DCAART 2014 Abstracts


Short Papers
Paper Nr: 4
Title:

Novel Feature Selection Methods for High Dimensional Data

Authors:

Verónica Bolón-Canedo, Noelia Sánchez-Maroño and Amparo Alonso-Betanzos

Abstract: Over the last few years, the dimensionality of datasets involved in data mining applications has increased dramatically. In this situation, feature selection becomes indispensable as it allows for dimensionality reduction and relevance detection. This paper is devoted to study the impact of feature selection on high-dimensonal data as well as to present novel methods. After demonstrating the adequacy of feature selection on real applications, new methods are described which cover different topics such as ensemble learning, distributed learning, scalability of algorithms or cost-based feature selection.

Paper Nr: 7
Title:

Households’ Behaviors and Systemic Financial Instability - Experimental Insights and Agent-based Simulations for Macroeconomic Policy Analyses

Authors:

Paola D'Orazio

Abstract: Economics profession is currently engaged in a debate on which are the best methodological tools in order to study the dynamics of the real economy and adequately address important policy issues and social concerns. The present paper suggests the development of an experimentally microfounded Agent-based model in order to cope with the complexity and instability of the macroeconomic environment. The focus of the paper is both on the micro and the macro level, i.e., agents and the environment in which they act and interact. For the micro level, I suggest to design an experiment in order to gain insights into agents’ behaviors. For the macro level, I plan to build an ABM where agents are estimated, rather than calibrated, by using experimental data.

Paper Nr: 8
Title:

Research on Techniques for Building Energy Model

Authors:

Dimitrios-Stavros Kapetanakis, Eleni Mangina and Donal Finn

Abstract: Forecasting of building thermal and cooling loads, without the use of simulation software, can be achieved using data from Building Energy Management Systems (BEMS). Experience in building modelling has shown that data analysis is a key factor in order to produce accurate results. Commercial buildings incorporate BEMS to control the Heating Ventilation and Air-Conditioning (HVAC) system and to monitor the indoor environment conditions. Measurements of temperature, humidity and energy consumption are typically stored within BEMS. These measurements include underlying information regarding buildings thermal response. This project focuses on a novel approach for cost-effective modelling of actual data from commercial buildings, with models that can be assembled rapidly and deployed easily. This approach will constitute a practical research testbed to optimise multiple objectives related to the buildings’ energy modelling research area: i) development of a novel approach for predicting thermal and cooling loads of commercial buildings; ii) highly accurate predictions in terms of thermal and cooling loads; iii) scalability of the new approach to any commercial building and iv) minimum commissioning and maintenance effort requirements.

Paper Nr: 9
Title:

Overview of the PhD Project: Agile Control Architecture for Reconfigurable Manufacturing Systems - Bringing Flexible Manufacturing to the Next Level

Authors:

Daniël Telgen, Erik Puik, Leo van Moergestel and John-Jules Meyer

Abstract: Agile Manufacturing is a paradigm that creates a shorter time to market in the manufacturing industry. In this project a hybrid architecture using Robot Operating Nodes and Multi Agent Systems is developed that is used to create flexible manufacturing systems where all systems are autonomous. In this system both products and manufacturing systems have an agent that represents and controls the hardware. The research problems focuses on several aspects of this project, including a quick overview of the hybrid architecture that has been developed and several problems that will need to be researched in the project. These problems include ways to automatically generate control instructions on the machines based on a products demand, safety aspects of reconfigurable machines and practical problems like scheduling. Related work is discussed to show that the current work is state of the art and that the agent approach for this research is valid. The impact of this project is to enable the use of agent technology in the manufacturing industry and to research & develop a proof of concept system that can be used for experimental research and test by industry.

Paper Nr: 10
Title:

Towards Attack Trees Synthesis for Computer-aided Risk Analysis

Authors:

Stéphanie Georges and Sophie Pinchinat

Abstract: Risk Analysis is a discipline consisting in identifying and evaluating risks that threaten a given system in order to reduce or annihilate them by defining actions to engage (risk management). We developed a method aiming at building attack scenarios against a system that has to be protected. We describe this method in the present article. The main principle is to extract from a dedicated model of the system the scenarios (expressed as a succession of elementary actions) allowing to reach, from an initial state, a given goal (expressed as a set of damages against which we want to provide countermeasures). Thanks to the use of high-level actions, these scenarios are then gathered into an attack tree allowing after specific processing to highlight flaws to bring down.

Paper Nr: 11
Title:

Leveraging Adaptive Sessions Based on Therapeutic Empathy Through a Virtual Agent

Authors:

Adrián Bresó, Juan Martínez-Miranda and Juan Miguel García-Gómez

Abstract: This Doctoral Consortium paper outlines the author's proposed research for the improvement of the Human-Agent Interaction (HAI) applied to the computerised based treatment of Major Depression. In patients with chronic conditions such as Major Depression, the non-adherence to the treatment is a serious problem that increases the economic and clinical burden, and worsens the clinical status of the patient. The research work presented extends the work done in the previous author’s MSc Thesis and promotes a better adherence to the treatment through two main contributions: (1) planning a flexible, personalized and adaptive daily interactive sessions, and (2) generating an adequate emotional response in a Virtual Agent used as the main system’s interface based on a modelling of Therapeutic Empathy.

Paper Nr: 12
Title:

Networks and Imitations in an Agent based Asset Market

Authors:

Souhir Masmoudi

Abstract: We propose an agent-based approach to analyze the influence of network structure and investors’ mimicking behavior on price dynamics and the market share of agents. We consider a model that involves two different forecasting strategies for investors: chartists and fundamentalists. Investors switch between the two strategies by either copying (a) the strategy used by the most profitable agent in her neighborhood (most profitable rule), or (b) the strategy with the highest average profitability in her neighborhood (average rule). Our results show that the most profitable rule exhibits greater volatility in terms of the fraction of agents using each strategy. This volatility is higher when (i) there are more random links given the size of local neighborhood, and (ii) the size of neighborhood is larger. Because the price volatility increases monotonically with an increase in the proportion of chartists, such volatility in the fraction of strategies used by agents leads to unstable prices.

Paper Nr: 13
Title:

A Network Dispersion Problem for Non-communicating Agents

Authors:

Martin N. Simanjuntak

Abstract: Alpern and Reyniers (2002) initiated the study of spatial dispersion of agents in the context of non-communicating agents who could move between any two locations, based only on the knowledge of the populations at all locations. The research problem addressed in this paper is to see how the problem of Alpern and Reyniers changes if a network (graph) structure is imposed on the set of locations: An agent only knows the population at his current node (location) and he can only stay still or move to an adjacent node. He can see the number of arcs at his node, but if he moves he must choose among them equiprobably. As we are dealing with non-communicating agents with limited amount of global knowledge, we shall limit our discussion to simple and myopic strategies. In particular, our agents adopt common (Markovian) stunted random walk strategy, completely specified by the probability p of staying still for another period at the same node. Given such strategies, how long will it take, on average, for the agents to achieve dispersion? Our objective is to determine the minimum dispersal time and optimal levels of p for several classes of networks, including line networks, cycle networks, and complete graphs.