IJCCI 2017 Abstracts


Area 1 - Cognitive and Hybrid Systems

Full Papers
Paper Nr: 7
Title:

Various Approaches to the Application of Answer Set Programming in Order-picking Systems with Intelligent Vehicles

Authors:

Steffen Schieweck, Gabriele Kern-Isberner and Michael ten Hompel

Abstract: Intelligent, self-driving vehicles on public roads are widely noted in the media nowadays. In warehouse and production systems, such vehicles have been common for a number of years, even though their intelligence has only come to awareness in the recent years. Those systems are especially tailored to work in volatile environments which change (to some degree) every few weeks or months. Thus, programming is required to be as flexible as possible while still providing high efficiency. Answer set programming is a well known paradigm which has received remarkable attention in the recent years. In this paper several approaches are presented and evaluated to apply answer set programming to an order-picking system with intelligent vehicles. The interconnected planning tasks are the dispatching of vehicles to driving jobs and the assignment of customer orders to picking stations.

Short Papers
Paper Nr: 59
Title:

Attractor Neural States: A Brain-Inspired Complementary Approach to Reinforcement Learning

Authors:

Oussama H. Hamid and Jochen Braun

Abstract: It is widely accepted that reinforcement learning (RL) mechanisms are optimal only if there is a predefined set of distinct states that are predictive of reward. This poses a cognitive challenge as to which events or combinations of events could potentially predict reward in a non-stationary environment. In addition, the computational discrepancy between two families of RL algorithms, model-free and model-based RL, creates a stability-plasticity dilemma, which in the case of interactive and competitive multiple brain systems poses a question of how to guide optimal decision-making control when there is competition between two systems implementing different types of RL methods. We argue that both computational and cognitive challenges can be met by infusing the RL framework as an algorithmic theory of human behavior with the strengths of the attractor framework at the level of neural implementation. Our position is supported by the hypothesis that ‘attractor states’ which are stable patterns of self-sustained and reverberating brain activity, are a manifestation of the collective dynamics of neuronal populations in the brain. Hence, when neuronal activity is described at an appropriate level of abstraction, simulations of spiking neuronal populations capture the collective dynamics of the network in response to recurrent interactions between these populations.

Posters
Paper Nr: 60
Title:

Challenging Recommendation Engines Evaluation Metrics and Mitigating Bias Problem of Information Cascades and Confirmation Biases

Authors:

Guillaume Blot, Francis Rousseaux and Pierre Saurel

Abstract: Digital knowledge gave birth to massive communication spaces, new access paths and new cleavages. Our experiment deals with the challenging issue of accessing this knowledge on the Internet. Computer scientists set up prediction algorithms and recommender engines. This way, knowledge access is partly automatized. Using a real-life dataset, our goal is to simulate the iterative behavior shift produced by most used recommender engines. On this basis, we show that in the context of recommendation, existing evaluation metrics are driven by prediction testing methods and we argue that ambiguity has to be raised between prediction and recommendation. Secondly, we propose alternative evaluation metrics for recommendation systems, targeting mitigating the bias problem of information cascades and confirmation biases.

Area 2 - Evolutionary Computation

Full Papers
Paper Nr: 11
Title:

Towards Real-Time Fleet-Event-Handling for the Dynamic Vehicle Routing Problem

Authors:

Simon Anderer, Max Halbich, Bernd Scheuermann and Sanaz Mostaghim

Abstract: This paper proposes an approach to real-time fleet event handling for the Dynamic Vehicle Routing Problem based on evolutionary computation. For this purpose, a communication protocol between a fleet of vehicles and an optimization back-end is presented and the related changes to the evolutionary algorithm are illustrated. This allows for information exchange and event-handling in real-time. Furthermore, this paper describes the adaption of benchmark files for the static Vehicle Routing Problem to a dynamic real-time scenario including time-dependent travel times, as well as dynamic travel and service times. The adapted benchmark files are then used for the evaluation of the proposed system.

Paper Nr: 22
Title:

The Behavior of Deep Statistical Comparison Approach for Different Criteria of Comparing Distributions

Authors:

Tome Eftimov, Peter Korošec and Barbara Koroušić Seljak

Abstract: Deep Statistical Comparison (DSC) is a recently proposed approach for the statistical comparison of meta-heuristic stochastic algorithms for single-objective optimization. The main contribution of the DSC is a ranking scheme, which is based on the whole distribution, instead of using only one statistic, such as average or median, which are commonly used. Contrary to common approach, the DSC gives more robust statistical results, which are not affected by outliers or misleading ranking scheme. The DSC ranking scheme uses a statistical test for comparing distributions in order to rank the algorithms. DSC was tested using the two-sample Kolmogorov-Smirnov (KS) test. However, distributions can be compared using different criteria, statistical tests. In this paper, we analyze the behavior of the DSC using two different criteria, the two-sample Kolmogorov-Smirnov (KS) test and the Anderson-Darling (AD) test. Experimental results from benchmark tests consisting of single-objective problems, show that both criteria behave similarly. However, when algorithms are compared on a single problem, it is better to use the AD test because it is more powerful and can better detect differences than the KS test when the distributions vary in shift only, in scale only, in symmetry only, or have the same mean and standard deviation but differ on the tail ends only. This influence is not emphasized when the approach is used for multiple-problem analysis.

Paper Nr: 36
Title:

Enhanced Symbolic Regression Through Local Variable Transformations

Authors:

Jirí Kubalík, Erik Derner and Robert Babuška

Abstract: Genetic programming (GP) is a technique widely used in a range of symbolic regression problems, in particular when there is no prior knowledge about the symbolic function sought. In this paper, we present a GP extension introducing a new concept of local transformed variables, based on a locally applied affine transformation of the original variables. This approach facilitates finding accurate parsimonious models. We have evaluated the proposed extension in the context of the Single Node Genetic Programming (SNGP) algorithm on synthetic as well as real-problem datasets. The results confirm our hypothesis that the transformed variables significantly improve the performance of the standard SNGP algorithm.

Paper Nr: 42
Title:

Self-learning Smart Cameras - Harnessing the Generalization Capability of XCS

Authors:

Anthony Stein, Stefan Rudolph, Sven Tomforde and Jörg Hähner

Abstract: In this paper, we show how an evolutionary rule-based machine learning technique can be applied to tackle the task of self-configuration of smart camera networks. More precisely, the Extended Classifier System (XCS) is utilized to learn a configuration strategy for the pan, tilt, and zoom of smart cameras. Thereby, we extend our previous approach, which is based on Q-Learning, by harnessing the generalization capability of Learning Classifier Systems (LCS), i.e. avoiding to separately approximate the quality of each possible (re-)configuration (action) in reaction to a certain situation (state). Instead, situations in which the same reconfiguration is adequate are grouped to one single rule. We demonstrate that our XCS-based approach outperforms the Q-learning method on the basis of empirical evaluations on scenarios of different severity.

Short Papers
Paper Nr: 9
Title:

Ant Colony Optimization Approaches for the Tree t-Spanner Problem

Authors:

Manisha Israni and Shyam Sundar

Abstract: A tree $t$-spanner of a given connected graph is a spanning tree $T$ in which the ratio of distance between every pair of vertices is at most $t$ times their distance in the graph, where $t$ is a parameter known as stretch factor of $T$. The tree $t$-spanner problem deals with finding a spanning tree in a connected graph whose stretch factor is minimum amongst all spanning trees of the graph. For unweighted graph, this problem is $\mathcal{NP}$-Hard for any fixed $t \geq 4$, whereas for weighted graph, this problem is $\mathcal{NP}$-Hard for any fixed ~ $t > 1$. This paper concerns this problem for connected, undirected, and weighted graph and proposes three variants of ant colony optimization (ACO) approach for this problem. ACO approach is a swarm intelligence technique inspired by the foraging behavior of real ants. All three variants of ACO approach have been tested on a set of randomly generated graph instances. Computational results show the effectiveness of all three variants of ACO approach.

Paper Nr: 26
Title:

Particle Stability in PSO under Stagnation Assumption

Authors:

Krzysztof Trojanowski and Tomasz Kulpa

Abstract: Convergence properties of a particle and a swarm decide about their performance. Particularly, one of these properties is a time, that is, a number of particle steps necessary to reach an equilibrium state. This is a subject of presented analysis. Generalized weak versions of measures: particle convergence expected time (pcet) and the particle location variance convergence time (pvct) as well as a new measure for evaluation of steps number necessary to reach equilibrium state, namely particle stability time, are proposed. For all the measures graphs of estimated and recorded values are presented.

Paper Nr: 28
Title:

Evolving Illumination Design Following Genetic Strategies

Authors:

Alice Plebe, Vincenzo Cutello and Mario Pavone

Abstract: Interior lighting design is a challenging task where are involved multiple constraints that need to be optimized for producing an accurate illumination avoiding possible glare. This paper, then, takes up the issue of providing a computational tool able to produce a proper lighting plan in interior spaces for a comfortable and optimal vision in all environments, taking also into account the energy consumption as little as possible. For finding acceptable solutions we have used the metaphor of the genetic evolution in a multi-objective model, where individuals are lists of possible light sources, their positions and lighting levels. For finding acceptable solutions we have used the metaphor of the genetic evolution in a multi-objective model, where every individual is a list of light sources; their positions; and lighting levels. Further, for properly evaluating each individual, we have developed two conflicting objective functions, one for optimizing the level of brightness, and the second one for maximising the energy saving, satisfying, obviously, the additional constraints to respect the architectural structure to be lighted. From the randomly initial population of individuals generations are constructed using crossover and mutation operators, whilst the fittest offspring is preserved via an elitist Pareto-dominance selection approach. In addition to the multi-objective genetic algorithm, the 3D graphic software Blender has been used in order to reproduce the architectural space to be lighted, with the aim to evaluate then, the accuracy and uniformity of the produced lighting through a physical simulation of its brightness. The main goal of the developed tool is to provide to the designer (i.e. the decision maker) a set of interiors illumination design options, for the given environment to be lit, ensuring (i) uniform illumination distribution; (ii) accuracy of the illumination produced; (iii) avoiding harsh brightness, and glare; and (iv) low energy consumptions. Two case studies have been considered in our evaluation experiments, and for each of these the algorithm was performed on two different instances and with different types of complexity respectively.

Paper Nr: 30
Title:

Enhancing Student Engagement via Reduction of Frustration with Programming Assignments using Machine Learning

Authors:

Mario Garcia Valdez, Amaury Hernandez Aguila, Juan-J. Merelo and Alejandra Mancilla Soto

Abstract: Learning to program is often regarded as a difficult task. When selecting an appropriate programming exercise, experienced instructors gauge a student´s affective state and skills to then assign an activity with the appropriate level of difficulty. This work is focused on the prediction of the affective states of programmers with different levels of expertise when learning a new programming language. For this, an interactive webbased programming platform is proposed. The platform is designed to collect data from the studentsínteraction for data analysis. Current work is focused on the prediction of affective states using non-obtrusive sensors. Specifically, the aim of this research is to evaluate the use of keyboard and mouse dynamics as an appropriate sensory input for an affective recognition system. The proposed method uses feature vectors obtained by mining data generated from both keyboard and mouse dynamics of students as they work in basic Python programming assignments, which were used to train different classification algorithms to classify learners into five different affective states: boredom, frustration, distraction, relaxation and engagement. Accuracy achieved was around 75% with J48 obtaining the best results, proving that data gathered from non-obtrusive sensors can successfully be used as another input to classification models in order to predict an individual´s affective states.

Paper Nr: 46
Title:

Experiments with Lazy Evaluation of Classification Decision Trees Made with Genetic Programming

Authors:

Sašo Karakatič, Marjan Heričko and Vili Podgorelec

Abstract: In this paper, we present a lazy evaluation approach of classification decision trees with genetic programming. We describe and experiment with the lazy evaluation that does not evaluate the whole population but evaluates only the individuals that are chosen to participate in the tournament selection method. Further on, we used dynamic weights for the classification instances, that are linked to the chance of that instance getting picked for the evaluation process. These weights change based on the misclassification rate of the instance. We test our lazy evaluation approach on 10 standard classification benchmark datasets and show that not only lazy evaluation approach uses less time to evolve the good solution, but can even produce better solution due to changing instance weights and thus preventing the overfitting of the solutions.

Posters
Paper Nr: 8
Title:

Parallelization of Real-time Control Algorithms on Multi-core Architectures using Ant Colony Optimization

Authors:

Oliver Gerlach, Florian Frick, Armin Lechler and Alexander Verl

Abstract: The emerging digitalization of production is accelerating the transformation of Industrial Control Systems (ICSs) from simple logic controllers to sophisticated systems utilizing complex algorithms which are running under strict real-time conditions. The required performance has now reached the limitations of single-core processors, making a transition to multi-core systems necessary. The parallelization of the currently monolithic and sequentially designed control algorithms is a complex problem that is further complicated by inherent hardware dependencies and real-time requirements. A fine-grained distribution of the algorithms on multiple cores while maintaining deterministic behavior is required but cannot be achieved with state of the art parallelization and scheduling algorithms. This paper presents a new parallelization approach for mapping and scheduling of model-based designs of control algorithms onto ICSs. Since the mapping and scheduling problem is NP-complete, an ACO algorithm is utilized and the solution is validated by experimental results.

Paper Nr: 10
Title:

Hierarchy Influenced Differential Evolution: A Motor Operation Inspired Approach

Authors:

Shubham Dokania, Ayush Chopra, Feroz Ahmad and Anil Singh Parihar

Abstract: Operational maturity of biological control systems have fuelled the inspiration for a large number of mathematical and logical models for control, automation and optimisation. The human brain represents the most sophisticated control architecture known to us and is a central motivation for several research attempts across various domains. In the present work, we introduce an algorithm for mathematical optimisation that derives its intuition from the hierarchical and distributed operations of the human motor system. The system comprises global leaders, local leaders and an effector population that adapt dynamically to attain global optimisation via a feedback mechanism coupled with the structural hierarchy. The hierarchical system operation is distributed into local control for movement and global controllers that facilitate gross motion and decision making. We present our algorithm as a variant of the classical Differential Evolution algorithm, introducing a hierarchical crossover operation. The discussed approach is tested exhaustively on standard test functions as well as the CEC 2017 benchmark. Our algorithm significantly outperforms various standard algorithms as well as their popular variants as discussed in the results.

Paper Nr: 17
Title:

Genetic Programming Hyper-heuristic for the Automated Synthesis of Selection Operators in Genetic Algorithms

Authors:

Evgenii Sopov

Abstract: Genetic algorithms have proved their efficiency with many hard optimization problems, but in order to achive the best results they must be fine-tuned. One such method of fine-tuning is a synthesis of new genetic operators. Hyper-heuristics represent search techniques that can be used for automating the process of selecting or generating simpler heuristics with the aim of designing new metaheuristic algorithms. In this study, we have proposed a new hyper-heuristic based on genetic programming for the automated synthesis of a selection operator in genetic algorithms. Black-Box Optimization Benchmarking is used as a training set for the genetic programming algorithm and as a test set for estimating the generalization ability of a synthesized selection operator. The results of numerical experiments are presented and discussed. The experiments have shown that the proposed approach can be used for designing new selection operators that outperform standard selection operators on average with new, previously unseen instances of hard black-box optimization problems.

Paper Nr: 18
Title:

Comparing Small Population Genetic Algorithms over Changing Landscapes

Authors:

Michael Curley and Seamus Hill

Abstract: This paper examines the performance and adaptability of a number of small population Genetic Algorithms (GAs) over a selection of dynamic landscapes. Much of the research in this area tends to focus on GA with relatively large populations for problem optimisation. However there is research, which suggests that GAs with smaller populations can also be effective over changing landscapes. This research compares the performance and adaptability of a number of these small population GA over changing landscapes. With small population GAs, convergence can occur quickly, which in turn affects the adaptability of a GA over dynamic landscapes. In this paper five GA variants using small population sizes are run over well-known unimodal and multimodal problems, which were tailored to produce dynamic landscapes. Adaptability within the population is considered a desirable feature for a GA to optimise a changing landscape and different methods are used to maintain a level of diversity within a population to avoid the problem of premature convergence, thereby allowing the GA population adapt to the dynamic nature of the search space. Initial results indicate that small population GAs can perform well in searching changing landscapes, with GAs which possess the ability to maintain diversity within the population, outperforming those that do not.

Paper Nr: 25
Title:

An EVEBO-Based BTS Localization Algorithm

Authors:

Koorosh Navi, Manoochehr Kelarestaghi and Farshad Eshghi

Abstract: In this paper, we use EVEBO, an election-inspired optimization algorithm, to solve the BTS (i.e. transceiver) localization problem. The proposed method tries to solve the classic and very important problem of achieving maximum coverage with minimum number of BTSs in a specified geographical area. It also tries to reduce the over-coverage rate, one of the undesirable phenomena in cellular networks. The EVEBO’s merit in solving the problem is measured by a common fitness function, and speed of convergence. Simulation results show that EVEBO solves the problem in much less number of evaluations compared to the best results reported in the literature for square-coverage transceivers. We also show that it can be used in a scenario involving more challenging non-square-coverage (almost circular) transceiver type with satisfactory results.

Paper Nr: 34
Title:

Analysis of a Batch Strategy for a Master-Worker Adaptive Selection Algorithm Framework

Authors:

Christopher Jankee, Sébastien Verel, Bilel Derbel and Cyril Fonlupt

Abstract: We look into the design of a parallel adaptive algorithm embedded in a master-slave scheme. The adaptive algorithm under study selects online and in parallel for each slave-node one algorithm from a portfolio. Indeed, many open questions still arise when designing an online distributed strategy that attributes optimally algorithms to distribute resources. We suggest to analyze the relevance of existing sequential adaptive strategies related to multi-armed bandits to the master-slave distributed framework. In particular, the comprehensive experimental study focuses on the gain of computing power, the adaptive ability of selection strategies, and the communication cost of the parallel system. In fact, we propose an adaptive batch mode in which a sequence of algorithms is submitted to each slave computing node to face a possibly high communication cost.

Paper Nr: 58
Title:

Termination Criteria in Evolutionary Algorithms: A Survey

Authors:

Seyyedeh Newsha Ghoreishi, Anders Clausen and Bo Noerregaard Joergensen

Abstract: Over the last decades, evolutionary algorithms have been extensively used to solve multi-objective optimization problems. However, the number of required function evaluations is not determined by nature of these algorithms which is often seen as a drawback. Therefore, a robust and reliable termination criterion is needed to stop the algorithm. There is a huge amount of knowledge encapsulated in the studies targeting termination criteria in evolutionary algorithms, but an updated integrated overview of this knowledge is missing. For this reason, we aim to conduct a systematic research through a comprehensive literature study. We extended the basic categorization of termination criteria to a more advanced one that takes the most common used termination criteria into consideration based on their specifications and the way they have been evolved over time. The survey is concluded by suggesting a road-map for future research directions.

Paper Nr: 64
Title:

Extreme Gradient Boosting Approach with Differential Evolution Tuning Applied on Service Time Prediction of Fire Events

Authors:

Marco A. R. Boaretto, Rafael Commim Busatto and Leandro dos S. Coelho

Abstract: The problems related with urban fires brings destruction, casualties and economic issues to the country. On aiming to decrease the risk specially on populous areas, the development of classification and/or forecasting systems are needed to detect these events. This paper presents a new case of study related with fire incidents to create a predictor with a machine learning technique allied with an optimization technique. Differential evolution (DE) is a population-based stochastic search evolutionary algorithm applied to nonlinear, nondifferentiable, and continuous space functions. It has shown good performance on a wide range of benchmark problems as well as many real-world applications. Extreme Gradient boosting machine (XGBoost) is an efficient boosting technique that is getting a lot of attention because of its success on winning several of competitions hosted by the site Kaggle. However, XGBoost’s hyper-parameters tuning process demands a high effort and computational time, hence the application of an optimization technique as DE. In this context, DE showed to be an efficient and robust optimization tool, presenting good and stable results on the adopted tests. Along with the XGBoost hyper-parameter tuning optimized approach, this paper presents a novel database from service times of real fire incidents of the years 2013 and 2014 that were gathered with the Fire Department of the state of Paraná on the city of Curitiba in Brazil (FDPRC). The regression results could contribute with FDPRC, to speed up the location reckoning process of the fire incidents, improving the answering time of the calling.

Area 3 - Fuzzy Computation

Full Papers
Paper Nr: 21
Title:

A Fuzzy Logic Approach to Improve Phone Segmentation - A Case Study of the Dutch Language

Authors:

Victor Milewski, Aysenur Bilgin and Tufan Kumbasar

Abstract: Phone segmentation is an essential task for Automatic Speech Recognition (ASR) systems, which still lack in performance when compared to the ability of humans' speech recognition. In this paper, we propose novel Fuzzy Logic (FL) based approaches for the prediction of phone durations using linguistic features. To the best of our knowledge, this is the first development and deployment of FL based approaches in the area of phone segmentation. In this study, we perform a case study on the Dutch IFA corpus, which consists of 50000 words. Different experiments are conducted on tuned FL Systems (FLSs) and Neural Networks (NNs). The experimental results show that FLSs are more efficient in phone duration prediction in comparison to their Neural Network counterparts. Furthermore, we observe that differentiating between the vowels and the consonants improves the performance of predictions, which can facilitate enhanced ASR systems. The FLS with the differentiation between vowels and consonants had an average Mean Average Precision Error of 43.3396\% on a k=3 fold. We believe that this first attempt of the employment of FL based approaches will be an important step for a wider deployment of FL in the area of ASR systems.

Paper Nr: 29
Title:

Determining Firing Strengths Through a Novel Similarity Measure to Enhance Uncertainty Handling in Non-singleton Fuzzy Logic Systems

Authors:

Direnc Pekaslan, Shaily Kabir, Jonathan M. Garibaldi and Christian Wagner

Abstract: Non-singleton Fuzzy Logic Systems have the potential to tackle uncertainty within the design of fuzzy systems. The inference process has a major role in determining results, being partly based on the interaction of input and antecedent fuzzy sets (in generating firing levels). Recent studies have shown that the standard technique for determining firing strengths risks substantial information loss in terms of the interaction of the input and antecedents. To address this issue, alternative approaches, which employ the centroid of intersections and similarity measures, have been developed. More recently, a novel similarity measure for fuzzy sets has been introduced, but as yet this has not been used for non-singleton fuzzy logic systems. This paper focuses on exploring the potential of this new similarity measure in combination with the similarity based inferencing approach to generate a more suitable firing level for non-singleton input. Experiments are presented for fuzzy systems trained using both noisy and noise-free time series. The prediction results of non-singleton fuzzy logic systems for the novel similarity measure and the current approaches are compared. Analysis of the results shows that the novel similarity measure, used within the similarity based inferencing approach, can be a stable and suitable method to be used in real world applications.

Short Papers
Paper Nr: 24
Title:

Testing Fuzzy Hypotheses with Fuzzy Data and Defuzzification of the Fuzzy p-value by the Signed Distance Method

Authors:

Rédina Berkachy and Laurent Donzé

Abstract: We extend the classical approach of hypothesis testing to the fuzzy environment. We propose a method based on fuzziness of data and on fuzziness of hypotheses at the same time. The fuzzy p-value with its α-cuts is provided and we show how to defuzzify it by the signed distance method. We illustrate our method by numerical applications where we treat a one and a two sided test. For the one-sided test, applying our method to the same data and performing tests on the same significance level, we compare the defuzzified p-values between different cases of null and alternative hypotheses.

Paper Nr: 27
Title:

A Framework for Automatic Generation of Fuzzy Evaluation Systems for Embedded Applications

Authors:

Daniele De Martini, Gianluca Roveda, Alessandro Bertini, Agnese Marchini and Tullio Facchinetti

Abstract: Fuzzy logic is a powerful modelling approach to build control applications and to generate knowledge-based evaluation indices. In both cases, however, the applicability to complex systems is limited by the effort required to formulate the rules, whose number grows rapidly with the number of input variables and membership functions. This work presents a framework that implements the F-IND fuzzy model to simplify the formulation of fuzzy indices, where the rules are automatically generated on the basis of the specification of best and worst cases on the membership functions of each input variable. The paper discusses the method and presents the organization of the framework that allows automatic code generation, targeting the efficient execution of the calculations on an embedded system. The framework has been tested and validated on real hardware.

Paper Nr: 53
Title:

Solving Fuzzy Answer Set Programs in Product Logic

Authors:

Ivor Uhliarik

Abstract: In recent years, foundations have been laid for a turn in logic programming paradigms in continuous domains. Fuzzy answer set programming (FASP) has emerged as a combination of a tool for non-monotonic reasoning and solving combinatorial problems (ASP) and a knowledge representation formalism that allows for modeling partial truth (fuzzy logic). There have been various attempts at designing a solver for FASP, but they either make use of transformations into optimization programs with scaling problems, operate only on finite-valued Łukasiewicz logic, or yield only approximate answer sets. Moreover, there has been no research focused on the product logic semantics in FASP. In this work we investigate the methods used in state-of-the-art classical ASP solvers with the aim of designing a FASP solver for product propositional logic. In particular, we base our approach on the conversion into fuzzy SAT (satisfiability problem) and the fuzzy generalization of the DPLL algorithm. Since both Łukasiewicz and (extended) Gödel logic can be embedded into product logic, the resulting system should be able to operate on all three logics uniformly.

Posters
Paper Nr: 44
Title:

ANFIS Synthesis by Clustering for Microgrids EMS Design

Authors:

Stefano Leonori, Alessio Martino, Antonello Rizzi and Fabio Massimo Frattale Mascioli

Abstract: Microgrids (MGs) play a crucial role for the development of Smart Grids. They are conceived to intelligently integrate the generation from Distributed Energy Resources, to improve Demand Response (DR) services, to reduce pollutant emissions and curtail power losses, assuring the continuity of services to the loads as well. In this work it is proposed a novel synthesis procedure for modelling an Adaptive Neuro-Fuzzy Inference System (ANFIS) featured by multivariate Gaussian Membership Functions (MFs) and first order Takagi-Sugeno rules. The Fuzzy Rule Base is the core inference engine of an Energy Management System (EMS) for a gridconnected MG equipped with a photovoltaic power plant, an aggregated load and an Energy Storage System (ESS). The EMS is designed to operate in real time by defining the ESS energy flow in order to maximize the revenues generated by the energy trade with the distribution grid. The ANFIS EMS is synthesized through a data driven approach that relies on a clustering algorithm which defines the MFs and the rule consequent hyperplanes. Moreover, three clustering algorithms are investigated. Results show that the adoption of kmedoids based on Mahalanobis (dis)similarity measure is more efficient with respect to the k-means, although affected by some variety in clusters composition.

Area 4 - Neural Computation

Full Papers
Paper Nr: 15
Title:

Performance of Complex-Valued Multilayer Perceptrons Largely Depends on Learning Methods

Authors:

Seiya Satoh and Ryohei Nakano

Abstract: Complex-valued multilayer perceptrons (C-MLPs) can naturally treat complex numbers, and therefore can work well for the processing of signals such as radio waves and sound waves, which are naturally expressed as complex numbers. The performance of C-MLPs can be measured by solution quality and processing time. We believe the performance seriously depends on which learning methods we employ since in the search space there exist many local minima and singular regions, which prevent learning methods from finding excellent solutions. Complex-valued backpropagation (C-BP) and complex-valued BFGS method (C-BFGS) are well-known for learning C-MLPs. Moreover, complex-valued singularity stairs following (C-SSF) has recently been proposed as a new learning method, which achieves successive learning by utilizing singular regions and guarantees monotonic decrease of training errors. Through experiments using five datasets, this paper evaluates how the performance of C-MLPs changes depending on learning methods.

Paper Nr: 20
Title:

Efficient Implementation of Self-Organizing Map for Sparse Input Data

Authors:

Josué Melka and Jean-Jacques Mariage

Abstract: Neural-based learning algorithms, which in most cases implement a lengthy iterative convergence procedure, are often hardly adapted to very sparse input data, both due to practical issues concerning time and memory usage, and to the inherent difficulty of learning in high dimensional space. However, the description of many real-world data sets is sparse by nature, and learning algorithms must circumvent this barrier. This paper proposes adaptations of the standard and the batch versions of the Self-Organizing Map algorithm, specifically fine-tuned for high dimensional sparse data, with parallel implementation efficiency in mind. We extensively evaluate the performance of both adaptations on a set of experiments carried out on several real and artificial large benchmark datasets of sparse format from the LIBSVM Data: Classi­fication. Results show that our approach brings a significant improvement in execution time.

Paper Nr: 38
Title:

Towards the Enrichment of Arabic WordNet with Big Corpora

Authors:

Georges Lebboss, Gilles Bernard, Noureddine Aliane and Mohammad Hajjar

Abstract: This paper presents a method aiming to enrich Arabic WordNet with semantic clusters extracted from a large general corpus. As the Arabic language is poor in open digital linguistic resources, we built such a corpus (more than 7.5 billion words) with ad-hoc tools. We then applied GraPaVec, a new method for word vectorization using automatically generated frequency patterns, as well as state-of-the-art Word2Vec and Glove methods. Word vectors were fed to a Self Organizing Map neural network model; the clusterings produced were then compared for evaluation with Arabic WordNet existing synsets (sets of synonymous words). The evaluation yields a F-score of 82.1 % for GrapaVec, 55.1 % for Word2Vec's Skipgram, 52.2 % for CBOW and 56.6 % for Glove, which at least shows the interest of the context that GraPaVec takes into account. We end up by discussing parameters and possible biases.

Paper Nr: 40
Title:

R-FCN Object Detection Ensemble based on Object Resolution and Image Quality

Authors:

Christoffer Bøgelund Rasmussen, Kamal Nasrollahi and Thomas B. Moeslund

Abstract: Object detection can be difficult due to challenges such as variations in objects both inter- and intra-class. Additionally, variations can also be present between images. Based on this, research was conducted into creating an ensemble of Region-based Fully Convolutional Networks (R-FCN) object detectors. Ensemble strategies explored were firstly data sampling and selection and secondly combination strategies. Data sampling and selection aimed to create different subsets of data with respect to object size and image quality such that expert R-FCN ensemble members could be trained. Two combination strategies were explored for combining the individual member detections into an ensemble result, namely average and a weighted average. R-FCNs were trained and tested on the PASCAL VOC benchmark object detection dataset. Results proved positive with an increase in Average Precision (AP), compared to state-of-the-art similar systems, when ensemble members were combined appropriately.

Paper Nr: 41
Title:

Depth Value Pre-Processing for Accurate Transfer Learning based RGB-D Object Recognition

Authors:

Andreas Aakerberg, Kamal Nasrollahi, Christoffer B. Rasmussen and Thomas B. Moeslund

Abstract: Object recognition is one of the important tasks in computer vision which has found enormous applications.Depth modality is proven to provide supplementary information to the common RGB modality for objectrecognition. In this paper, we propose methods to improve the recognition performance of an existing deeplearning based RGB-D object recognition model, namely the FusionNet proposed by Eitel et al. First, we showthat encoding the depth values as colorized surface normals is beneficial, when the model is initialized withweights learned from training on ImageNet data. Additionally, we show that the RGB stream of the FusionNetmodel can benefit from using deeper network architectures, namely the 16-layered VGGNet, in exchange forthe 8-layered CaffeNet. In combination, these changes improves the recognition performance with 2.2% incomparison to the original FusionNet, when evaluating on the Washington RGB-D Object Dataset.

Paper Nr: 43
Title:

Entorhinal Grid Cells May Facilitate Pattern Separation in the Hippocampus

Authors:

Jochen Kerdels and Gabriele Peters

Abstract: The dentate gyrus (DG) in the hippocampus of the mammalian brain is known to exhibit strong pattern separation. However, how this pattern separation arises in the DG is not well understood. Here we offer a novel hypothesis regarding this problem by demonstrating that pattern separation can already be performed by entorhinal grid cells, which are located just one synapse upstream of the DG. For our simulations we utilize a recently introduced grid cell model that interprets the behavior of grid cells as just one instance of a general information processing scheme. The obtained results challenge the established view that pattern separation occurs primarily in the DG, and they uncover a common misconception regarding the specificity of ensemble activity in grid cells.

Paper Nr: 57
Title:

Higher Order Neural Units for Efficient Adaptive Control of Weakly Nonlinear Systems

Authors:

Ivo Bukovsky, Jan Voracek, Kei Ichiji and Homma Noriyasu

Abstract: The paper reviews the nonlinear polynomial neural architectures (HONUs) and their fundamental supervised batch learning algorithms for both plant identification and neuronal controller training. As a novel contribution to adaptive control with HONUs, Conjugate Gradient batch learning for weakly nonlinear plant identification with HONUs is presented as efficient learning improvement. Further, a straightforward MRAC strategy with efficient controller learning for linear and weakly nonlinear plants is proposed with static HONUs that avoids recurrent computations, and its potentials and limitations with respect to plant nonlinearity are discussed.

Paper Nr: 61
Title:

CNN Patch--Based Voting for Fingerprint Liveness Detection

Authors:

Amirhosein Toosi, Sandro Cumani and Andrea Bottino

Abstract: Biometric identification systems based on fingerprints are vulnerable to attacks that use fake replicas of real fingerprints. One possible countermeasure to this issue consists in developing software modules capable of telling the liveness of an input image and, thus, of discarding fakes prior to the recognition step. This paper presents a fingerprint liveness detection method founded on a patch--based voting approach. Fingerprint images are first segmented to discard background information. Then, small--sized foreground patches are extracted and processed by a well--know Convolutional Neural Network model adapted to the problem at hand. Finally, the patch scores are combined to draw the final fingerprint label. Experimental results on well--established benchmarks demonstrate a promising performance of the proposed method compared with several state-of-the-art algorithms.

Paper Nr: 62
Title:

Environment Recognition based on Images using Bag-of-Words

Authors:

Taurius Petraitis, Rytis Maskeliūnas, Robertas Damaševičius, Dawid Połap, Marcin Woźniak and Marcin Gabryel

Abstract: Object and scene recognition solutions have a wide application field from entertainment apps, and medical tools to security systems. In this paper, scene recognition methods and applications are analysed, and the Bag of Words (BoW), a local image feature based scene classification model is implemented. In the BoW model every picture is encoded by a bag of visual features, which shows the quantities of different visual features of an image, but disregards any spatial information. Five different feature detectors and two feature descriptors were analyzed and two best approaches were experimentally chosen as being most effective classifying images into eight outdoor categories: forced feature detection with a grid and description using SIFT descriptor, and feature detection with SURF and description with U-SURF. Support vector machines were used for classification. We also have found that for the task of scene recognition not just the distinct features which are found by common feature detectors are important, but also the features that are uninteresting for them. Indoor scenes were experimentally classified into five categories and worse results were achieved. This shows that indoor scene classification is a much harder task and a model which does not take into account any mid-level scene information like objects of the scene is not sufficient for the task. A computer application was written in order to demonstrate the algorithm, which allows training new classifiers with different parameters and using the trained classifiers to predict the classes of new images.

Short Papers
Paper Nr: 4
Title:

Neural Network Inverse Model for Quality Monitoring - Application to a High Quality Lackering Process

Authors:

Philippe Thomas, Marie-Christine Suhner, Emmanuel Zimmermann, Hind Bril El Haouzi, André Thomas and Mélanie Noyel

Abstract: The quality requirement is an important issue for modern companies. Many tools and philosophies have been proposed to monitor quality, including the seven basic tools or the experimental design. However, high quality requirement may lead companies to work near their technological limit capabilities. In this case, classical approaches to monitor quality may be insufficient. That is why on line quality monitoring based on the neural network prediction model has been proposed. Within this philosophy, the dataset is used in order to determine the optimal setting considering the operating point and the product routing. An inverse model approach is proposed here in order to determine directly the optimal setting in order to avoid defects production. A comparison between the use of a classical multi-inputs multi-outputs NN model and a sequence of different multi-inputs single-output NN models is performed. The proposed approach is tested on a real application case.

Paper Nr: 12
Title:

Towards a Better Understanding of Deep Neural Networks Representations using Deep Generative Networks

Authors:

Jérémie Despraz, Stéphane Gomez, Héctor F. Satizábal and Carlos Andrés Peña-Reyes

Abstract: This paper presents a novel approach to deep-dream-like image generation for convolutional neural networks (CNNs). Images are produced by a deep generative network from a smaller dimensional feature vector. This method allows for the generation of more realistic looking images than traditional activation-maximization methods and gives insight into the CNN’s internal representations. Training is achieved by standard backpropagation algorithms.

Paper Nr: 19
Title:

FPGA Implementation of the Huber-Braun Neuron Model

Authors:

Marcel Beuler, Alexander Krum, Werner Bonath and Hartmut Hillmer

Abstract: The Hodgkin-Huxley model (HH) describes the initiation and propagation of action potentials in neurons closed to the biological conditions, but it is not well suited for large scale simulation of neuronal networks. In this paper, an implementation of the Huber-Braun model is presented. It is a simplified HH-type model and able to reproduce a wide variety of spiking patterns. An FPGA is selected as a reconfigurable hardware implementation platform to simulate the network functionality of the neurons. The 32-bit floating-point format and computation techniques (i.e. CORDIC) instead of LUTs are used to avoid loss of physiological information. We validated our design with a C++ program and report the synthesis result based on Xilinx Virtex 6 FPGA.

Paper Nr: 31
Title:

Unsupervised Segmentation of Hyper-spectral Images via Diffusion Bases

Authors:

Alon Schclar and Amir Averbuch

Abstract: In the field of hyper-spectral sensing, sensors capture images at hundreds and even thousands of wavelengths. These hyper-spectral images, which are composed of hyper-pixels, offer extensive intensity information which can be utilized to obtain segmentation results which are superior to those that are obtained using RGB images. However, straightforward application of segmentation is impractical due to the large number of wavelength images, noisy wavelengths and inter-wavelength correlations. Accordingly, in order to efficiently segment the image, each pixel needs to be represented by a small number of features which capture the structure of the image. In this paper we propose the diffusion bases dimensionality reduction algorithm to derive the features which are needed for the segmentation. We also propose a simple algorithm for the segmentation of the dimensionality reduced image. We demonstrate the proposed framework when applied to hyper-spectral microscopic images and hyper-spectral images obtained from an airborne hyper-spectral camera.

Paper Nr: 45
Title:

Efficient Approaches for Solving the Large-Scale k-medoids Problem

Authors:

Alessio Martino, Antonello Rizzi and Fabio Massimo Frattale Mascioli

Abstract: In this paper, we propose a novel implementation for solving the large-scale k-medoids clustering problem. Conversely to the most famous k-means, k-medoids suffers from a computationally intensive phase for medoids evaluation, whose complexity is quadratic in space and time; thus solving this task for large datasets and, specifically, for large clusters might be unfeasible. In order to overcome this problem, we propose two alternatives for medoids update, one exact method and one approximate method: the former based on solving, in a distributed fashion, the quadratic medoid update problem; the latter based on a scan and replacement procedure. We implemented and tested our approach using the Apache Spark framework for parallel and distributed processing on several datasets of increasing dimensions, both in terms of patterns and dimensionality, and computational results show that both approaches are efficient and effective, able to converge to the same solutions provided by state-of-the-art k-medoids implementations and, at the same time, able to scale very well as the dataset size and/or number of working units increase.

Paper Nr: 52
Title:

Automated Diagnostic Model Based on Heart Tissue Isoline Map Analysis

Authors:

Olga Senyukova, Danuta Brotikovskaya, Svetlana Gorokhova and Ekaterina Tebenkova

Abstract: Automated heart disease diagnostics is an important problem, especially for tissue structure defect cases. A new approach to automated diagnostics based on supervised machine learning algorithms is described in this paper. Main heart tissue layer, left ventricle myocardium, characteristics based on isoline map analysis are utilized at feature model construction stage. Histogram-based features are also extracted for comparison with the proposed method. Feature selection using chi-squared test and information gain is performed. SVM and Random Forest classifiers are used for normal/abnormal classification of left ventricle myocardium images. Different combinations of feature models and classifiers were evaluated and promising results were achieved. Isoline map-based features demonstrated superiority over histogram-based feature model and the best F-score value was above 96% on real data.

Posters
Paper Nr: 3
Title:

Emotion Recognition from Speech using Representation Learning in Extreme Learning Machines

Authors:

Stefan Glüge, Ronald Böck and Thomas Ott

Abstract: We propose the use of an Extreme Learning Machine initialised as auto-encoder for emotion recognition from speech. This method is evaluated on three different speech corpora, namely EMO-DB, eNTERFACE and SmartKom. We compare our approach against state-of-the-art recognition rates achieved by Support Vector Machines (SVMs) and a deep learning approach based on Generalised Discriminant Analysis (GerDA). We could improve the recognition rate compared to SVMs by 3%-14% on all three corpora and those compared to GerDA by 8%-13% on two of the three corpora.

Paper Nr: 14
Title:

Impact of Hidden Weights Choice on Accuracy of MLP with Randomly Fixed Hidden Neurons for Regression Problems

Authors:

Marie-Christine Suhner and Philippe Thomas

Abstract: Neural network is a well-known tool able to learn model from data with a good accuracy. However, this tool suffers from an important computational time which may be too expansive. One alternative is to fix the weights and biases connecting the input to the hidden layer. This approach has been denoted recently extreme learning machine (ELM) which is able to learn quickly a model. Multilayers perceptron and ELM have identical structure, the main difference is that only the parameters linking hidden to output layers are learned. The weights and biases which connect the input to the hidden layers are randomly chosen and they don’t evolved during the learning. The impact of the choice of these random parameters on the model accuracy is not studied in the literature. This paper draws on extensive literature concerning the feedforward neural networks initialization problem. Different feedforward neural network initialisation algorithms are recalled, and used for the determination of ELM parameters connecting input to hidden layers. These algorithms are tested and compared on several regression benchmark problems.

Paper Nr: 39
Title:

Real-Time Barcode Detection and Classification using Deep Learning

Authors:

Daniel Kold Hansen, Kamal Nasrollahi, Christoffer B. Rasmusen and Thomas B. Moeslund

Abstract: Barcodes, in their different forms, can be found on almost any packages available in the market. Detecting and then decoding of barcodes have therefore great applications. We describe how to adapt the state-of-the-art deep learning-based detector of You Only Look Once (YOLO) for the purpose of detecting barcodes in a fast and reliable way. The detector is capable of detecting both 1D and QR barcodes. The detector achieves state-of-the-art results on the benchmark dataset of Muenster BarcodeDB with a detection rate of 0.991. The developed system can also find the rotation of both the 1D and QR barcodes, which gives the opportunity of rotating the detection accordingly which is shown to benefit the decoding process in a positive way. Both the detection and the rotation prediction shows real-time performance.

Paper Nr: 47
Title:

Using Artificial Intelligence to Improve the Evaluation of Human Blastocyst Morphology

Authors:

José Celso Rocha, Diogo Lima Bezerra da Silva, João Guilherme Cândido dos Santos, Lucy Benham Whyte, Cristina Hickman, Stuart Lavery and Marcelo Fábio Gouveia Nogueira

Abstract: The morphology of the human embryo produced by in vitro fertilized (IVF) is historically used as a predictive marker of gestational success. Although there are several different proposed methods to improve determination of embryo morphology, currently, all methods rely on a manual, optical and subjective evaluation done by an embryologist. Given that tiredness, mood and distinct experience could influence the accuracy of the evaluation, the results found are very different from embryologist to embryologist and from clinic to clinic. We propose the use of an objective evaluation, with repeatability and automatization, of the human blastocyst by image processing and the use of Artificial Neural Network (i.e., Artificial Intelligence).

Paper Nr: 63
Title:

Early Prediction of the Winner in StarCraft Matches

Authors:

Antonio Álvarez-Caballero, J. J. Merelo, Pablo García Sánchez and A. Fernández-Ares

Abstract: A fast and precise prediction of the outcome of a game is essential for the design of bots that play the game; it can be used either offline as a fast way to design bot strategies or online for conserving resources and conceding defeat or speed up victory, as well as evaluating the consequences of actions. The objective of this paper is predicting the winner of a StarCraft match as soon as possible. This study is done with supervised learning, because a lot of suitable data is available. The main problem of this approach is the big amount of generated data, so it has to be selected and organised properly and be treated with proper tools. A set of six learning algorithms is used, from simpler ones to more complex algorithms. Spark and MLlib are used due to their capabilities to deal with big amounts of data. With the learned models, time of matches are restricted, trying to get a time bound for predicting results. With this approach we get that it is not necessary to play a whole match to predict its winner with high accuracy: with 10 minutes we can predict the outcome with 90% of accuracy.