Recently there are a considerable amount of work devoted to the study of the algorithmic stability and generalization for stochastic gradient descent (SGD). However, the existing stability analysis requires to impose restrictive assumptions on the boundedness of gradients, strong smoothness and convexity of loss functions. In this paper, we provide a fine-grained analysis of stability and generalization for SGD by substantially relaxing these assumptions. Firstly, we establish stability and generalization for SGD by removing the existing bounded gradient assumptions. The key idea is the introduction of a new stability measure called on-average model stability, for which we develop novel bounds controlled by the risks of SGD iterates. This yields generalization bounds depending on the behavior of the best model, and leads to the first-ever-known fast bounds in the low-noise setting using stability approach. Secondly, the smoothness assumption is relaxed by considering loss functions with Holder continuous (sub)gradients for which we show that optimal bounds are still achieved by balancing computation and stability. To our best knowledge, this gives the first-ever-known stability and generalization bounds for SGD with even non-differentiable loss functions. Finally, we study learning problems with (strongly) convex objectives but non-convex loss functions.

Clinical trials play important roles in drug development but often suffer from expensive, inaccurate and insufficient patient recruitment. The availability of massive electronic health records (EHR) data and trial eligibility criteria (EC) bring a new opportunity to data driven patient recruitment. One key task named patient-trial matching is to find qualified patients for clinical trials given structured EHR and unstructured EC text (both inclusion and exclusion criteria). How to match complex EC text with longitudinal patient EHRs? How to embed many-to-many relationships between patients and trials? How to explicitly handle the difference between inclusion and exclusion criteria? In this paper, we proposed CrOss-Modal PseudO-SiamEse network (COMPOSE) to address these challenges for patient-trial matching. One path of the network encodes EC using convolutional highway network. The other path processes EHR with multi-granularity memory network that encodes structured patient records into multiple levels based on medical ontology. Using the EC embedding as query, COMPOSE performs attentional record alignment and thus enables dynamic patient-trial matching. COMPOSE also introduces a composite loss term to maximize the similarity between patient records and inclusion criteria while minimize the similarity to the exclusion criteria. Experiment results show COMPOSE can reach 98.0% AUC on patient-criteria matching and 83.7% accuracy on patient-trial matching, which leads 24.3% improvement over the best baseline on real-world patient-trial matching tasks.

The broader ambition of this article is to popularize an approach for the fair distribution of the quantity of a system's output to its subsystems, while allowing for underlying complex subsystem level interactions. Particularly, we present a data-driven approach to vehicle price modeling and its component price estimation by leveraging a combination of concepts from machine learning and game theory. We show an alternative to common teardown methodologies and surveying approaches for component and vehicle price estimation at the manufacturer's suggested retail price (MSRP) level that has the advantage of bypassing the uncertainties involved in 1) the gathering of teardown data, 2) the need to perform expensive and biased surveying, and 3) the need to perform retail price equivalent (RPE) or indirect cost multiplier (ICM) adjustments to mark up direct manufacturing costs to MSRP. This novel exercise not only provides accurate pricing of the technologies at the customer level, but also shows the, a priori known, large gaps in pricing strategies between manufacturers, vehicle sizes, classes, market segments, and other factors. There is also clear synergism or interaction between the price of certain technologies and other specifications present in the same vehicle. Those (unsurprising) results are indication that old methods of manufacturer-level component costing, aggregation, and the application of a flat and rigid RPE or ICM adjustment factor should be carefully examined. The findings are based on an extensive database, developed by Argonne National Laboratory, that includes more than 64,000 vehicles covering MY1990 to MY2020 over hundreds of vehicle specs.

In this work we solve the Dynamic Vehicle Routing Problem (DVRP). DVRP is a modification of the Vehicle Routing Problem, in which the clients' requests (cities) number and location might not be known at the beginning of the working day Additionally, all requests must be served during one working day by a fleet of vehicles with limited capacity. In this work we propose a Monte Carlo method (MCTree), which directly approaches the dynamic nature of arriving requests in the DVRP. The method is also hybridized (MCTree+PSO) with our previous Two-Phase Multi-swarm Particle Swarm Optimization (2MPSO) algorithm. Our method is based on two assumptions. First, that we know a bounding rectangle of the area in which the requests might appear. Second, that the initial requests' sizes and frequency of appearance are representative for the yet unknown clients' requests. In order to solve the DVRP we divide the working day into several time slices in which we solve a static problem. In our Monte Carlo approach we randomly generate the unknown clients' requests with uniform spatial distribution over the bounding rectangle and requests' sizes uniformly sampled from the already known requests' sizes. The solution proposal is constructed with the application of a clustering algorithm and a route construction algorithm. The MCTree method is tested on a well established set of benchmarks proposed by Kilby et al. and is compared with the results achieved by applying our previous 2MPSO algorithm and other literature results. The proposed MCTree approach achieves a better time to quality trade-off then plain heuristic algorithms. Moreover, a hybrid MCTree+PSO approach achieves better time to quality trade-off then 2MPSO for small optimization time limits, making the hybrid a good candidate for handling real world scale goods delivery problems.

Organizations that own data face increasing legal liability for its discriminatory use against protected demographic groups, extending to contractual transactions involving third parties access and use of the data. This is problematic, since the original data owner cannot ex-ante anticipate all its future uses by downstream users. This paper explores the upstream ability to preemptively remove the correlations between features and sensitive attributes by mapping features to a fair representation space. Our main result shows that the fairness measured by the demographic parity of the representation distribution can be certified from a finite sample if and only if the chi-squared mutual information between features and representations is finite. Empirically, we find that smoothing the representation distribution provides generalization guarantees of fairness certificates, which improves upon existing fair representation learning approaches. Moreover, we do not observe that smoothing the representation distribution degrades the accuracy of downstream tasks compared to state-of-the-art methods in fair representation learning.

We study the distribution and uncertainty of nonconvex optimization for noisy tensor completion -- the problem of estimating a low-rank tensor given incomplete and corrupted observations of its entries. Focusing on a two-stage estimation algorithm proposed by Cai et al. (2019), we characterize the distribution of this nonconvex estimator down to fine scales. This distributional theory in turn allows one to construct valid and short confidence intervals for both the unseen tensor entries and the unknown tensor factors. The proposed inferential procedure enjoys several important features: (1) it is fully adaptive to noise heteroscedasticity, and (2) it is data-driven and automatically adapts to unknown noise distributions. Furthermore, our findings unveil the statistical optimality of nonconvex tensor completion: it attains un-improvable $\ell_{2}$ accuracy -- including both the rates and the pre-constants -- when estimating both the unknown tensor and the underlying tensor factors.

Binary classification is widely used in ML production systems. Monitoring classifiers in a constrained event space is well known. However, real world production systems often lack the ground truth these methods require. Privacy concerns may also require that the ground truth needed to evaluate the classifiers cannot be made available. In these autonomous settings, non-parametric estimators of performance are an attractive solution. They do not require theoretical models about how the classifiers made errors in any given sample. They just estimate how many errors there are in a sample of an industrial or robotic datastream. We construct one such non-parametric estimator of the sample errors for an ensemble of weak binary classifiers. Our approach uses algebraic geometry to reformulate the self-assessment problem for ensembles of binary classifiers as an exact polynomial system. The polynomial formulation can then be used to prove - as an algebraic geometry algorithm - that no general solution to the self-assessment problem is possible. However, specific solutions are possible in settings where the engineering context puts the classifiers close to independent errors. The practical utility of the method is illustrated on a real-world dataset from an online advertising campaign and a sample of common classification benchmarks. The accuracy estimators in the experiments where we have ground truth are better than one part in a hundred. The online advertising campaign data, where we do not have ground truth data, is verified by an internal consistency approach whose validity we conjecture as an algebraic geometry theorem. We call this approach - algebraic ground truth inference.

We study the role of $L_2$ regularization in deep learning, and uncover simple relations between the performance of the model, the $L_2$ coefficient, the learning rate, and the number of training steps. These empirical relations hold when the network is overparameterized. They can be used to predict the optimal regularization parameter of a given model. In addition, based on these observations we propose a dynamical schedule for the regularization parameter that improves performance and speeds up training. We test these proposals in modern image classification settings. Finally, we show that these empirical relations can be understood theoretically in the context of infinitely wide networks. We derive the gradient flow dynamics of such networks, and compare the role of $L_2$ regularization in this context with that of linear models.

This article applies the principle of Occam's Razor to non-parametric model building of statistical data, by finding a model with the minimal number of bits, leading to an exceptionally effective regularization method for probability density estimators. The idea comes from the fact that likelihood maximization also minimizes the number of bits required to encode a dataset. However, traditional methods overlook that the optimization of model parameters may also inadvertently play the part in encoding data points. The article shows how to extend the bit counting to the model parameters as well, providing the first true measure of complexity for parametric models. Minimizing the total bit requirement of a model of a dataset favors smaller derivatives, smoother probability density function estimates and most importantly, a phase space with fewer relevant parameters. In fact, it is able prune parameters and detect features with small probability at the same time. It is also shown, how it can be applied to any smooth, non-parametric probability density estimator.

Malware classification is a difficult problem, to which machine learning methods have been applied for decades. Yet progress has often been slow, in part due to a number of unique difficulties with the task that occur through all stages of the developing a machine learning system: data collection, labeling, feature creation and selection, model selection, and evaluation. In this survey we will review a number of the current methods and challenges related to malware classification, including data collection, feature extraction, and model construction, and evaluation. Our discussion will include thoughts on the constraints that must be considered for machine learning based solutions in this domain, and yet to be tackled problems for which machine learning could also provide a solution. This survey aims to be useful both to cybersecurity practitioners who wish to learn more about how machine learning can be applied to the malware problem, and to give data scientists the necessary background into the challenges in this uniquely complicated space.