Platform as a service gateway for the Fog of Things

Internet of Things (IoT), one of the key research topics in recent years, together with concepts from Fog Computing, brings rapid advancements in Smart City, Monitoring Systems, industrial control, transportation and other fields. These applications require a reconfigurable sensor architecture that can span multiple scenarios, devices and use cases that allow storage, networking and computational resources to be efficiently used on the edge of the network. There are a number of platforms and gateway architectures that have been proposed to manage these components and enable application deployment. These approaches lack horizontal integration between multiple providers as well as higher order functionalities like load balancing and clustering. This is partly due to the strongly coupled nature of the deployed applications, a lack of abstraction of device communication layers as well as a lock-in for communication protocols. This limitation is a major obstacle for the development of a protocol agnostic application environment that allows for single application to be migrated and to work with multiple peripheral devices with varying protocols from different local gateways. This research looks at existing platforms and their shortcomings as well as proposes a messaging based modular gateway platform that enables clustering of gateways and the abstraction of peripheral communication protocol details. These novelties allow applications to send and receive messages regardless of their deployment location and destination device protocol, creating a more uniform development environment. Furthermore, it results in a more streamlined application development and testing while providing more efficient use of the gateway’s resources. Our evaluation of a prototype for the system shows the need for the migration of resources and the QoS advantages of such a system. The examined use case scenarios show that clustering proves to be an advantage in certain use cases as well as presenting the deployment of a larger testing and control environment through the platform.     

Modeling under-resourced languages for speech recognition

One particular problem in large vocabulary continuous speech recognition for low-resourced languages is finding relevant training data for the statistical language models. Large amount of data is required, because models should estimate the probability for all possible word sequences. For Finnish, Estonian and the other fenno-ugric languages a special problem with the data is the huge amount of different word forms that are common in normal speech. The same problem exists also in other language technology applications such as machine translation, information retrieval, and in some extent also in other morphologically rich languages. In this paper we present methods and evaluations in four recent language modeling topics: selecting conversational data from the Internet, adapting models for foreign words, multi-domain and adapted neural network language modeling, and decoding with subword units. Our evaluations show that the same methods work in more than one language and that they scale down to smaller data resources.     

Toward large-scale Hybrid Monte Carlo simulations of the Hubbard model on graphics processing units

One of the most efficient non-perturbative methods for the calculation of thermal properties of quantum systems is the Hybrid Monte Carlo algorithm, as evidenced by its use in large-scale lattice quantum chromodynamics calculations. The performance of this algorithm is determined by the speed at which the fermion operator is applied to a given vector, as it is the central operation in the preconditioned conjugate gradient iteration. We study a simple implementation of these operations for the fermion matrix of the Hubbard model in d+1 spacetime dimensions, and report a performance comparison between a 2.66 GHz Intel Xeon E5430 CPU and an NVIDIA Tesla C1060 GPU using double-precision arithmetic. We find speedup factors ranging between 30 and 350 for d=1, and in excess of 40 for d=3. We argue that such speedups are of considerable impact for large-scale simulational studies of quantum many-body systems.     

Bidding strategies for real-life small loan auctions

We define and identify bidding strategies in real-life small loan auctions (Prosper.com). In such auctions, lenders bid for borrowers' loan listings and the winners get to fund the loan at an interest rate determined by the auction. The exceptionally large empirical database provided by Prosper.com offers a unique opportunity to test and further develop the theory of online auctions. This study shows that bidding behavior is not homogeneous among bidders, as the traditional auction theory suggests. Instead, bidders use many different bidding strategies. Moreover, learning and bidders' consistency over time in different auctions is studied.     

Water Resource Models in the Mekong Basin: A Review

Development of the water resources of the Mekong Basin is the subject of intense debate both within the Mekong region and internationally. Water resources modelling is playing an increasingly important role in the debate, with significant effort in building integrated modelling platforms to describe the hydrological, ecological, social and economic impacts of water resource development. In the hydrological domain, a comprehensive set of models has been effective in building understanding of the system, and in identifying and describing the issues and trade-offs involved in basin-scale water planning. In the ecological and social domains, quantitative modelling has not progressed very far; geo-spatial analysis and qualitative frameworks remain the most commonly used tools. Economic models have been used to assess the costs and benefits of water resources development and to describe the trade-offs between different sectors and users. These analyses are likely to play an important role in the policy and planning debate, but are hampered by uncertainties in valuation of ecosystem services. Future efforts should focus on optimising the use of existing model platforms for the Mekong, including structured comparison of multiple hydrological models to quantify errors and identify an optimum set of modelling tools for different applications. A comprehensive research effort is needed to incorporate groundwater into hydrological models for regional planning. Options for social impact assessment should be reassessed before major investments are made in complex modelling platforms, and participatory social survey methods evaluated as part of an integrated assessment framework.     

Nitrate estimation in the denitrifying post-filtration unit of a municipal wastewater treatment plant: the Viikinm?ki case

In this work we present and discuss the design of an array of soft-sensors to estimate the nitrate concentration in the denitrifying post-filtration unit at the Viikinm?ki wastewater treatment plant in Helsinki (Finland). The developed sensors aim at supporting the existing hardware analyzers by providing a reliable back-up system in case of malfunction of the instruments. In the attempt to design easy to implement and interpretable sensors, computationally light linear models have been considered. However, due to the intrinsic nonlinearity of the process, also nonlinear but still computationally affordable models have been considered for comparison. The experimental results demonstrate the potential of the developed soft-sensors and the possibility for an on-line implementation in the plant's control system as alternative monitoring devices.     

Non-myopic vehicle and route selection in dynamic DARP with travel time and workload objectives

In dial-a-ride problems, a fleet of n vehicles is routed to transport people between pick-up and delivery locations. We consider an elementary version of the problem where trip requests arrive in time and require an immediate vehicle assignment (which triggers an appropriate route update of the selected vehicle). In this context, a relatively general objective can be stated as a weighted sum of the system's effort and the customers' inconvenience. However, optimizing almost any objective in this immensely complex stochastic system is prohibitively difficult. Thus the earlier work has largely resorted to heuristic cost functions that arise, e.g., from the corresponding static systems. By using the framework of Markov decision processes and the classical M/M/1 queue as a highly abstract model for a single vehicle, we explain why certain intuitive cost functions indeed give satisfactory results in the dynamic system, and also give an explicit interpretation of different components appearing in a general cost function. The resulting family of heuristic control policies is demonstrated to offer a desired type of performance thus justifying the assumed analogy between a multi-queue and dial-a-ride systems.     

Eigenclassifiers for combining correlated classifiers

In practice, classifiers in an ensemble are not independent. This paper is the continuation of our previous work on ensemble subset selection [A. Ula?, M. Semerci, O.T. Y?ld?z, E. Alpayd?n, Incremental construction of classifier and discriminant ensembles, Information Sciences, 179 (9) (2009) 1298–1318] and has two parts: first, we investigate the effect of four factors on correlation: (i) algorithms used for training, (ii) hyperparameters of the algorithms, (iii) resampled training sets, (iv) input feature subsets. Simulations using 14 classifiers on 38 data sets indicate that hyperparameters and overlapping training sets have higher effect on positive correlation than features and algorithms. Second, we propose postprocessing before fusing using principal component analysis (PCA) to form uncorrelated eigenclassifiers from a set of correlated experts. Combining the information from all classifiers may be better than subset selection where some base classifiers are pruned before combination, because using all allows redundancy.