Accepted Papers

  • Music Mood Dataset Creation Based On Last.Fm Tags
    Erion Çano and Maurizio Morisio, Department of Control and Computer Engineering, Polytechnic University of Turin, Italy
    ABSTRACT
    Music emotion recognition today is based on techniques that require high quality and large emotionally labeled sets of songs to train algorithms. Manual and professional annotations of songs are costly and hardly accomplished. There is a high need for datasets that are public, highly polarized, large in size and following popular emotion representation models. In this paper we present the steps we followed to create two such datasets using intelligence of last.fm community tags. In the first dataset, songs are categorized based on an emotion space of four clusters we adopted from literature observations. The second dataset discriminates between positive and negative songs only. We also observed that last.fm mood tags are biased towards positive emotions. This imbalance of tags was reflected in cluster sizes of the resulting datasets we obtained; they contain more positive songs than negative ones..
  • Effective Vector Representations for Variable Length Symbol Sequences
    Gustavo Lado, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires
    ABSTRACT
    Machine learning techniques have demonstrated their versatility and have been successfully applied to a wide variety of problems, however, one of their major limitations is the treatment of sequential information. In general the input and output for these methods is expressed as fixed-dimension vectors, but in many problem domains, as in natural language processing, the information is represented by variable-length sequences. For most cases, it is possible to use some methods that convert these variable length sequences into fixed dimension vectors, however each of these methods has its own disadvantages. In this paper we propose an alternative to obtain vector representations of fixed dimension from symbol's sequences of variable length and their potential applications for natural language processing.
  • Clustering for Different Scales of Measurement - the Gap-Ratio Weighted K-means Algorithm
    Joris Guerin, Olivier Gibaru, Stephane Thiery, and Eric Nyiri, Laboratoire des Sciences de l'Information et des Systemes, France
    ABSTRACT
    This paper describes a method for clustering data that are spread out over large regions and which dimensions are on di erent scales of measurement. Such an algorithm was developed to implement a robotics application consisting in sorting and storing objects in an unsupervised way. The toy dataset used to validate such application consists of Lego bricks of di erent shapes and colors. The uncontrolled lighting conditions together with the use of RGB color features, respectively involve data with a large spread and di erent levels of measurement between data dimensions. To overcomethe combination of these two characteristics in the data, we have developed a new weighted K-means algorithm, called gap-ratio K-means, which consists in weighting each dimension of the feature space before running the K-means algorithm. The weight associated with a feature is proportional to the ratio of the biggest gap between two consecutive data points, and the average of all the other gaps. This method is compared with two other variants of K-means on the Lego bricks clustering problem as well as two other common classi cation datasets.
  • A Self-Organizing Recurrent Neural Network Based On Dynamic Analysis
    Qili Chen1 Junfei Qiao2 and Yi Ming Zou3, 1Beijing Information Science and Technology University, Beijing, China2Beijing University of Technology, Beijing, Chaina, 3 University of Wisconsin-Milwaukee, Milwaukee, USA
    ABSTRACT
    A recurrent neural network with a self-organizing structure based on the dynamic analysis of a task is presented in this paper. The stability of the recurrent neural network is guaranteed by design. A dynamic analysis method to sequence the subsystems of the recurrent neural network according to the fitness between the subsystems and the target system is developed. The network is trained with the network's structure self-organized by dynamically activating subsystems of the network according to tasks. The experiments showed the proposed network is capable of activating appropriate subsystems to approximate different nonlinear dynamic systems regardless of the inputs. When the network was applied to the problem of simultaneously soft measuring the chemical oxygen demand (COD) and NH3-N in wastewater treatment process, it showed its ability of avoiding the coupling influence of the two parameters and thus achieved a more desirable outcome.
  • Autonomous Generation Of Conflict-Free Examination Timetable Using Constraint Satisfaction Modelling
    Tarek Elsaka, University of Sharjah, Sharjah, UAE
    ABSTRACT
    Examination timetable (ETT) is a complex administrative task at educational institutions that must fulfil various constraints to generate the ETT to schedule exam sessions within a precise period. The ETT problem could be modelled as Constraint Satisfaction Problems (CSPs). In addition, it could be particularly investigated by Constraint Logic Programming (CLP) approach. This paper uses a real examination dataset from the Community College (CC), University of Sharjah (UoS). This dataset contains very rich data with many practical constraints to be satisfied such as a course taught at many campuses must has the same exam date and an invigilator can invigilate at any campus. This paper applies the CSP definitions as well as the Optimization Programming Language (OPL) to model the ETT dataset and automatically generate a conflict-free ETT solution using a CLP Solver. Finally, it uses the results to satisfy the proposed constraints in the model.
  • A Biologically Inspired Reinforcement Learning Based Algorithm For Autonomous Exploration
    Amir Ramezani1, Byambaa Dorj1, Abdur Razzaq Fayjie1, Oualid Doukhi1 and Deok Jin Lee1, Kunsan National University, Gunsan, Republic of Korea
    ABSTRACT
    Ability of learning is a crucial aspect of intelligence where when it is autonomous, machine is able to interact with its environment and learn from it. As an intelligent entity, this kind of machine can learn from their own experiences as a self-learner and use it to adapt themselves with the unknown environment. In this paper, we discuss about the algorithm of a self-learner robot that is able to interact with its unknown environment and learn from its own experiences to explore and avoid obstacles in an autonomous and adaptive way. In order to implement our idea, we have used a deep reinforcement learning method in order to design an autonomous end-to-end exploration and obstacle avoidance method useful for ground and aerial robots. To make our algorithm robust, we have used a sensor-fusion method in order to measure disparity map related to robot safely. Our algorithm is tested in a simulated ground robot in Gazebo and where our result shows that the robot is able to explore autonomously and in an adaptive and autonomous way while time goes to infinity.