positions are generated automatically.

Regression methods have provided a foundation for modeling structure-property relationships in materials science1,2,3,4,5,6. These methods take as input a data set of known material compositions along with some property (e.g. Each material, in turn, is described in terms of one or more features that represent aspects of structure, chemistry, bonding and/or microstructure in an abstract, high-dimensional space. The success of regression is based on its ability to capture the relative variation in a property as a function of the features, eventually culminating in the prediction of new materials with desired properties. Materials design is an optimization problem with the goal of maximizing (or minimizing) some desired property of a material, denoted by y, by varying certain features, denoted by x.

A highly influential ingredient of many techniques designed to exploit sparsity in numerical optimization is the so-called chordal extension of a graph representation of the optimization problem. The definitive relation between chordal extension and the performance of the optimization algorithm that uses the extension is not a mathematically understood task. For this reason, we follow the current research trend of looking at Combinatorial Optimization tasks by using a Machine Learning lens, and we devise a framework for learning elimination rules yielding high-quality chordal extensions. As a first building block of the learning framework, we propose an on-policy imitation learning scheme that mimics the elimination ordering provided by the (classical) minimum degree rule. The results show that our on-policy imitation learning approach is effective in learning the minimum degree policy and, consequently, produces graphs with desirable fill-in characteristics.

Adversarial representation learning is a promising paradigm for obtaining data representations that are invariant to certain sensitive attributes while retaining the information necessary for predicting target attributes. Existing approaches solve this problem through iterative adversarial minimax optimization and lack theoretical guarantees. In this paper, we first study the "linear" form of this problem i.e., the setting where all the players are linear functions. We show that the resulting optimization problem is both non-convex and non-differentiable. We obtain an exact closed-form expression for its global optima through spectral learning and provide performance guarantees in terms of analytical bounds on the achievable utility and invariance. We then extend this solution and analysis to non-linear functions through kernel representation. Numerical experiments on UCI, Extended Yale B and CIFAR-100 datasets indicate that, (a) practically, our solution is ideal for "imparting" provable invariance to any biased pre-trained data representation, and (b) empirically, the trade-off between utility and invariance provided by our solution is comparable to iterative minimax optimization of existing deep neural network based approaches. Code is available at https://github.com/human-analysis/Kernel-ARL

We introduce a new meta-learning procedure, called non-Euclidean upgrading (NEU), which learns algorithm-specific geometries by deforming the ambient space until the algorithm can achieve optimal performance. We prove that these deformations have several novel and semi-classical universal approximation properties. These deformations can be used to approximate any continuous, Borel, or modular-Lebesgue integrable functions to arbitrary precision. Further, these deformations can transport any data-set into any other data-set in a finite number of iterations while leaving most of the space fixed. The NEU meta-algorithm embeds these deformations into a wide range of learning algorithms. We prove that the NEU version of the original algorithm must perform better than the original learning algorithm. Moreover, by quantifying model-free learning algorithms as specific unconstrained optimization problems, we find that the NEU version of a learning algorithm must perform better than the model-free extension of the original algorithm. The properties and performance of the NEU meta-algorithm are examined in various simulation studies and applications to financial data.

Infinite horizon off-policy policy evaluation is a highly challenging task due to the excessively large variance of typical importance sampling (IS) estimators. Recently, Liu et al. (2018a) proposed an approach that significantly reduces the variance of infinite-horizon off-policy evaluation by estimating the stationary density ratio, but at the cost of introducing potentially high biases due to the error in density ratio estimation. In this paper, we develop a bias-reduced augmentation of their method, which can take advantage of a learned value function to obtain higher accuracy. Our method is doubly robust in that the bias vanishes when either the density ratio or the value function estimation is perfect. In general, when either of them is accurate, the bias can also be reduced. Both theoretical and empirical results show that our method yields significant advantages over previous methods.

Variable metric proximal gradient (VM-PG) is a widely used class of convex optimization method. Lately, there has been a lot of research on the theoretical guarantees of VM-PG with different metric selections. However, most such metric selections are dependent on (an expensive) Hessian, or limited to scalar stepsizes like the Barzilai-Borwein (BB) stepsize with lots of safeguarding. Instead, in this paper we propose an adaptive metric selection strategy called the diagonal Barzilai-Borwein (BB) stepsize. The proposed diagonal selection better captures the local geometry of the problem while keeping per-step computation cost similar to the scalar BB stepsize i.e. $O(n)$. Under this metric selection for VM-PG, the theoretical convergence is analyzed. Our empirical studies illustrate the improved convergence results under the proposed diagonal BB stepsize, specifically for ill-conditioned machine learning problems for both synthetic and real-world datasets.

Bayesian optimization (BO) is a model-based approach to sequentially optimize expensive black-box functions, such as the validation error of a deep neural network with respect to its hyperparameters. In many real-world scenarios, the optimization is further subject to a priori unknown constraints. For example, training a deep network configuration may fail with an out-of-memory error when the model is too large. In this work, we focus on a general formulation of Gaussian process-based BO with continuous or binary constraints. We propose constrained Max-value Entropy Search (cMES), a novel information theoretic-based acquisition function implementing this formulation. We also revisit the validity of the factorized approximation adopted for rapid computation of the MES acquisition function, showing empirically that this leads to inaccurate results. On an extensive set of real-world constrained hyperparameter optimization problems we show that cMES compares favourably to prior work, while being simpler to implement and faster than other constrained extensions of Entropy Search.

We propose a novel strategy for extracting features in supervised learning that can be used to construct a classifier which is more robust to small perturbations in the input space. Our method builds upon the idea of the information bottleneck by introducing an additional penalty term that encourages the Fisher information of the extracted features to be small, when parametrized by the inputs. By tuning the regularization parameter, we can explicitly trade off the opposing desiderata of robustness and accuracy when constructing a classifier. We derive the optimal solution to the robust information bottleneck when the inputs and outputs are jointly Gaussian, proving that the optimally robust features are also jointly Gaussian in that setting. Furthermore, we propose a method for optimizing a variational bound on the robust information bottleneck objective in general settings using stochastic gradient descent, which may be implemented efficiently in neural networks. Our experimental results for synthetic and real data sets show that the proposed feature extraction method indeed produces classifiers with increased robustness to perturbations.

The main goal of this work is equipping convex and nonconvex problems with Barzilai-Borwein (BB) step size. With the adaptivity of BB step sizes granted, they can fail when the objective function is not strongly convex. To overcome this challenge, the key idea here is to bridge (non)convex problems and strongly convex ones via regularization. The proposed regularization schemes are \textit{simple} yet effective. Wedding the BB step size with a variance reduction method, known as SARAH, offers a free lunch compared with vanilla SARAH in convex problems. The convergence of BB step sizes in nonconvex problems is also established and its complexity is no worse than other adaptive step sizes such as AdaGrad. As a byproduct, our regularized SARAH methods for convex functions ensure that the complexity to find $\mathbb{E}[\| \nabla f(\mathbf{x}) \|^2]\leq \epsilon$ is ${\cal O}\big( (n+\frac{1}{\sqrt{\epsilon}})\ln{\frac{1}{\epsilon}}\big)$, improving $\epsilon$ dependence over existing results. Numerical tests further validate the merits of proposed approaches.