The Steiner tree problem (STP) is a challenging NP-Hard combinatorial optimization problem. The STP with revenue, budget and hop constraints (STPRBH) determines a subtree of a given undirected graph with the defined constraints. In this study, we propose a novel self-adaptive and stagnation-aware breakout local search (BLS) algorithm (Grid-BLS) for the solution of the STPRBH. The proposed Grid-BLS is a parallel algorithm and keeps the parameters of the BLS heuristic in a population at the master node and tunes/updates them with the best performing parameters sent by the slave nodes. The parameter tuning of the BLS heuristic is considered as another optimization job and processed by a genetic algorithm that runs on the master node. The slave nodes perform BLS search and use a multistarting technique that prevents them to get stuck in a local optima by restarting the search processes. A master and slave communication topology is used for communicating with the slave processors. In order to evaluate the performance of the Grid-BLS algorithm, experiments are carried out on 240 benchmark problem instances. The solutions for 226 of these problems are reported to be optimal or the best solutions. The Grid-BLS achieves 21 new best solutions (graphs) that have never been found by any heuristic algorithm so far and performs better than the state-of-the-art heuristic algorithms Greedy, Destroy&Repair, Tabu Search, and Dynamic Memetic.

Performance and compliance objectives have become crucial for critical sites. Balancing security, energy and costs with operational objectives is the current main challenge for critical site administrators. While current systems are not integrated enough to provide an overall view of the site performances on such targets, the FUSE-IT project intend to propose a new paradigm: the convergence of monitoring on energy, building and facilities, cyber and physical security, and ICT will leverage critical sites activities and offer new opportunity in the detection of threats or savings. In this paper, we detail the first building block of the FUSE-IT concept which lies in the definition of cross-domain (i.e. on energy, security, ICT and facility) Key Performance Indicators (KPIs). These KPIs are the main decision variables for an enhanced Decision Support System (DSS) meeting the next century impediments of critical sites.

Bu çalışmanın hedefi, Dil Modeli (DM) üretmek için kullanılan metin derlem büyüklüğünün, Ses Tanıma Sistemleri (STS) üzerindeki etkisini araştırmaktır. Çalışmada ayrıca DM elde etmek için yapılması gereken işler detaylı olarak anlatılmaktadır. DM istatistiksel olarak oluşturulduğundan, eğitim verisinde bulunan veri miktarı arttıkça STS doğruluğunun artması beklenmektedir. Fakat Türkçe gibi sondan eklemeli dillerde, kullanılan derlemin büyüklüğünün hangi noktaya kadar sistemin doğruluk oranı üzerinde etkin olacağı önem taşımaktadır. Bu çalışmada, toplanan farklı büyüklükteki metin derlemleri ile konuşma tanıma sisteminde Dil Model Ağırlığı (DMA) ve Aktif Token Sayısı (ATS) parametrelerini değiştirerek yapılan deneyler yer almaktadır. Bu çalışma DM boyutu büyüdükçe Türkçe konuşma tanıma başarımının yükseldiğini göstermektedir. Ancak, DMA ve ATS değerlerinde yapılan ayarlamaların tanıma başarımına olumlu bir etki yaptığı gözlemlenememiştir.

In this paper we aimed at investigating the effect of Language Model (LM) size on Speech Recognition (SR) accuracy. We also provided details of our approach for obtaining the LM for Turkish. Since LM is obtained by statistical processing of raw text, we expect that by increasing the size of available data for training the LM, SR accuracy will improve. Since this study is based on recognition of Turkish, which is a highly agglutinative language, it is important to find out the appropriate size for the training data. The minimum required data size is expected to be much higher than the data needed to train a language model for a language with low level of agglutination such as English. In the experiments we also tried to adjust the Language Model Weight (LMW) and Active Token Count (ATC) parameters of LM as these are expected to be different for a highly agglutinative language. We showed that by increasing the training data size to an appropriate level, the recognition accuracy improved on the other hand changes on LMW and ATC did not have a positive effect on Turkish speech recognition accuracy.

The supply-demand equilibrium is the main criteria for determination of electricity pricing for both electrical power production companies and ordinary (household) users. The companies must be sure about future demands of electricity for uninterrupted efficient electrical supply. The demand of electricity is affected from weather conditions, process of economy, working and nonworking days of a year, etc. Therefore, forecasting demand by using current and historical data is very important for electricity trading and producing companies. In this study, a cloud-based forecasting service which is based on neural network model is proposed for long-term electricity demand forecasting of Turkey. Cloud based nature of the proposed system help continuous training and improved forecasting capability over time from the system. Following year overall electric demand is approximately estimated with neural network and artificial neuro-fuzzy inference systems.

Original scientific paper Confidence measures are expected to give a measure of reliability on the result of a speech/speaker recognition system. Most commonly used confidence measures are based on posterior word or phoneme probabilities which can be obtained from the output of the recognizer. In this paper we introduced a linear interpretation of posterior probability based confidence measure by using inverse Fisher transformation. Speaker adaptation consists in updating model parameters of a speaker independent model to have a better representation of the current speaker. Confidence measures give more reliable selection criteria to select the utterances which best represent the speaker. A linear interpretation of confidence measure is very important to select the most representative data for adaptation.

Confidence measures are expected to give a measure of reliability on the result of a speech/speaker recognition system. Most commonly used confidence measures are based on posterior word or phoneme probabilities which can be obtained from the output of the recognizer. In this paper we introduced a linear interpretation of posterior probability based confidence measure by using inverse Fisher transformation. Speaker adaptation consists in updating model parameters of a speaker independent model to have a better representation of the current speaker. Confidence measures give more reliable selection criteria to select the utterances which best represent the speaker. A linear interpretation of confidence measure is very important to select the most representative data for adaptation.

The long term electricity demand forecasting is evaluated by using past demand and weather data. The aim of this study is to present the effect of the variety of inputs on performance. Therefore neural network models are evaluated due to their performance and common acceptance. Different varieties of inputs are applied to the model and results are compared with each other.

This paper considers some specific issues relating to model-driven system management applied to complex systems. Exami ning dynamically coupled systems-of-systems on the one hand and high ly distributed devices for service access on the other, we define a common met aodel of (semi-) automated management applicable in both domains. Ta king monitoring by way of illustration, we then show how this meta-model i s put into practice along two complementary aspects: management modelling and run time event processing support.

This paper describes IBM's and Telefónica's joint work in progress in the field of Domain-Specific Modelling applied to Autonomic Network Management in the context of the European IST FP6 project MODELPLEX. In modelling edge-of-network devices which act as service access points from consumers to the telephone network, we have introduced a dynamic aspect to the network topology which affects the way Service Level Agreements (SLAs) should be modelled and processed within an autonomic management control loop. It is suggested that it is not possible to separate system and rule for managing services across a service-provider network.

Multilingual access to information and services is a key requirement in any pervasive or ubiquitous computing environment. In this paper we review the design of a common alphabet for up to fifteen languages and describe its application to multilingual speech recognition in low-resource devices in real-time. We give an overview of the special requirements for acoustic modeling in such environments and present initial results of a technique that aims on a more efficient discrimination between languages in training while keeping low memory footprint. We also report the usefulness of a multilingual recognizer as a language-independent system to bootstrap a new language.

This paper provides an automated system management and service provisioning approach for distributed devices. We consider a functional process model of the necessary phases for runtime management (MAPE-K proposal from the Autonomic Computing Initiative). This approach defines discrete steps within runtime system management which provide a simplified view of the system for service provisioning. Two issues emerge: communication from managed entities (or monitoring) may need special consideration; and this leads on to modeling communication between the system management phases proposed by MAPE-K.

In this paper, a new acoustic confidence measure of automatic speech recognition hypothesis is proposed and it is compared to approaches proposed in the literature. This approach takes into account prior information on the acoustic model performance specific to each phoneme. The new method is tested on two types of recognition errors: the out-of-vocabulary words and the errors due to additive noise. An efficient way to interpret the raw confidence measure as a correctness prior probability is also proposed in the paper.

Confidence measures are expected give a measure of reliability on the result of a speech/speaker recognition system. Most commonly used confidence measures are based on posterior word or phoneme probabilities which can be obtained from the output of the recognizer. In this paper we introduced a linear interpretation of posterior probability based confidence measure by using inverse Fisher transformation. Speaker adaptation consists in updating model parameters of a speaker independent model to have a better representation of the current speaker. Confidence measures give a more reliable selection criteria to select the utterances which best represents the speaker. A linear interpretation of confidence measure is very important to select the most representative data for adaptation.

In this paper, we propose a new acoustic confidence measure of ASR hypothesis and compare it to approaches proposed in the literature. This approach takes into account prior information on the acoustic model performance specific to each phoneme. The new method is tested on two types of recognition errors: the out-of-vocabulary words and the errors due to additive noise. We then propose an efficient way to interpret the raw confidence measure as a correctness prior probability.

Turkish language is an agglutinative language. It is possible to produce a very high number of words from the same root with suffixes [1]. Language modeling for agglutinative languages needs to be different than modeling of languages like English. Such languages also have inflections but not as many as an agglutinative language. Techniques which can be used for modeling agglutinative languages are presented in this work. Turkish is one of the least studied language for speech recognition. For this reason the first step for Turkish speech recognition is preparing a database. The texts to record the database were selected from television programs and newspaper articles. Selection criterion was to cover various subject and to create a phonetically balanced corpus. Additionally it is important to include as many different word as possible. The Speech Training and Recognition Unified Tool (STRUT)1 has been used for training and testing systems for preliminary recognition experiments.

A confidence measure is defined as the posterior probability of word correctness given the values of confidence indicators [8]. Confidence measures can be calculated from a posteriori probability of recognized word or sub-word unit inferred from some acoustic models and language models, including various normalization techniques. In this work we present several confidence measures and propose to apply them to automatic recognition of the Turkish language. Turkish language is an inflected language. It is possible to produce a very high number of words from the same root with suffixes [4]. Hence confidence measures appear to be promising techniques to improve the current performance of the speech recognition for Turkish language. For experimental purpose, a Turkish database has been developed. Its content will be described in this communication. The Speech Training and Recognition Unified Tool (STRUT) toolkit has been used for training of the models used in the speech recognition system. Keywords— Confidence measure, Turkish speech recognition, Language modelling for agglutinative languages, Hybrid speech recognition, Phonetic labeling.