Despite the rapid expansion of smart grids and large volumes of data at the individual consumer level, there are still various cases where adequate data collection to train accurate load forecasting models is challenging or even impossible. This paper proposes adapting an established Model-Agnostic Meta-Learning algorithm for short-term load forecasting in the context of few-shot learning. Specifically, the proposed method can rapidly adapt and generalize within any unknown load time series of arbitrary length using only minimal training samples. In this context, the meta-learning model learns an optimal set of initial parameters for a base-level learner recurrent neural network. The proposed model is evaluated using a dataset of historical load consumption data from real-world consumers. Despite the examined load series’ short length, it produces accurate forecasts outperforming transfer learning and task-specific machine learning methods by 12.5%. To enhance robustness and fairness during model evaluation, a novel metric, mean average log percentage error, is proposed that alleviates the bias introduced by the commonly used MAPE metric. Finally, a series of studies to evaluate the model’s robustness under different hyperparameters and time series lengths is also conducted, demonstrating that the proposed approach consistently outperforms all other models.

This paper presents the outcomes of the project titled: ‘’Feasibility Study for a Black Sea SDGs Observatory’’, financed by the Black Sea Project Promotion Facility. The project was assigned to the Aristotle University of Thessaloniki and the Financial University of Russia. The Observatory will constitute one of the main divisions of the regional UN SDSN (United Nations Sustainable Development Solutions Network) Black Sea. The paper presents the need for establishing a Black Sea SDGs (Sustainable Development Goals) Observatory, the methodology followed, and the expected results of its creation.

Nowadays, the fashion clothing industry is moving towards "fast" fashion, offering a wide variety of products based on different patterns and styles, usually characterized by lower costs and ambiguous quality. The retails markets are trying to present regularly new fashion collections while trying to follow the latest fashion trends at the same time. The main reason is to remain competitive and keep up with ever-changing customer demands. Fashion designers draw inspiration from social media, e-shops, and fashion shows that set the new fashion trends. In this direction, we propose Science4Fashion, an AI end-to-end system that facilitates fashion designers by collecting and analyzing data from many different sources and suggesting products according to their needs. An overview of the system's modules is presented, emphasizing data collection, data annotation using deep learning models, and product recommendation and user feedback processes. The experiments presented in this paper are twofold: (a) experiments regarding the evaluation of clothing attribute classification, and (b) experiments regarding product recommendation using the baseline kNN enriched by the frequency-based clustering algorithm (FBC), achieving promising results.

In the clothing industry, design, development, and procurement teams have been affected more than any other industry and are constantly under pressure to present more products with fewer resources in a shorter time. The diversity of garment designs created as new products is not found in any other industry and is almost independent of the size of the business. Science4Fashion is a semi-autonomous intelligent personal assistant for fashion product designers. Our system consists of an interactive environment where a user utilizes different modules responsible for a) data collection from online sources, b) knowledge extraction, c) clustering, and d) trend/product recommendation. This paper is focusing on two core modules of the implemented system. The Clustering Module combines various clustering algorithms and offers a consensus that arranges data in clusters. At the same time, the Product Recommender and Feedback module receives the designer’s input on different fashion products and recommends more relevant items based on their preferences. The experimental results highlight the usefulness and the efficiency of the proposed subsystems in aiding the creative fashion process.

Background Antigen receptors are characterized by an extreme diversity of specificities, which poses major computational and analytical challenges, particularly in the era of high-throughput immunoprofiling by next generation sequencing (NGS). The T cell Receptor/Immunoglobulin Profiler (TRIP) tool offers the opportunity for an in-depth analysis based on the processing of the output files of the IMGT/HighV-Quest tool, a standard in NGS immunoprofiling, through a number of interoperable modules. These provide detailed information about antigen receptor gene rearrangements, including variable (V), diversity (D) and joining (J) gene usage, CDR3 amino acid and nucleotide composition and clonality of both T cell receptors (TR) and B cell receptor immunoglobulins (BcR IG), and characteristics of the somatic hypermutation within the BcR IG genes. TRIP is a web application implemented in R shiny. Results Two sets of experiments have been performed in order to evaluate the efficiency and performance of the TRIP tool. The first used a number of synthetic datasets, ranging from 250k to 1M sequences, and established the linear response time of the tool (about 6 h for 1M sequences processed through the entire BcR IG data pipeline). The reproducibility of the tool was tested comparing the results produced by the main TRIP workflow with the results from a previous pipeline used on the Galaxy platform. As expected, no significant differences were noted between the two tools; although the preselection process seems to be stricter within the TRIP pipeline, about 0.1% more rearrangements were filtered out, with no impact on the final results. Conclusions TRIP is a software framework that provides analytical services on antigen receptor gene sequence data. It is accurate and contains functions for data wrangling, cleaning, analysis and visualization, enabling the user to build a pipeline tailored to their needs. TRIP is publicly available at https://bio.tools/TRIP_-_T-cell_Receptor_Immunoglobulin_Profiler .

In contemporary clothing industry, design, development and procurement teams are constantly asked to present more products with fewer resources in a shorter time. Thus, clothing companies that aim to remain competitive in today’s market have to deploy new Artificial Intelligence techniques aiming at the automation of their traditional procedures. In this direction, the presented approach utilizes a deep learning model to accurately classify fashion images. The predictions are intended to be used on a personalized recommendation system, that acts as an assistant for the fashion designers. Two well established architectures are studied, VGG and ResNet, as well as a variation of ResNet. The realized experiments include: (a) architecture comparison, (b) hyperparameter tuning and classification, and (c) transfer learning. Two fashion datasets are used for the model training and classification: DeepFashion (for training the model from scratch) and iMaterialist (used to evaluate the transferability of the produced model). The results show that the first set of experiments achieved 80.5% accuracy, whereas the pre-trained model used on the second dataset led to a decrease of 60% on training time, while attaining satisfying results. CCS Concepts: • Computing methodologies → Object recognition; Supervised learning by classification; Neural networks; • Applied computing → Consumer products.

Scalable big data analysis frameworks are of paramount importance in the modern web society, which is characterized by a huge number of resources, including electronic text documents. Document clustering is an important field in text mining and is commonly used for document organization, browsing, summarization and classification. Hierarchical clustering methods construct a hierarchy structure that, combined with the produced clusters, can be useful in managing documents, thus making the browsing and navigation process easier and quicker, and providing only relevant information to the users’ queries by leveraging the structure relationships. Nevertheless, the high computational cost and memory usage of baseline hierarchical clustering algorithms render them inappropriate for the vast number of documents that must be handled daily. In this paper, we propose a new scalable hierarchical clustering framework, which uses the frequency of the topics in the documents to overcome these limitations. Our work consists of a binary tree construction algorithm that creates a hierarchy of the documents using three metrics (Identity, Entropy, Bin Similarity), and a branch breaking algorithm which composes the final clusters by applying thresholds to each branch of the tree. The clustering algorithm is followed by a meta-clustering module which makes use of graph theory to obtain insights in the leaf clusters’ connections. The feature vectors representing each document derive from topic modeling. At the implementation level, the clustering method has been dockerized in order to facilitate its deployment on cloud computing infrastructures. Finally, the proposed framework is evaluated on several datasets of varying size and content, achieving significant reduction in both memory consumption and computational time over existing hierarchical clustering algorithms. The experiments also include performance testing on cloud resources using different setups and the results are promising.

Nowadays, the fashion industry is moving towards fast fashion, offering a large selection of garment products in a quicker and cheaper manner. To this end, the fashion designers are required to come up with a wide and diverse amount of fashion products in a short time frame. At the same time, the fashion retailers are oriented towards using technology, in order to design and provide products tailored to their consumers' needs, in sync with the newest fashion trends. In this paper, we propose an artificial intelligence system which operates as a personal assistant to a fashion product designer. The system's architecture and all its components are presented, with emphasis on the data collection and data clustering subsystems. In our use case scenario, datasets of garment products are retrieved from two different sources and are transformed into a specific format by making use of Natural Language Processes. The two datasets are clustered separately using different mixed-type clustering algorithms and comparative results are provided, highlighting the usefulness of the clustering procedure in the clothing product recommendation problem.

Hyper-personalization intends to maximize the opportunities a marketer has to tailor content that fits each and every customer's wants and needs. Naturally, gathering and analyzing more data is the key to those opportunities. This is were the "Conversation Web" comes in, which in the near future is expected to transform to so much more than just conversational interfaces (chat-bots). In a truly Conversation Web, websites and users implicitly "discuss" in the form of clicks, mouse scrolls and movements, as well as page views and product purchases. Websites use this information for decoding user interests and profile and provide customized one-to-one services. In this work we proposed an integrated architecture for the conversational Web; consequently we propose a novel hybrid approach for recommendations using offline and online analysis, as well as we propose a novel personalized search strategy that takes into account the strict time performance limitations applied in e-commerce. We evaluate the proposed methods on three different datasets and we show that our personalized search approach provides considerably improvements in search results while being suitable for near realtime search in commercial environments. Regarding personalized recommendations, the proposed approach outperforms current state-of-art methods in small-medium datasets and improves performance in large datasets when combined with other methods.

Nowadays, a wide range of sciences are moving towards the Big Data era, producing large volumes of data that require processing for new knowledge extraction. Scientific workflows are often the key tools for solving problems characterized by computational complexity and data diversity, whereas cloud computing can effectively facilitate their efficient execution. In this paper, we present a generative big data analysis workflow that can provide analytics, clustering, prediction and visualization services to datasets coming from various scientific fields, by transforming input data into strings. The workflow consists of novel algorithms for data processing and relationship discovery, that are scalable and suitable for cloud infrastructures. Domain experts can interact with the workflow components, set their parameters, run personalized pipelines and have support for decision-making processes. As case studies in this paper, two datasets consisting of (i) Documents and (ii) Gene sequence data are used, showing promising results in terms of efficiency and performance.

The identification of meaningful groups of proteins has always been a major area of interest for structural and functional genomics. Successful protein clustering can lead to significant insight, assisting in both tracing the evolutionary history of the respective molecules as well as in identifying potential functions and interactions of novel sequences. Here we propose a clustering algorithm for same-length sequences, which allows the construction of subset hierarchy and facilitates the identification of the underlying patterns for any given subset. The proposed method utilizes the metrics of sequence identity and amino-acid similarity simultaneously as direct measures. The algorithm was applied on a real-world dataset consisting of clonotypic immunoglobulin (IG) sequences from Chronic lymphocytic leukemia (CLL) patients, showing promising results.

In the recent years, we have witnessed the rapid adoption of social media platforms, such as Twitter, Facebook and YouTube, and their use as part of the everyday life of billions of people worldwide. Given the habit of people to use these platforms to share thoughts, daily activities and experiences it is not surprising that the amount of user generated content has reached unprecedented levels, with a substantial part of that content being related to real-world events, i.e. actions or occurrences taking place at a certain time and location. Given the key role of events in our life, the task of annotating and organizing social media content around them is of crucial importance for ensuring real-time and future access to multimedia content about an event of interest. In this chapter, we present several research efforts from recent years that tackle two main problems: a) event detection and b) event-based media retrieval and summarization. Given archived collections or live streams of social media items, the purpose of event detection methods is to identify previously unknown events in the form of sets of items that describe them. In general, the events could be of any type, but there are also approaches aiming at events of specific type. Given a target event the goal of event summarization is first to identify relevant content and then to represent it in a concise way, selecting the most appealing and representative content.

Recent developments in deep reinforcement learning have enabled the creation of agents for solving a large variety of games given a visual input. These methods have been proven successful for 2D games, like the Atari games, or for simple tasks, like navigating in mazes. It is still an open question, how to address more complex environments, in which the reward is sparse and the state space is huge. In this paper we propose a divide and conquer deep reinforcement learning solution and we test our agent in the first person shooter (FPS) game of Doom. Our work is based on previous works in deep reinforcement learning and in Doom agents. We also present how our agent is able to perform better in unknown environments compared to a state of the art reinforcement learning algorithm.

In this paper we redefine the concept of Conversation Web in the context of hyper-personalization. We argue that hyper-personalization in the WWW is only possible within a conversational web where websites and users continuously“discuss” (interact in any way). We present a modular system architecture for the conversational WWW, given that adapting to various user profiles and multivariate websites in terms of size and user traffic is necessary, especially in e-commerce. Obviously there cannot be a unique fit-to-all algorithm, but numerous complementary personalization algorithms and techniques are needed. In this context, we propose PRCW, a novel hybrid approach combining offline and online recommendations using RFMG, an extension of RFM modeling. We evaluate our approach against the results of a deep neural network in two datasets coming from different online retailers. Our evaluation indicates that a) the proposed approach outperforms current state-of-art methods in small-medium datasets and can improve performance in large datasets when combined with other methods, b) results can greatly vary in different datasets, depending on size and characteristics, thus locating the proper method for each dataset can be a rather complex task, and c) offline algorithms should be combined with online methods in order to get optimal results since offline algorithms tend to offer better performance but online algorithms are necessary for exploiting new users and trends that turn up.

An important problem in the live music industry is finding venues that help expose artists to wider audiences. However, it is often difficult to obtain live music audience data to tackle this task. In this work, we investigate whether important venues can instead be inferred through social media data. Our approach consists of employing bipartite graph ranking algorithms to help discover important venues in artist-venue graphs mined from Facebook. We use both well-established algorithms, such as BiRank, and a modification of their common iterative scheme that avoids the impact of possibly erroneous heuristics to the ranking, which we call VenueRank. Resulting venue ranks are compared to those obtained from feature extraction for predicting the most listened artists and large listener increments in Spotify. This comparison yields high correlation between venue importance for listener prediction and bipartite graph ranking algorithms, with VenueRank found more robust against overfitting.

The paper describes a technical demonstration of an open-source framework for monitoring, analysis and search over multiple social media platforms. The framework is intended to be a valuable tool for media intelligence professionals, as well as a framework and testbed for scientists and developers with interest in social media research.

Contemporary electricity markets necessitate balancing supply and demand in real time, as well as the active involvement of all participating entities in the price-setting procedure. To this end, Demand Response pricing schemes are often introduced, encouraging customers to contribute to power system stability by reducing their load during peak periods. In recent years, various Time-of-Use strategies have been designed and applied in real life scenarios without yielding the expected results. Analysis of several unsuccessful test cases revealed that customers are not willing to alter their consumption habits and reduce their comfort by responding to price signals, unless they are presented with tangible benefits (cost reduction, better energy services, etc.). In this context, current work introduces an optimization methodology for individualized ToU pricing policies, building upon authors' previous work on small-scale consumer activity and response modelling. Using a multi-objective particle swarm optimization mechanism, the appropriate rates for each implemented pricing policy are identified, leading to consumers' cost and peak load reduction, in addition to higher acceptance rates. Experiments indicate the capabilities and effectiveness of the proposed approach on simulated datasets extracted from Power TAC competition platform, as well as real-life measurements acquired from a multi-residential building in Sweden.

Hermes introduces a new “describe once, run anywhere” paradigm for the execution of bioinformatics workflows in hybrid cloud environments. It combines the traditional features of parallelization-enabled workflow management systems and of distributed computing platforms in a container-based approach. It offers seamless deployment, overcoming the burden of setting up and configuring the software and network requirements. Most importantly, Hermes fosters the reproducibility of scientific workflows by supporting standardization of the software execution environment, thus leading to consistent scientific workflow results and accelerating scientific output.

Nowadays, the construction of a complex robotic system requires a high level of specialization in a large number of diverse scientific areas. It is reasonable that a single researcher cannot create from scratch the entirety of this system, as it is impossible for him to have the necessary skills in the necessary fields. This obstacle is being surpassed with the existent robotic frameworks. This paper tries to give an extensive review of the most famous robotic frameworks and middleware, as well as to provide the means to effortlessly compare them. Additionally, we try to investigate the differences between the definitions of a robotic framework, a robotic middleware and a robotic architecture.

In recent years, the liberation of distribution and energy services has led towards competitive Energy Market environments. In these Markets, the participating suppliers need to provide more reliable services, specifically tailored to each customer or group of customers with similar needs. Thus, it is important to identify the consumer types in their portfolio, through Customer Load Profiling. In this paper, algorithms that provide robust and reliable clustering results are examined, as tools for meaningful Low Voltage consumers' segmentation. A number of experiments on two different data sets were implemented to provide an insight to the proposed attributes' input selection. Additionally, a set of custom pricing schemes was produced based on the segmentation results and the peak and cost reduction are estimated. The results are promising and enhance our understanding of the long-term gains that can be obtained by well-defined customer segmentation and the design of individualized pricing schemes.