Abstract
This paper addresses the issue of learning time series forecasting models in changing environments by leveraging the predictive power of ensemble methods. Concept drift adaptation is performed in an active manner, by dynamically combining base learners according to their recent performance using a non-linear function. Diversity in the ensembles is encouraged with several strategies that include heterogeneity among learners, sampling techniques and computation of summary statistics as extra predictors. Heterogeneity is used with the goal of better coping with different dynamic regimes of the time series. The driving hypotheses of this work are that (i) heterogeneous ensembles should better fit different dynamic regimes and (ii) dynamic aggregation should allow for fast detection and adaptation to regime changes. We extend some strategies typically used in classification tasks to time series forecasting. The proposed methods are validated using Monte Carlo simulations on 16 real-world univariate time series with numerical outcome as well as an artificial series with clear regime shifts. The results provide strong empirical evidence for our hypotheses. To encourage reproducibility the proposed method is publicly available as a software package.

Abstract
The structural patterns in the neighborhood of nodes assign unique roles to the nodes. Mining the set of existing roles in a network provides a descriptive profile of the network and draws its general picture. This paper proposes a new method to determine structural roles in a dynamic network based on the current position of nodes and their historic behavior. We develop a temporal ensemble clustering technique to dynamically find groups of nodes, holding similar tempo-structural roles. We compare two weighting functions, based on age and distribution of data, to incorporate temporal behavior of nodes in the role discovery. To evaluate the performance of the proposed method, we assess the results from two points of view: 1) goodness of fit to current structure of the network; 2) consistency with historic data. We conduct the evaluation using different ensemble clustering techniques. The results on real world networks demonstrate that our method can detect tempo-structural roles that simultaneously depict the topology of a network and reflect its dynamics with high accuracy. Copyright 2017 ACM.

Abstract

Abstract
The brain emotional learning (BEL) control paradigm has been gathering increased interest by the control systems design community. However, the lack of a consistent mathematical formulation and computer based tools are factors that have prevented its more widespread use. In this article both features are tackled by providing a coherent mathematical framework for both the continuous and discrete-time formulations and by presenting a SIMULINK (R) computational tool that can be easily used for fast prototyping BEL based control systems.

Abstract
Optimising the design of water distribution networks (WDNs) is a well-known problem that has been studied by numerous researchers. This work proposes a heuristic based on simulated annealing and improved by using concepts from the cross-entropy method. The proposed optimization approach is presented and used in two case studies of different complexity. The results show not only a fall in the computational effort of the new approach relative to simulated annealing but also include a comparison with other heuristic results from the literature, used to solve the same problems.

Abstract
Context. Z CMa is a complex pre-main sequence binary with a current separation of 110 mas, known to consist of an FU Orionis star (SE component) and an embedded Herbig Be star (NW component). Although it represents a well-studied and characterized system, the origin of photometric variabilities, the component properties, and the physical configuration of the system remain mostly unknown. Aims. Immediately when the late-2008 outburst of Z CMa was announced to the community, we initiated a high angular resolution imaging campaign aimed at characterizing the outburst state of both components of the system in the near-infrared. Methods. We used the VLT/NACO and the Keck/NIRC2 near-infrared adaptive optics instrument to monitor the astrometric position and the near-infrared photometry of the Z CMa components during the outburst phase and one year after. The VLT/SINFONI and Keck/OSIRIS integral field spectroscrographs were in addition used to characterize for the first time the resolved spectral properties of the FU Orionis and the Herbig Be component during and after the outburst. Results. We confirm that the NW star dominates the system flux in the 1.1-3.8 mu m range and is responsible for the photometric outburst. We extract the first medium-resolution (R similar to 2000-4000) near-infrared (1.1-2.4 mu m) spectra of the individual components. The SE component has a spectrum typical of FU Orionis objects. The NW component spectrum is characteristic of embedded outbursting protostars and EX Or objects. It displays numerous emission lines whose intensity correlates with the system activity. In particular, we find a correlation between the Br gamma equivalent width and the system brightness. The bluing of the continuum of the NW component along with the absolute flux and color-variation of the system during the outburst suggests that the outburst was caused by a complex interplay between a variation of the extinction in the line of sight of the NW component on one hand, and the emission of shocked regions close to the NW component on the other. We confirm the recently reported wiggling of the SE component jet from [Fe II] line emission. We find a point-like structure associated with a peak emission at 2.098 mu m coincidental with the clump or arm seen in broadband polarization di ff erential imaging as well as additional di ff use emission along a PA = 214 degrees. The origin of these two structures is unclear and deserves further investigation.