The very particular characteristics of electricity markets, require deep studies of the interactions between the involved players. MASCEM is a market simulator developed to allow studying electricity market negotiations. This paper presents a new proposal for the definition of MASCEM players' strategies to negotiate in the market. The proposed methodology is implemented as a multiagent system, using reinforcement learning algorithms to provide players with the capabilities to perceive the changes in the environment, while adapting their bids formulation according to their needs, using a set of different techniques that are at their disposal. This paper also presents a methodology to define players' models based on the historic of their past actions, interpreting how their choices are affected by past experience, and competition.
This paper proposes the design, development and evaluation of a hybrid video recommendation system. The proposed hybrid video recommendation system is based on a graph algorithm called Adsorption. Adsorption is a collaborative filtering algorithm in which relations between users are used to make recommendations. In this paper, Adsorption algorithm is enriched by content based filtering to provide better suggestions. Thus, collaborative recommendations are empowered considering item similarities. Therefore, the developed hybrid system combines both collaborative and content based approaches to produce more effective suggestions.
This paper describes the use of a support vector machine algorithm for the classification of molecules database in order for the prediction of the activity of drugs. Molecules database are fragmented, and each molecule is represented by a set of contained fragments. Molecular weighted descriptors are tested for the representation of molecular fragments in order to represent the dataset as a MxF array where each element takes the value of the molecular weighted descriptor calculated for the fragment. As weighted descriptors take into account distances and heteroatoms present in the fragments, the representation space allows the discrimination of similar structural fragments. A Support Vector Machine algorithm is used for the classification process for a training set. Prediction of the activity of the test set is carried out in function of results of training stage and the application of a proposed heuristic. Results obtained shows that the use of weighted molecular descriptors improves the prediction of drug activity for heterogeneous datasets.
The Automated Negotiation Agents Competition (ANAC-10) was held and our agent won the tournament. Our agent estimates the alternatives the opponent will offer based on the history of the opponent’s offers. In addition, our agent tries to compromise to the estimated maximum utility of the opponent by the end of the negotiation. Also, we modify the basic strategy to not exceed the limit of compromise when the opponent is uncooperative. We can adjust the speed of compromise depending on the negotiation. We introduce the new ratio(t) to control our agent’s actions at the final phase. With this improvement, our agents try to reach agreement when the opponent’s proposal is closer to our estimated maximum values. The main reason our agent outperforms the others is its ability to reach a last-minute agreement as often as possible.
This paper describes a novel meta-learning (MTL) based methodology used to optimize a neural network based inference system. The inference system being optimized is part of a bioinformatic application built to implement a systematic search scheme for the identification of genes which encode enzymes of metabolic pathways. Different MTL implementations are contrasted with manually optimized inference systems. The MTL based approach was found to be flexible and able to produce better results than manual optimization.
The fastest Learning Automata (LA) algorithms currently available come from the family of estimator algorithms. The Pursuit algorithm (PST), a pioneering scheme in the estimator family, obtains its superior learning speed by using Maximum Likelihood (ML) estimates to pursue the action currently perceived as being optimal. Recently, a Bayesian LA (BLA) was introduced, and empirical results that demonstrated its advantages over established top performers, including the PST scheme, were reported. The BLA scheme is inherently Bayesian in nature, but it succeeds in avoiding the computational intractability by merely relying on updating the hyper-parameters of sibling conjugate priors, and on random sampling from the resulting posteriors. In this paper, we integrate the foundational learning principles motivating the design of the BLA, with the principles of the PST. By doing this, we have succeeded in obtaining a completely novel, and rather pioneering, approach to solving LA-like problems, namely, by designing the Bayesian Pursuit algorithm (BPST). As in the BLA, the estimates are truly Bayesian (as opposed to ML) in nature. However, the action selection probability vector of the PST is used for its exploration purposes. Also, unlike the ML estimate, which is usually a single value, the use of a posterior distribution permits us to choose any one of a spectrum of values in the posterior, as the appropriate estimate. Thus, in this paper, we have chosen a 95% percentile value of the posterior (instead of the mean) to pursue the most promising actions. Further, as advocated in [7], the pursuit has been done using both the Linear Reward-Penalty and Reward-Inaction philosophies, leading to the corresponding BPSTRP and BPSTRI schemes respectively. It turns out that the BPST is superior to the PST, with the BPSTRI being even more robust than the BPSTRP. Moreover, by controlling the learning speed of the BPST, the BPST schemes perform either better or comparable to the BLA. We thus believe that the BPST constitutes a new avenue of research, in which the performance benefits of the PST and the BLA are mutually augmented, opening up for improved performance in a number of applications, currently being tested.
Maximizing occupant comfort and minimizing energy costs are two challenging tasks in the efficient operation of any office building. Often these objectives cannot be achieved simultaneously which asks for methods that resolve this trade-off in the best way. Several approaches deal with this problem by focusing on optimizing one of the above criteria while keeping the other one within an acceptable range. However, defining the latter can be very difficult in practice. In particular, setting the acceptable comfort range for offices shared by multiple occupants with conflicting thermal comfort preferences is a challenging problem. This paper presents an intelligent decision support system that assists a building operator in resolving the trade-off between energy efficiency and occupant comfort. Its key component is a case-based reasoning algorithm for predicting the distribution of the occupants’ thermal preferences.
Good project planning provides the basis for successful offshore well drilling projects. In this domain, planning occurs in two phases: an onshore phase develops a project plan; and an offshore phase implements the plan and tracks progress. The Performance Tracker applies a case-based reasoning approach to support the reuse of project plans. Cases comprise problem parts that store project initiation data, and solution parts that record the tasks and subtasks of actual plans. An initial evaluation shows that nearest neighbour retrieval identifies projects in which the retrieved tasks and subtasks are relevant for the new project. The Performance Tracker can be viewed as a recommender system in which recommendations are plans. Thus the data that is routinely captured as part of the performance tracking during offshore implementation is utilised as experiences.
In this paper an agent-based analysis is made of patterns in crowd behaviour, in particular to simulate a real-life incident that took place on May 4, 2010 in Amsterdam. As a basis, an existing agent-based model is used for contagion of emotions, beliefs and intentions. From available video material and witness reports, useful empirical data were extracted. Similar patterns were achieved in simulations, whereby some of the parameters of the model were tuned to the case addressed, and most parameters were assigned default values. The results show the inclusion of contagion of belief, emotion, and intention states of agents results in better reproduction of the incident than non-inclusion.
A fault diagnosis framework for electrical power transmission networks, which combines Hybrid Bayesian Networks (HBN) and Wavelets is proposed. HBN is a probabilistic graphical model in which discrete and continuous data are analyzed. In this work, power network's protection breakers are modeled as discrete nodes, and information extracted from voltages measured in every electrical network node represent the continuous nodes. Protection breakers are devices with the function to isolate faulty nodes by opening the circuit, and are considered to be working in one of three states: OK, OPEN, and FAIL. On the other hand, node voltages data are processed with wavelets, delivering specific coefficients patterns which are encoded into probability distributions of continuous HBN nodes. Experimental tests show a good performance of the diagnostic system when simultaneous multiple faults are simulated in a 24 nodes electrical network, in comparison with a previous approach in the same domain.
We present hardware testing of an evolutionary algorithm known as biogeography-based optimization (BBO) and extend it to distributed learning. BBO is an evolutionary algorithm based on the theory of biogeography, which describes how nature geographically distributes organisms. We introduce a new BBO algorithm that does not use a centralized computer, and which we call distributed BBO. BBO and distributed BBO have been developed by mimicking nature to obtain an algorithm that optimizes solutions for different situations and problems. We use fourteen common benchmark functions to obtain results from BBO and distributed BBO, and we also use both algorithms to optimize robot control algorithms. We present not only simulation results, but also experimental results using BBO to optimize the control algorithms of mobile robots. The results show that centralized BBO generally gives better optimization results and would generally be a better choice than any of the newly proposed forms of distributed BBO. However, distributed BBO allows the user to find a less optimal solution to a problem while avoiding the need for centralized, coordinated control.
In this paper, we focus on protein contact map prediction, one of the most important intermediate steps of the protein folding problem. We describe a method where contact maps of proteins are predicted with decision trees, using as input codings the information obtained from all possible pairs of amino acids that were formed in the training data set. As a result, the algorithm creates a model that consists of 400 decision trees (one for each possible amino acids pair), which takes into account the amino acids frequency in the subsequence existent between the couple of amino acids analyzed. In order to evaluate the method generalization capabilities, we carry out an experiment using 173 non-homologous proteins of known structures, selected from the protein databank (PBD). Our results indicate that the method can assign protein contacts with an average accuracy of 0.34, superior to the 0.25 obtained by the FNETCSS method. This shows that our algorithm improves the accuracy with respect to the methods compared, especially with the increase of protein length.
In this paper, we present an architecture and sample implementation of a system which allows us to push latest up-to-date related contents to any Web news article or page in real-time. The architecture makes use of page Agents which recommend the contents and are persistent as well as synchronized over all page instances in browsers. The Agents are easy to incorporate on any Web page and make use of state-of-the-art Web technology. In our sample implementation, we show how our Agents, coupled with a Complementary Naive Bayes classifier, can recommend latest contents related to 47 Japanese prefectures and over 1700 Japanese cities. We show performance results and conclude on further research to improve the affiliate and user experience on the Web.
Negotiation support systems (NSSs) aim to assist people during the complex process of negotiation.We argue that having a shared mental model of the negotiation task enables and enhances the collaboration between the human negotiator and the NSS. This paper presents an analysis of negotiation that results in a set of concepts that a shared mental model of the user and the NSS should contain. Discrepancies between the individual mental models can arise for various reasons, such as the constructive nature of preferences. Explanation can increase user understanding of the NSS's reasoning, allowing the user to detect and resolve discrepancies. We therefore propose using explanation to achieve and maintain sharedness. We present a framework that provides a means to generate content for such explanations, where we focus on the mental models of user and opponent preferences.
Data discretization is an important task for certain types of data mining algorithms such as association rule discovery and Bayesian learning. For those algorithms, proper discretization not only can significantly improve the quality and understandability of discovered knowledge, but also can reduce the running time. We present a Global Unsupervised Discretization Algorithm based on Collective Correlation Coefficient (GUDA-CCC) that provides the following attractive merits. 1) It does not require class labels from training data. 2) It preserves the ranks of attribute importance in a data set and meanwhile minimizes the information loss measured by mean square error. The attribute importance is calibrated by the CCC derived from principal component analysis (PCA). The idea behind GUDA-CCC is that to stick closely to an original data set might be the best policy, especially when other available information is not reliable enough to be leveraged in the discretization. Experiments on benchmark data sets illustrate the effectiveness of the GUDA-CCC algorithm.
Our goal is to achieve a robot audition system that is capable of recognizing multiple environmental sounds and making use of them in human-robot interaction. The main problems in environmental sound recognition in robot audition are: (1) recognition under a large amount of background noise including the noise from the robot itself, and (2) the necessity of robust feature extraction against spectrum distortion due to separation of multiple sound sources. This paper presents the environmental recognition of two sound sources fired simultaneously using matching pursuit (MP) with the Gabor wavelet, which extracts salient audio features from a signal. The two environmental sounds come from different directions, and they are localized by multiple signal classification and, using their geometric information, separated by geometric source separation with the aid of measured head-related transfer functions. The experimental results show the noise-robustness of MP although the performance depends on the properties of the sound sources.
In this paper, we propose a greedy heuristic for the 2D rectangular packing problem that represents packings using a skyline. To solve the 2D rectangular strip packing problem, we make use of this heuristic as a subroutine in an iterative deepening binary search on the maximum height of the packing that incorporates a tabu search mechanic. Our approach outperforms all existing approaches on several sets of standard benchmark test cases for the 2D strip packing problem.
This paper describes aspects of a training environment for crisis decision makers who, notoriously, operate in highly stressful and unpredictable situations. Training such decision makers is the goal of Pandora-Box, a system which is able to teach a class of trainees representing different authorities that coordinate their interventions in critical situations. This paper dwells on the selection and modeling of the relevant human features that are shown to have an influence in decision making under crisis. The chosen features are used to create a trainee model on the basis of which the system adjusts the training exercises with the ultimate goal of maximizing the effectiveness of training. Trainees models are built by merging physiological and psychological data, and are represented by means of a timeline-based approach, a representation derived from planning technology. The infrastructure built for the trainee modeling constitutes the basis to assess the influence of specific variables (e.g., personality traits, self efficacy, stress and anxiety) on the performance of crisis managers during the training.
