Uncertainty analysis in the form of probabilistic forecasting can provide significant improvements in decision making processes in the smart power gird for better integrating renewable energies such as wind. Whereas point forecasting provides a single expected value, probabilistic forecasts provide more information in the form of quantiles, prediction intervals, or full predictive densities. This paper analyzes the effectiveness of an approach for nonparametric probabilistic forecasting of wind power that combines support vector machines and nonlinear quantile regression with non-crossing constraints. A numerical case study is conducted using publicly available wind data from the Global Energy Forecasting Competition 2014. Multiple quantiles are estimated to form 20%, 40%, 60% and 80% prediction intervals which are evaluated using the pinball loss function and reliability measures. Three benchmark models are used for comparison where results demonstrate the proposed approach leads to significantly better performance while preventing the problem of overlapping quantile estimates.

As datasets for machine learning grow larger, parallelization strategies become more and more important. Recent approaches to distributed modelfitting rely heavily either on consensus ADMM, where each node solves smallsub-problems using only local data, or on stochastic gradient methods thatdon't scale well to large numbers of cores in a cluster setting. For this reason, GPU clusters have become common prerequisites to large-scale machinelearning. This paper describes an unconventional training method that uses alternating direction methods and Bregman iteration to train a variety of machine learning models on CPUs while avoiding the drawbacks of consensus methods and without gradient descent steps. Using transpose reduction strategies, the proposed method reduces the optimization problems to a sequence of minimization sub-steps that can each be solved globally in closed form. The method provides strong scaling in the distributed setting, yielding linear speedups even when split over thousands of cores.

Current practice in whole time series clustering of residential meter data focuses on aggregated or subsampled load data at the customer level, which ignores day-to-day differences within customers. This information is critical to determine each customer’s suitability to various demand side management strategies that support intelligent power grids and smart energy management. Clustering daily load shapes provides fine-grained information on customer attributes and sources of variation for subsequent models and customer segmentation. In this paper, we apply 11 clustering methods to daily residential meter data. We evaluate their parameter settings and suitability based on 6 generic performance metrics and post-checking of resulting clusters. Finally, we recommend suitable techniques and parameters based on the goal of discovering diverse daily load patterns among residential customers. To the authors’ knowledge, this paper is the first robust comparative review of clustering techniques applied to daily residential load shape time series in the power systems’ literature.

We present a framework for automated analysis and categorization of .onion websites in the darkweb to facilitate analyst situational awareness of new content that emerges from this dynamic landscape. Over the last two years, our team has developed a large-scale darkweb crawling infrastructure called OnionCrawler that acquires new onion domains on a daily basis, and crawls and indexes millions of pages from these new and previously known .onion sites. It stores this data into a research repository designed to help better understand Tor’s hidden service ecosystem. The analysis component of our framework is called Automated Tool for Onion Labeling (ATOL), which introduces a two-stage thematic labeling strategy: (1) it learns descriptive and discriminative keywords for different categories, and (2) uses these terms to map onion site content to a set of thematic labels. We also present empirical results of ATOL and our ongoing experimentation with it, as we have gained experience applying it to the entirety of our darkweb repository, now over 70 million indexed pages. We find that ATOL can perform site-level thematic label assignment more accurately than keywordbased schemes developed by domain experts — we expand the analyst-provided keywords using an automatic keyword discovery algorithm, and get 12% gain in accuracy by using a machine learning classification model. We also show how ATOL can discover categories on previously unlabeled onions and discuss applications of ATOL in supporting various analyses and investigations of the darkweb.

Recent developments in SCADA (Supervisory Control and Data Acquisition) systems for physical infrastructure, such as high pressure gas pipeline systems and electric grids, have generated enormous amounts of time series data. This data brings great opportunities for advanced knowledge discovery and data mining methods to identify system failures faster and earlier than operation experts. This paper presents our effort in collaboration with a utility company to solve a grand challenge; namely, to use advanced data mining methods to detect leaks on a high pressure gas transmission system. Leak detection models with unsupervised learning tasks were developed analyzing billions of data records to identify leaks of different sizes and impacts, with very low false positive rates. In particular, our solution was able to identify small leaks leading to rupture events. The model also identified small leaks not identifiable with current detection systems. Such high-fidelity early identification enables operation personnel to take preventive measures against possible catastrophic events. We then formulate several generic detection methods with models derived from time series anomaly detection methods. We show that our leak detection models are superior to the SCADA alarm system, a mass balance model and other generic time series anomaly detection models in terms of both detection accuracy and computation time.

Learning approximate payoff functions is important to understand the dynamics in multi-player interactions. In general repeat games, each player's payoff can be represented as a combination of all other players' action choices using normal forms, which grow exponentially as the number of action choices increases. Graphical games, however, provide a compact representation to specify the inter-relations where one player's action choice is influenced by its neighbourhood. In this paper, we present how to learn players' approximate payoff functions from normal-form representations, yet also learn a compact graphical game representation of the inter-relations among the players. In this normal form representation, we explore the structural connections of mutual influence between players' action choices in game playing. We formally describe the problem of learning a player influence network and give a novel reward structure-learning algorithm for multiagent graphical games, called the Multi-Descendent Regression Learning Structure Algorithm (MDRLSA). We evaluate MDRLSA on random graphical games generated in GAMUT. Experiments show that MDRLSA can efficiently identify the independence among players and extract the influence graph accurately. The running time of MDRLSA increases linearly with the number of strategy profiles of a game. Compared with state-of-the-art graphical game model learning methods, MDRLSA shows efficiency in terms of time and accuracy.

In this paper, we study inexact damped Newton method implemented in a distributed environment. We are motivated by the original DiSCO algorithm [Communication-Efficient Distributed Optimization of Self-Concordant Empirical Loss, Yuchen Zhang and Lin Xiao, 2015].We show that this algorithm may not scale well and propose algorithmic modifications which lead to fewer communications and better load-balancing between nodes. Those modifications lead to a more efficient algorithm with better scaling. This was made possibly by introducing our new pre-conditioner which is specially designed so that the preconditioning step can be solved exactly and efficiently.Numerical experiments for minimization of regularized empirical loss with a 273GB instance shows the efficiency of proposed algorithm.

In distributed machine learning, data is dispatched to multiple machines for processing. Motivated by the fact that similar data points often belong to the same or similar classes, and more generally, classification rules of high accuracy tend to be "locally simple but globally complex" (Vapnik and Bottou 1993), we propose data dependent dispatching that takes advantage of such structure. We present an in-depth analysis of this model, providing new algorithms with provable worst-case guarantees, analysis proving existing scalable heuristics perform well in natural non worst-case conditions, and techniques for extending a dispatching rule from a small sample to the entire distribution. We overcome novel technical challenges to satisfy important conditions for accurate distributed learning, including fault tolerance and balancedness. We empirically compare our approach with baselines based on random partitioning, balanced partition trees, and locality sensitive hashing, showing that we achieve significantly higher accuracy on both synthetic and real world image and advertising datasets. We also demonstrate that our technique strongly scales with the available computing power.

We introduce a novel approach for parallelizing MCMC inference in models with spatially determined conditional independence relationships, for which existing techniques exploiting graphical model structure are not applicable. Our approach is motivated by a model of seismic events and signals, where events detected in distant regions are approximately independent given those in intermediate regions. We perform parallel inference by coloring a factor graph defined over regions of latent space, rather than individual model variables. Evaluating on a model of seismic event detection, we achieve significant speedups over serial MCMC with no degradation in inference quality.

Two popular approaches for distributed training of SVMs on big data are parameter averaging and alternating direction method of multipliers (ADMM). Parameter averaging is efficient but suffers from loss of accuracy with increase in number of partitions, while ADMM in the feature space is accurate but suffers from slow convergence. In this paper, we report a hybrid approach called weighted parameter averaging (WPA), which optimizes the regularized hinge loss with respect to weights on parameters. The problem is shown to be same as solving SVM in a projected space. We also demonstrate an O(1/N) stability bound on final hypothesis given by WPA, using novel proof techniques. Experimental results on a variety of toy and real world datasets show that our approach is significantly more accurate than parameter averaging for high number of partitions. It is also seen the proposed method enjoys much faster convergence compared to ADMM in feature space.