Search

Creating an intelligent system that is able to generalize and reach human or above-human performance in a variety of tasks will be part of the crowning achievement of Artificial General Intelligence. However, even though many steps have been taken towards this direction, they have critical shortcomings that prevent the research community from drawing a clear path towards that goal, such as limited learning capacity of a model, sample-inefficiency or low overall performance. In this paper, we propose GENEREIT, a meta-Reinforcement Learning model in which a single Deep Reinforcement Learning agent (meta-learner) is able to produce high-performance agents (inner-learners) for solving different environments under a single training session, in a sample-efficient way, as shown by primary results in a set of various toy-like environments. This is partially due to the fixed subset selection strategy implementation that allows the meta-learner to focus on tuning specific traits of the generated agents rather than tuning them completely. This, combined with our system’s modular design for introducing higher levels in the meta-learning hierarchy, can also be potentially immune to catastrophic forgetting and provide ample learning capacity.

Personality Traits are one of the most important aspects of human behaviour analysis that can be widely used in a plethora of scientific fields. Nowadays, a variety of industries use self-assessment techniques to evaluate their employees or candidates to form effective teams based on the relationships between the personnel. The Big Five personality traits, also known as the OCEAN Five model, encompass an individual’s behaviour on five factors: Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. Currently, the evaluation is based on questionnaires, which are prone to subjectivity and can be biased. In this paper, we analyse the behaviours on which each trait is based, in a series of 3D Escape Rooms with the aim of collecting metrics in a number of settings/situations to effectively assess them, with the use of gamification. Each Escape Room has unique features and simulates special occasions to monitor the player's reactions as well as gameplay style to generate specific models that calculate the level of each personality trait. The final profiling, however, is based on a combination of metrics from the different rooms since the behaviours are common and overlap with each other but can be evaluated in different situations. Results validate the mathematical models of metrics for each behaviour and crystallise the nature of the situations in which these behaviours can be modelled. Therefore, through gamification, a new generic model for categorising players, based on the Ocean five personality trait model, is generated.

Cryptocurrency markets experienced a significant increase in the popularity, which motivated many financial traders to seek high profits in cryptocurrency trading. The predominant tool that traders use to identify profitable opportunities is technical analysis. Some investors and researchers also combined technical analysis with machine learning, in order to forecast upcoming trends in the market. However, even with the use of these methods, developing successful trading strategies is still regarded as an extremely challenging task. Recently, deep reinforcement learning (DRL) algorithms demonstrated satisfying performance in solving complicated problems, including the formulation of profitable trading strategies. While some DRL techniques have been successful in increasing profit and loss (PNL) measures, these techniques are not much risk-aware and present difficulty in maximizing PNL and lowering trading risks simultaneously. This research proposes the combination of DRL approaches with rule-based safety mechanisms to both maximize PNL returns and minimize trading risk. First, a DRL agent is trained to maximize PNL returns, using a novel reward function. Then, during the exploitation phase, a rule-based mechanism is deployed to prevent uncertain actions from being executed. Finally, another novel safety mechanism is proposed, which considers the actions of a more conservatively trained agent, in order to identify high-risk trading periods and avoid trading. Our experiments on 5 popular cryptocurrencies show that the integration of these three methods achieves very promising results.

Distribution of IGHV4-34 CLL according to IGHV mutational status and subset membership.

Purpose: We sought to investigate whether B cell receptor immunoglobulin (BcR IG) stereotypy is associated with particular clinicobiological features among chronic lymphocytic leukemia (CLL) patients expressing mutated BcR IG (M-CLL) encoded by the IGHV4-34 gene, and also ascertain whether these associations could refine prognostication.Experimental Design: In a series of 19,907 CLL cases with available immunogenetic information, we identified 339 IGHV4-34–expressing cases assigned to one of the four largest stereotyped M-CLL subsets, namely subsets #4, #16, #29 and #201, and investigated in detail their clinicobiological characteristics and disease outcomes.Results: We identified shared and subset-specific patterns of somatic hypermutation (SHM) among patients assigned to these subsets. The greatest similarity was observed between subsets #4 and #16, both including IgG-switched cases (IgG-CLL). In contrast, the least similarity was detected between subsets #16 and #201, the latter concerning IgM/D-expressing CLL. Significant differences between subsets also involved disease stage at diagnosis and the presence of specific genomic aberrations. IgG subsets #4 and #16 emerged as particularly indolent with a significantly (P < 0.05) longer time-to-first-treatment (TTFT; median TTFT: not yet reached) compared with the IgM/D subsets #29 and #201 (median TTFT: 11 and 12 years, respectively).Conclusions: Our findings support the notion that BcR IG stereotypy further refines prognostication in CLL, superseding the immunogenetic distinction based solely on SHM load. In addition, the observed distinct genetic aberration landscapes and clinical heterogeneity suggest that not all M-CLL cases are equal, prompting further research into the underlying biological background with the ultimate aim of tailored patient management. Clin Cancer Res; 23(17); 5292–301. ©2017 AACR.

Univariable and multivariable analysis for TTFT within M-CLL.

BACKGROUND Attention Deficit Hyperactivity Disorder (ADHD) is one of the most common neurodevelopmental disorders during childhood, however the diagnosis procedure remains challenging as it is non-standardized, multi-parametric and highly dependent on subjective evaluation of the perceived behavior. OBJECTIVE To address the challenges of existing procedures for ADHD diagnosis, the ADHD360 project aims to develop a platform for (a) early detection of ADHD by assessing the user’s likelihood of having ADHD characteristics and (b) providing complementary training for ADHD management. METHODS A two-phase pilot study was designed to evaluate the ADHD360 platform, including ADHD and non-ADHD participants aged 7-16 years. Machine Learning methods were used to detect discriminative gameplay patterns among the two groups (ADHD, non-ADHD) and estimate a player’s likelihood of having ADHD characteristics. RESULTS A preliminary analysis of collected data showed that the trained models achieve high performance in correctly predicting a user’s label (ADHD or non-ADHD) from his gameplay session in the ADHD360 platform. CONCLUSIONS ADHD360 is characterized by notable capacity to discriminate player gameplay behavior as either ADHD or non-ADHD. Therefore, the ADHD360 platform could be a valuable complementary tool for early ADHD detection. CLINICALTRIAL ClinicalTrials.gov NCT04362982

Deep Reinforcement Learning is a topic that has gained a lot of attention recently, due to the unprecedented achievements and remarkable performance of such algorithms in various benchmark tests and environmental setups. The power of such methods comes from the combination of an already established and strong field of Deep Learning, with the unique nature of Reinforcement Learning methods. It is, however, deemed necessary to provide a compact, accurate and comparable view of these methods and their results for the means of gaining valuable technical and practical insights. In this work we gather the essential methods related to Deep Reinforcement Learning, extracting common property structures for three complementary core categories: a) Model-Free, b) Model-Based and c) Modular algorithms. For each category, we present, analyze and compare state-of-the-art Deep Reinforcement Learning algorithms that achieve high performance in various environments and tackle challenging problems in complex and demanding tasks. In order to give a compact and practical overview of their differences, we present comprehensive comparison figures and tables, produced by reported performances of the algorithms under two popular simulation platforms: the Atari Learning Environment and the MuJoCo physics simulation platform. We discuss the key differences of the various kinds of algorithms, indicate their potential and limitations, as well as provide insights to researchers regarding future directions of the field.

This study proposes a modular water monitoring IoT system that enables quantitative and qualitative measuring of water in terms of an upgraded version of the water infrastructure to sustain operational reliability. The proposed method could be used in urban and rural areas for consumption and quality monitoring, or eventually scaled up to a contemporary water infrastructure enabling water providers and/or decision makers (i.e., governmental authorities, global water organization, etc.) to supervise and drive optimal decisions in challenging times. The inherent resilience and agility that the proposed system presents, along with the maturity of IoT communications and infrastructure, can lay the foundation for a robust smart water metering solution. Introducing a modular system can also allow for optimal consumer profiling while alleviating the upfront adoption cost by providers, environmental stewardship and an optimal response to emergencies. The provided system addresses the urbanization and technological gap in the smart water metering domain by presenting a modular IoT architecture with consumption and quality meters, along with machine learning capabilities to facilitate smart billing and user profiling.

Part 2: AI in Biomedical Applications; International audience; Biomedical Named Entity Recognition is a difficult task, aimed to identify all named entities in medical literature. The importance of the task becomes apparent as these entities are used to identify key features, enable better search results and can accelerate the process of reviewing related evidence to a medical case. This practice is known as Evidence-Based Medicine (EBM) and is globally used by medical practitioners who do not have the time to read all the latest developments in their respective fields. In this paper we propose a methodology which achieves state-of-the-art results in a plethora of Biomedical Named Entity Recognition datasets, with a lightweight approach that requires minimal training. Our model is end-to-end and capable of efficiently modeling significantly longer sequences than previous models, benefiting from inter-sentence dependencies.

Drug-Drug Interaction (DDI) extraction is the task of identifying drug entities and the potential interactions between drug pairs from biomedical literature. Computer-aided extraction of DDIs is vital for drug discovery, as this process remains extremely expensive and time consuming. Therefore, Machine Learning-based approaches can reduce the laborious task during the drug development cycle. Numerous traditional and Neural Network-based approaches for Drug Named Entity Recognition (DNER) and the classification of DDIs have been proposed over the years. However, despite the development of many effective methods, achieving good prediction accuracy is an area where significant improvement can be made. In this article, we present a novel end-to-end approach that tackles the overall DDI extraction task as a pipelined method via the Transformer model architecture and biomedical domain pre-trained weights. In our approach, the tasks of DNER and DDI classification are executed successively to extract the drug entities and to classify their relationship respectively. The proposed approach, TP-DDI, integrates prior knowledge by using pre-trained weights from BioBERT and improves in both the Drug Named Entity Recognition and the overall DDI extraction task over the current state-of-the-art approaches on the DDI Extraction 2013 corpus.

Argument Mining (AM) refers to the task of automatically identifying arguments in a text and finding their relations. In medical literature this is done by identifying Claims and Premises and classifying their relations as either Support or Attack. Evidence-Based Medicine (EBM) refers to the task of identifying all related evidence in medical literature to allow medical practitioners to make informed choices and form accurate treatment plans. This is achieved through the automatic identification of Population, Intervention, Comparator and Outcome entities (PICO) in the literature to limit the collection to only the most relevant documents. In this work, we combine EBM with AM in medical literature to increase the performance of the individual models and create high quality argument graphs, annotated with PICO entities. To that end, we introduce a state-of-the-art EBM model, used to predict the PICO entities and two novel Argument Identification and Argument Relation classification models that utilize the PICO entities to enhance their performance. Our final system works in a pipeline and is able to identify all PICO entities in a medical publication, the arguments presented in them and their relations.

Drug-drug interaction (DDI) identification is the task of identifying potential interactions between drugs when administered simultaneously. The interactions can be synergetic or antagonistic as one drug can affect the other. Adverse drug reactions caused by antagonistic DDI can pose a serious threat to health and potentially lead to greater increase in health care expenditure. Multiple excellent resources for DDI already exist, although unable to keep up with the exponential increase in published biomedical literature. Most existing systems rely on handcrafted features to extract and classify the relationships between drugs. In this paper, we present a deep learning method of stacked bidirectional Long Short Term Memory (Bi-LSTM) and Convolutional neural (CNN) networks that utilize word embeddings, part-of-speech tags and distance embeddings respectively to perform the DDI extraction task and aid the drug development cycle and drug repurposing. Furthermore, the model uses attention mechanism to better focus on importance of all the hidden states of the Bi-LSTM layers. Experimental results show that our method can better avoid misclassifications of instances with a minimal preprocessing.

The prediction of the average size of fragments in blasted rock piles produced after blasting in aggregate quarries is essential for decresing the cost of crushing and secondary breaking. There are several conventional and advanced processes to estimate the size of blasted rocks. Among these, the empirical prediction of the expected fragmentation in most cases is carried out by Kuznetsov’s equation (Sov Min Sci 9:144–148, 1973), modified by Lilly (1986) and Cunningham (1987). The present research focuses on the effect of the engineering geological factors and blasting process on the blasted fragments using a more powerful, advanced computational tool, an artificial neural network. In particular, the blast database consists of the blastability index of limestone on the pit face, the quantities of the explosives and of the blasted rock pile, assessing the interaction of these parameters on the blasted rocks. The data were collected from two aggregate quarries, Drymos and Tagarades, near Thessaloniki, in the Central Macedonia region of Greece. This approach indicates significant performance stability, providing the fragmentation size with high accuracy.

PaaS is a Cloud computing service that provides a computing platform to develop, run, and manage applications without the complexity of infrastructure maintenance. SMEs are reluctant to enter the growing PaaS market due to the possibility of being locked in to a certain platform, mostly provided by the market's giants. The PaaSport Marketplace aims to avoid the provider lock-in problem by allowing Platform provider SMEs to roll out semantically interoperable PaaS offerings and Software SMEs to deploy or migrate their applications on the best-matching offering, through a thin, non-intrusive Cloud broker. In this paper, we present the PaaSport semantic model, namely an OWL ontology, extension of the DUL ontology. The ontology is used for semantically representing (a) PaaS offering capabilities and (b) requirements of applications to be deployed. The ontology has been designed to optimally support a semantic matchmaking and ranking algorithm that recommends the best-matching PaaS offering to the application developer. The DUL ontology offers seamless extensibility, since both PaaS Characteristics and parameters are defined as classes; therefore, extending the ontology with new characteristics and parameters requires the addition of new specialized subclasses of the already existing classes, which is less complicated than adding ontology properties. The PaaSport ontology is evaluated through verification tools, competency questions, human experts, application tasks and query performance tests.

In this article, we study the transfer learning model of action advice under a budget. We focus on reinforcement learning teachers providing action advice to heterogeneous students playing the game of Pac-Man under a limited advice budget. First, we examine several critical factors affecting advice quality in this setting, such as the average performance of the teacher, its variance and the importance of reward discounting in advising. The experiments show that the best performers are not always the best teachers and reveal the non-trivial importance of the coefficient of variation (CV) as a statistic for choosing policies that generate advice. The CV statistic relates variance to the corresponding mean. Second, the article studies policy learning for distributing advice under a budget. Whereas most methods in the relevant literature rely on heuristics for advice distribution, we formulate the problem as a learning one and propose a novel reinforcement learning algorithm capable of learning when to advise or not. The proposed algorithm is able to advise even when it does not have knowledge of the student’s intended action and needs significantly less training time compared to previous learning approaches. Finally, in this article, we argue that learning to advise under a budget is an instance of a more generic learning problem: Constrained Exploitation Reinforcement Learning.

Background and objective Single Nucleotide Polymorphism (SNPs) are, nowadays, becoming the marker of choice for biological analyses involving a wide range of applications with great medical, biological, economic and environmental interest. Classification tasks i.e. the assignment of individuals to groups of origin based on their (multi-locus) genotypes, are performed in many fields such as forensic investigations, discrimination between wild and/or farmed populations and others. Τhese tasks, should be performed with a small number of loci, for computational as well as biological reasons. Thus, feature selection should precede classification tasks, especially for Single Nucleotide Polymorphism (SNP) datasets, where the number of features can amount to hundreds of thousands or millions. Methods In this paper, we present a novel data mining approach, called FIFS – Frequent Item Feature Selection, based on the use of frequent items for selection of the most informative markers from population genomic data. It is a modular method, consisting of two main components. The first one identifies the most frequent and unique genotypes for each sampled population. The second one selects the most appropriate among them, in order to create the informative SNP subsets to be returned. Results The proposed method (FIFS) was tested on a real dataset, which comprised of a comprehensive coverage of pig breed types present in Britain. This dataset consisted of 446 individuals divided in 14 sub-populations, genotyped at 59,436 SNPs. Our method outperforms the state-of-the-art and baseline methods in every case. More specifically, our method surpassed the assignment accuracy threshold of 95% needing only half the number of SNPs selected by other methods (FIFS: 28 SNPs, Delta: 70 SNPs Pairwise FST: 70 SNPs, In: 100 SNPs.) Conclusion Our approach successfully deals with the problem of informative marker selection in high dimensional genomic datasets. It offers better results compared to existing approaches and can aid biologists in selecting the most informative markers with maximum discrimination power for optimization of cost-effective panels with applications related to e.g. species identification, wildlife management, and forensics.

Entity Coreference Resolution is the task of resolving all mentions in a document that refer to the same real world entity and is considered as one of the most difficult tasks in natural language understanding. It is of great importance for downstream natural language processing tasks such as entity linking, machine translation, summarization, chatbots, etc. This work aims to give a detailed review of current progress on solving Coreference Resolution using neural-based approaches. It also provides a detailed appraisal of the datasets and evaluation metrics in the field, as well as the subtask of Pronoun Resolution that has seen various improvements in the recent years. We highlight the advantages and disadvantages of the approaches, the challenges of the task, the lack of agreed-upon standards in the task and propose a way to further expand the boundaries of the field., Comment: 52 pages, 8 figures, 4 tables, Published in Expert Systems with Applications

PaaSport is a cloud broker between PaaS providers and cloud application developers.PaaS offerings and application requirements are annotated with a PaaS OWL ontology.A semantic matchmaking and ranking algorithm recommends the best-matching offering.Cloud developers can deploy or migrate applications through the PaaSport broker.The algorithm scales linearly and is extensible. Platform as a service (PaaS) is one of the Cloud computing services that provide a computing platform in the Cloud, allowing customers to develop, run, and manage web applications without the complexity of building and maintaining the infrastructure. The primary disadvantage for an SME to enter the emerging PaaS market is the possibility of being locked into a certain platform, mostly provided by the market's giants. The PaaSport project focuses on facilitating SMEs to deploy business applications on the best-matching Cloud PaaS offering and to seamlessly migrate these applications on demand, via a thin, non-intrusive Cloud-broker, in the form of a Cloud PaaS Marketplace. PaaSport enables PaaS provider SMEs to roll out semantically interoperable PaaS offerings, by annotating them using a unified PaaS semantic model that has been defined as an OWL ontology. In this paper we focus on the recommendation algorithm that has been developed on top of the ontology, for providing the application developer with recommendations about the best-matching Cloud PaaS offering. The algorithm consists of: a) a matchmaking part, where the functional parameters of the application are taken into account to rule out inconsistent offerings, and b) a ranking part, where the non-functional parameters of the application are considered to score and rank offerings. źhe algorithm is extensively evaluated showing linear scalability to the number of offerings and application requirements. Furthermore, it is extensible upon future semantic model extensions, because it is agnostic to domain specific concepts and parameters, using SPARQL template queries.

The remarkable advances in biotechnology and health sciences have led to a significant production of data, such as high throughput genetic data and clinical information, generated from large Electronic Health Records (EHRs). To this end, application of machine learning and data mining methods in biosciences is presently, more than ever before, vital and indispensable in efforts to transform intelligently all available information into valuable knowledge. Diabetes mellitus (DM) is defined as a group of metabolic disorders exerting significant pressure on human health worldwide. Extensive research in all aspects of diabetes (diagnosis, etiopathophysiology, therapy, etc.) has led to the generation of huge amounts of data. The aim of the present study is to conduct a systematic review of the applications of machine learning, data mining techniques and tools in the field of diabetes research with respect to a) Prediction and Diagnosis, b) Diabetic Complications, c) Genetic Background and Environment, and e) Health Care and Management with the first category appearing to be the most popular. A wide range of machine learning algorithms were employed. In general, 85% of those used were characterized by supervised learning approaches and 15% by unsupervised ones, and more specifically, association rules. Support vector machines (SVM) arise as the most successful and widely used algorithm. Concerning the type of data, clinical datasets were mainly used. The title applications in the selected articles project the usefulness of extracting valuable knowledge leading to new hypotheses targeting deeper understanding and further investigation in DM. Keywords: Machine learning, Data mining, Diabetes mellitus, Diabetic complications, Disease prediction models, Biomarker(s) identification

Purpose: We sought to investigate whether B cell receptor immunoglobulin (BcR IG) stereotypy is associated with particular clinicobiological features among chronic lymphocytic leukemia (CLL) patients expressing mutated BcR IG (M-CLL) encoded by the IGHV4-34 gene, and also ascertain whether these associations could refine prognostication. Experimental Design: In a series of 19,907 CLL cases with available immunogenetic information, we identified 339 IGHV4-34–expressing cases assigned to one of the four largest stereotyped M-CLL subsets, namely subsets #4, #16, #29 and #201, and investigated in detail their clinicobiological characteristics and disease outcomes. Results: We identified shared and subset-specific patterns of somatic hypermutation (SHM) among patients assigned to these subsets. The greatest similarity was observed between subsets #4 and #16, both including IgG-switched cases (IgG-CLL). In contrast, the least similarity was detected between subsets #16 and #201, the latter concerning IgM/D-expressing CLL. Significant differences between subsets also involved disease stage at diagnosis and the presence of specific genomic aberrations. IgG subsets #4 and #16 emerged as particularly indolent with a significantly (P < 0.05) longer time-to-first-treatment (TTFT; median TTFT: not yet reached) compared with the IgM/D subsets #29 and #201 (median TTFT: 11 and 12 years, respectively). Conclusions: Our findings support the notion that BcR IG stereotypy further refines prognostication in CLL, superseding the immunogenetic distinction based solely on SHM load. In addition, the observed distinct genetic aberration landscapes and clinical heterogeneity suggest that not all M-CLL cases are equal, prompting further research into the underlying biological background with the ultimate aim of tailored patient management. Clin Cancer Res; 23(17); 5292–301. ©2017 AACR.

This work aims to resolve issues related to Web Service retrieval, also known as Service Selection, Discovery or essentially Matching, in two directions. Firstly, a novel matching algorithm for SAWSDL is introduced. The algorithm is hybrid in nature, combining novel and known concepts, such as a logic-based strategy and syntactic text-similarity measures on semantic annotations and textual descriptions. A plugin for the S3 contest environment was developed, in order to position Tomaco amongst state-of-the-art in an objective, reproducible manner. Evaluation showed that Tomaco ranks high amongst state of the art, especially for early recall levels. Secondly, this work introduces the Tomaco web application, which aims to accelerate the wide-spread adoption of Semantic Web Service technologies and algorithms while targeting the lack of user-friendly applications in this field. Tomaco integrates a variety of configurable matching algorithms proposed in this paper. It, finally, allows discovery of both existing and user-contributed service collections and ontologies, serving also as a service registry., Comment: Under review. Keywords: Web Services Discovery, Intelligent Web Services and Semantic Web, Internet reasoning services, Web-based services

Evidence-Based Medicine (EBM) has been an important practice for medical practitioners. However, as the number of medical publications increases dramatically, it is becoming extremely difficult for medical experts to review all the contents available and make an informative treatment plan for their patients. A variety of frameworks, including the PICO framework which is named after its elements (Population, Intervention, Comparison, Outcome), have been developed to enable fine-grained searches, as the first step to faster decision making. In this work, we propose a novel entity recognition system that identifies PICO entities within medical publications and achieves state-of-the-art performance in the task. This is achieved by the combination of four 2D Convolutional Neural Networks (CNNs) for character feature extraction, and a Highway Residual connection to facilitate deep Neural Network architectures. We further introduce a PICO Statement classifier, that identifies sentences that not only contain all PICO entities but also answer questions stated in PICO. To facilitate this task we also introduce a high quality dataset, manually annotated by medical practitioners. With the combination of our proposed PICO Entity Recognizer and PICO Statement classifier we aim to advance EBM and enable its faster and more accurate practice.

The widespread use of Social Media creates great opportunities for businesses to take advantage of. By combining clever techniques, companies can develop powerful data analysis systems to understand their customers. This paper presents a real time sentiment analysis and location inference system, showcased via AirSent, an R-based application designed to assist airline carriers in measuring their passengers' satisfaction. AirSent can download, classify and locate tweets within seconds, presenting the results in interactive maps.

This paper presents a novel real-world application for energy savings in a Smart Building environment. The proposed system unifies heterogeneous wireless sensor networks under a Semantic Web Service middleware. Two complementary and mutually exclusive rule-based approaches for enforcing energy-saving policies are proposed: a reactive agent based on production rules and a deliberative agent based on defeasible logic. The system was deployed at a Greek University, showing promising experimental results (at least 4% daily savings). Although the percentage of energy savings may seem low, the greatest merit of the method is ensuring that no energy is wasted by constantly enforcing the policies.

Dynamic (also known as instance-based) ensemble pruning selects a (potentially) different subset of models from an ensemble during prediction based on the given unknown instance with the goal of maximizing prediction accuracy. This paper models dynamic ensemble pruning as a multi-label classification task, by considering the members of the ensemble as labels. Multi-label training examples are constructed by evaluating whether ensemble members are accurate or not on the original training set via cross-validation. We show that classification accuracy is maximized when learning algorithms that optimize example-based precision are used in the multi-label classification task. Results comparing the proposed framework against state-of-the-art dynamic ensemble pruning approaches in a variety of datasets using a heterogeneous ensemble of 200 classifiers show that it leads to significantly improved accuracy.

The advent of high-throughput genomic technologies is enabling analyses on thousands or even millions of single-nucleotide polymorphisms (SNPs). At the same time, the selection of a minimum number of SNPs with the maximum information content is becoming increasingly problematic. Available locus ranking programs have been accused of providing upwardly biased results (concerning the predicted accuracy of the chosen set of markers for population assignment), cannot handle high-dimensional datasets, and some of them are computationally intensive. The toolbox for ranking and evaluation of SNPs (TRES) is a collection of algorithms built in a user-friendly and computationally efficient software that can manipulate and analyze datasets even in the order of millions of genotypes in a matter of seconds. It offers a variety of established methods for evaluating and ranking SNPs on user defined groups of populations and produces a set of predefined number of top ranked loci. Moreover, dataset manipulation algorithms enable users to convert datasets in different file formats, split the initial datasets into train and test sets, and finally create datasets containing only selected SNPs occurring from the SNP selection analysis for later on evaluation in dedicated software such as GENECLASS. This application can aid biologists to select loci with maximum power for optimization of cost-effective panels with applications related to e.g. species identification, wildlife management, and forensic problems. TRES is available for all operating systems at http://mlkd.csd.auth.gr/bio/tres.

In this short paper, we use machine learning methods to maximise the efficiency in the Hellenic Electricity Distribution Network Operator (HEDNO) S.A.'s project management by predicting which upcoming projects are going to be paid by the customer, thus enabling the company to fetch all needed materials in time, and to avoid unwanted delays in the operations of the company.

Biological systems ranging from simple ecology models to biochemical interactions in molecular biology, have profited profoundly by the advances and potentials of nanotechnology, especially those in lower dimensions. Molecular nanonetworks, the outcome of this adjacency, have been under great growth, the result of which was the establishment of molecular communication paradigm. Under this paradigm, we study the message dissemination dynamics in a biological communication system, by a model typical of electromagnetic networks. Applying a reaction-diffusion mathematical model of double diffusivity in a biological communication system, we compare the outcome with the simulation results. To this aim and based on a commercial tool, a set of efficient simulation scenarios, for the precise prediction of message dissemination dynamics were run, under the newly developed simulation framework. In particular, single and multiple sources of stimuli for different bacterial speeds and specimen geometries have been simulated. Results reveal that the message dissemination process in such bionanonetworks, can be successfully described by the reaction diffusion model, supporting the use of macro-scale simulation methods.

This paper deals with content-based large-scale image retrieval using the state-of-the-art framework of VLAD and Product Quantization proposed by Jegou as a starting point. Demonstrating an excellent accuracy-efficiency trade-off, this framework has attracted increased attention from the community and numerous extensions have been proposed. In this work, we make an in-depth analysis of the framework that aims at increasing our understanding of its different processing steps and boosting its overall performance. Our analysis involves the evaluation of numerous extensions (both existing and novel) as well as the study of the effects of several unexplored parameters. We specifically focus on: a) employing more efficient and discriminative local features; b) improving the quality of the aggregated representation; and c) optimizing the indexing scheme. Our thorough experimental evaluation provides new insights into extensions that consistently contribute, and others that do not, to performance improvement, and sheds light onto the effects of previously unexplored parameters of the framework. As a result, we develop an enhanced framework that significantly outperforms the previous best reported accuracy results on standard benchmarks and is more efficient.

A key challenge in information theoretic feature selection is to estimate mutual information expressions that capture three desirable terms&mdash, the relevancy of a feature with the output, the redundancy and the complementarity between groups of features. The challenge becomes more pronounced in multi-target problems, where the output space is multi-dimensional. Our work presents an algorithm that captures these three desirable terms and is suitable for the well-known multi-target prediction settings of multi-label/dimensional classification and multivariate regression. We achieve this by combining two ideas&mdash, deriving low-order information theoretic approximations for the input space and using quantization algorithms for deriving low-dimensional approximations of the output space. Under the above framework we derive a novel criterion, Group-JMI-Rand, which captures various high-order target interactions. In an extensive experimental study we showed that our suggested criterion achieves competing performance against various other information theoretic feature selection criteria suggested in the literature.

This article introduces a teacher–student framework for reinforcement learning, synthesising and extending material that appeared in conference proceedings [Torrey, L., & Taylor, M. E. (2013)]. Teaching on a budget: Agents advising agents in reinforcement learning. {Proceedings of the international conference on autonomous agents and multiagent systems}] and in a non-archival workshop paper [Carboni, N., &Taylor, M. E. (2013, May)]. Preliminary results for 1 vs. 1 tactics in StarCraft. {Proceedings of the adaptive and learning agents workshop (at AAMAS-13)}]. In this framework, a teacher agent instructs a student agent by suggesting actions the student should take as it learns. However, the teacher may only give such advice a limited number of times. We present several novel algorithms that teachers can use to budget their advice effectively, and we evaluate them in two complex video games: StarCraft and Pac-Man. Our results show that the same amount of advice, given at different moments, can have differe...

This work presents a Web Service Middleware infrastructure for Ambient Intelligence environments, named aWESoME. aWESoME is a vital part of the Smart IHU project, a large-scale Smart University deployment. The purpose of the proposed middleware within the project is twofold: for one, to ensure universal, homogeneous access to the system's functions and secondly, to fulfill functional and non-functional requirements of the system. Namely, the infrastructure itself should consume significantly low power (as it is meant for energy savings in addition to automations), without compromising reliability and fast response time. The infrastructure should enable fast and direct discovery, invocation and execution of services. Finally, on hardware level, the wireless sensor and actuator network should be optimally configured for speed and reliability as well. The proposed solution employs widely used web open standards for description and discovery to expose hardware and software functions and ensure interoperability, even outside the borders of this university deployment. It proposes a straightforward method to integrate low-cost and resource-constrained heterogeneous devices found in the market and a large-scale placement of servers and wireless sensor networks. Different server hardware installations have been evaluated to find the optimum trade-off between response time and power consumption. Finally, a range of client applications that exploit the middleware on different platforms are demonstrated, to prove its usability and effectiveness in enabling, in this scenario, energy monitoring and savings.

The main objective of transfer learning is to reuse knowledge acquired in a previous learned task, in order to enhance the learning procedure in a new and more complex task. Transfer learning comprises a suitable solution for speeding up the learning procedure in Reinforcement Learning tasks. This work proposes a novel method for transferring models to Reinforcement Learning agents. The models of the transition and reward functions of a source task, will be transferred to a relevant but different target task. The learning algorithm of the target task's agent takes a hybrid approach, implementing both model-free and model-based learning, in order to fully exploit the presence of a source task model. Moreover, a novel method is proposed for transferring models of potential-based reward shaping functions. The empirical evaluation, of the proposed approaches, demonstrated significant results and performance improvements in the 3D Mountain Car and Server Job Scheduling tasks, by successfully using the models generated from their corresponding source tasks.