Showing papers by "Enrique Alba published in 2018"

Evolutionary Deep Learning for Car Park Occupancy Prediction in Smart Cities

Abstract: This study presents a new technique based on Deep Learning with Recurrent Neural Networks to address the prediction of car park occupancy rate. This is an interesting problem in smart mobility and we here approach it in an innovative way, consisting in automatically design a deep network that encapsulates the behavior of the car occupancy and then is able to make an informed guess on the number of free parking spaces near to the medium time horizon. We analyze a real world case study consisting of the occupancy values of 29 car parks in Birmingham, UK, during eleven weeks and compare our results to other predictors in the state-of-the-art. The results show that our approach is accurate to the point of being useful for being used by citizens in their daily lives, as well as it outperforms the existing competitors.

BiPred: A Bilevel Evolutionary Algorithm for Prediction in Smart Mobility.

Abstract: This article develops the design, installation, exploitation, and final utilization of intelligent techniques, hardware, and software for understanding mobility in a modern city. We focus on a smart-campus initiative in the University of Malaga as the scenario for building this cyber⁻physical system at a low cost, and then present the details of a new proposed evolutionary algorithm used for better training machine-learning techniques: BiPred. We model and solve the task of reducing the size of the dataset used for learning about campus mobility. Our conclusions show an important reduction of the required data to learn mobility patterns by more than 90%, while improving (at the same time) the precision of the predictions of theapplied machine-learning method (up to 15%). All this was done along with the construction of a real system in a city, which hopefully resulted in a very comprehensive work in smart cities using sensors.

Low-Cost Recurrent Neural Network Expected Performance Evaluation.

Abstract: Recurrent neural networks are a powerful tool, but they are very sensitive to their hyper-parameter configuration. Moreover, training properly a recurrent neural network is a tough task, therefore selecting an appropriate configuration is critical. Varied strategies have been proposed to tackle this issue. However, most of them are still impractical because of the time/resources needed. In this study, we propose a low computational cost model to evaluate the expected performance of a given architecture based on the distribution of the error of random samples of the weights. We empirically validate our proposal using three use cases. The results suggest that this is a promising alternative to reduce the cost of exploration for hyper-parameter optimization.

CMI: An Online Multi-objective Genetic Autoscaler for Scientific and Engineering Workflows in Cloud Infrastructures with Unreliable Virtual Machines

Abstract: Cloud Computing is becoming the leading paradigm for executing scientific and engineering workflows. The large-scale nature of the experiments they model and their variable workloads make clouds the ideal execution environment due to prompt and elastic access to huge amounts of computing resources. Autoscalers are middleware-level software components that allow scaling up and down the computing platform by acquiring or terminating virtual machines (VM) at the time that workflow's tasks are being scheduled. In this work we propose a novel online multi-objective autoscaler for workflows denominated Cloud Multi-objective Intelligence (CMI), that aims at the minimization of makespan, monetary cost and the potential impact of errors derived from unreliable VMs. In addition, this problem is subject to monetary budget constraints. CMI is responsible for periodically solving the autoscaling problems encountered along the execution of a workflow. Simulation experiments on four well-known workflows exhibit that CMI significantly outperforms a state-of-the-art autoscaler of similar characteristics called Spot Instances Aware Autoscaling (SIAA). These results convey a solid base for deepening in the study of other meta-heuristic methods for autoscaling workflow applications using cheap but unreliable infrastructures.

A stop-and-start adaptive cellular genetic algorithm for mobility management of GSM-LTE cellular network users

Abstract: The optimisation of the user tracking process is one of the most challenging tasks in today’s advanced cellular networks. In this paper, we propose a new low-complexity adaptive cellular genetic algorithm to solve this problem. The proposed approach uses a torus-like structured population of candidate solutions and regulates interactions inside it by using a bi-dimensional neighbourhood. It also automatically adapts the algorithm’s parameters and regenerates the algorithm’s population using two algorithmically-light operators. In order to draw reliable conclusions and perform an encompassing assessment, extensive experiments have been conducted on 25 differently-sized realistic networks. The proposed approach has been compared against 26 state-of-the-art algorithms previously designed to solve the mobility management problem, and a thorough statistical analysis of results has been performed. The obtained results have shown that our proposal is more efficient and algorithmically less complex than most of the state-of-the-art solvers.

Waste Generation Prediction in Smart Cities Through Deep Neuroevolution

Abstract: Managing the waste collection service is a challenge in the fast-growing city context. A key to success in planning the collection is having an accurate prediction of the filling level of the waste containers. In this study we present a solution to the waste generation prediction problem based on recurrent neural networks. Particularly, we introduce a deep neuroevolutionary technique to automatically design a deep network that encapsulates the behavior of all the waste containers in a city. We analyze a real world case study consisting of one year of filling level values of 217 containers located in a city in the south of Spain and compare our results to the state-of-the-art. The results show that the predictions of our approach exceeds all its competitors and that its accuracy is a key enabler for an appropriate waste collection planning.

DLOPT: Deep Learning Optimization Library.

Abstract: Deep learning hyper-parameter optimization is a tough task. Finding an appropriate network configuration is a key to success, however most of the times this labor is roughly done. In this work we introduce a novel library to tackle this problem, the Deep Learning Optimization Library: DLOPT. We briefly describe its architecture and present a set of use examples. This is an open source project developed under the GNU GPL v3 license and it is freely available at this https URL

Comparing Deep Recurrent Networks Based on the MAE Random Sampling, a First Approach

Abstract: Recurrent neural networks have demonstrated to be good at tackling prediction problems, however due to their high sensitivity to hyper-parameter configuration, finding an appropriate network is a tough task. Automatic hyper-parameter optimization methods have emerged to find the most suitable configuration to a given problem, but these methods are not generally adopted because of their high computational cost. Therefore, in this study we extend the MAE random sampling, a low-cost method to compare single-hidden layer architectures, to multiple-hidden-layer ones. We validate empirically our proposal and show that it is possible to predict and compare the expected performance of an hyper-parameter configuration in a low-cost way.

Developing Genetic Algorithms Using Different MapReduce Frameworks: MPI vs. Hadoop

Abstract: MapReduce is a quite popular paradigm, which allows to no specialized users to use large parallel computational platforms in a transparent way. Hadoop is the most used implementation of this paradigm, and in fact, for a large amount of users the word Hadoop and MapReduce are interchangeable. But, there are other frameworks that implement this programming paradigm, such as MapReduce-MPI. Since, optimization techniques can be greatly beneficiary of this kind of data-intensive computing modeling, in this paper, we analyze the performance effect of developing genetic algorithms (GA) using different frameworks of MapReduce (MRGA). In particular, we implement MRGA using Hadoop and MR-MPI frameworks. We analyze and compare both implementations considering relevant aspects such as efficiency and scalability to solve a large dimension problem. The results show a similar efficiency level between the algorithms but Hadoop presents a better scalability.

Customer Segmentation Based on the Electricity Demand Signature: The Andalusian Case

Abstract: A smart meter enables electric utilities to get detailed insights into their customer needs, allowing them to offer tailored products and services, and to succeed in an increasingly competitive market. While in an ideal world companies would treat every customer as an individual, in practice this is rather difficult. For this reason, companies usually have to target smaller groups of customers that are similar. There are several ways of tackling this matter and finding the right approach is a key to success. Therefore, in this study we introduce the electricity demand signature, a novel approach to characterize and cluster electricity customers based on their demand habits. We test our proposal using the electricity demand of 64 buildings in Andalusia, Spain, and compare our results to the state-of-the-art. The results show that our proposal is useful for clustering customers in a meaningful way, and that it is an easy and friendly representation of the behavior of a customer that can be used for further analysis.

Running Genetic Algorithms in the Edge: A First Analysis

Abstract: Nowadays, the volume of data produced by different kinds of devices is continuously growing, making even more difficult to solve the many optimization problems that impact directly on our living quality. For instance, Cisco projected that by 2019 the volume of data will reach 507.5 zettabytes per year, and the cloud traffic will quadruple. This is not sustainable in the long term, so it is a need to move part of the intelligence from the cloud to a highly decentralized computing model. Considering this, we propose a ubiquitous intelligent system which is composed by different kinds of endpoint devices such as smartphones, tablets, routers, wearables, and any other CPU powered device. We want to use this to solve tasks useful for smart cities. In this paper, we analyze if these devices are suitable for this purpose and how we have to adapt the optimization algorithms to be efficient using heterogeneous hardware. To do this, we perform a set of experiments in which we measure the speed, memory usage, and battery consumption of these devices for a set of binary and combinatorial problems. Our conclusions reveal the strong and weak features of each device to run future algorihms in the border of the cyber-physical system.

Natural evolution tells us how to best make goods delivery: use vans

Abstract: Nowadays, city streets are populated not only by private vehicles but also by public transport, distribution of goods, and deliveries. Since each vehicle class has a maximum cargo capacity, we study in this article how authorities could improve the road traffic by changing the different vehicle proportions: sedans, minivans, full-size vans, trucks, and motorbikes, without losing the ability of moving cargo throughout the city. We have performed our study in a realistic scenario and defined a multi-objective optimization problem to be solved, so as to minimize these city metrics. Our results provide a scientific evidence that we can improve the delivery of goods in the city by reducing the number of heavy duty vehicles and fostering the use of full-size vans instead.

Finding Best Compiler Options for Critical Software Using Parallel Algorithms

Abstract: The efficiency of a software piece is a key factor for many systems Real-time programs, critical software, device drivers, kernel OS functions and many other software pieces which are executed thousands or even millions of times per day require a very efficient execution How this software is built can significantly affect the run time for these programs, since the context is that of compile-once/run-many In this sense, the optimization flags used during the compilation time are a crucial element for this goal and they could make a big difference in the final execution time In this paper, we use parallel metaheuristic techniques to automatically decide which optimization flags should be activated during the compilation on a set of benchmarking programs The using the appropriate flag configuration is a complex combinatorial problem, but our approach is able to adapt the flag tuning to the characteristics of the software, improving the final run times with respect to other spread practices

Studying Solutions of the p-Median Problem for the Location of Public Bike Stations

Abstract: The use of bicycles as a means of transport is becoming more and more popular today, especially in urban areas, to avoid the disadvantages of individual car traffic. In fact, city managers react to this trend and actively promote the use of bicycles by providing a network of bicycles for public use and stations where they can be stored. Establishing such a network involves the task of finding best locations for stations, which is, however, not a trivial task. In this work, we examine models to determine the best location of bike stations so that citizens will travel the shortest distance possible to one of them. Based on real data from the city of Malaga, we formulate our problem as a p-median problem and solve it with a variable neighborhood search algorithm that was automatically configured with irace. We compare the locations proposed by the algorithm with the real ones used currently by the city council. We also study where new locations should be placed if the network grows.

Measuring the Quality of Machine Learning and Optimization Frameworks

Abstract: Software frameworks are daily and extensively used in research, both for fundamental studies and applications. Researchers usually trust in the quality of these frameworks without any evidence that they are correctly build, indeed they could contain some defects that potentially could affect to thousands of already published and future papers. Considering the important role of these frameworks in the current state-of-the-art in research, their quality should be quantified to show the weaknesses and strengths of each software package.

Reduction of the Size of Datasets by Using Evolutionary Feature Selection: The Case of Noise in a Modern City

Abstract: Smart city initiatives have emerged to mitigate the negative effects of a very fast growth of urban areas. Most of the population in our cities are exposed to high levels of noise that generate discomfort and different health problems. These issues may be mitigated by applying different smart cities solutions, some of them require high accurate noise information to provide the best quality of serve possible. In this study, we have designed a machine learning approach based on genetic algorithms to analyze noise data captured in the university campus. This method reduces the amount of data required to classify the noise by addressing a feature selection optimization problem. The experimental results have shown that our approach improved the accuracy in 20% (achieving an accuracy of 87% with a reduction of up to 85% on the original dataset).

Smart Campus Human Tracking: The Case of University of Málaga

Abstract: Smart city initiatives have emerged to mitigate the negative effects of a very fast growth of urban areas. A number of universities are applying smart city solutions to face similar challenges in their campuses. In this study, we analyze the possibility of using low cost sensors based on detecting wireless signals of light commodity devices to track the movement of the members of the university community. This tracking information will help the university managers to provide the users with smart services. The first insight is that there were not detected barely movements through the campus during late-night/early morning hours (from 0:00H to 6:00H). In turn, the number of human flows sensed in a given direction is similar to the ones in the opposite one. The analysis of the sensed data has shown that the most mobility occurs during the opening and finishing school hours, as expected. Finally, we observed that the sensors are able to detect vehicular mobility.

An Intelligent Advisor for City Traffic Policies

Abstract: Nowadays, city streets are populated not only by private vehicles but also by public transport, fleets of workers, and deliveries. Since each vehicle class has a maximum cargo capacity, we study in this article how authorities could improve the road traffic by endorsing long term policies to change the different vehicle proportions: sedans, minivans, full size vans, trucks, and motorbikes, without losing the ability of moving cargo throughout the city. We have performed our study in a realistic scenario (map, road traffic characteristics, and number of vehicles) of the city of Malaga and captured the many details into the SUMO microsimulator. After analyzing the relationship between travel times, emissions, and fuel consumption, we have defined a multiobjective optimization problem to be solved, so as to minimize these city metrics. Our results provide a scientific evidence that we can improve the delivery of goods in the city by reducing the number of heavy duty vehicles and fostering the use of vans instead.

Un primer enfoque para medir la calidad de FIWARE

Abstract: Universidad de Malaga. Campus de Excelencia Internacional Andalucia Tech. Esta investigacion ha sido parcialmente financiada por CELTIC C2017/2-2 en colaboracion con las empresas EMERGYA y SECMOTIC en los contratos #8.06/5.47.4997 y #8.06/5.47.4996. Tambien agradecemos el apoyo del Ministerio de Economia y Competitividad y de los fondos FEDER, proyectos: TIN2014-57341-R (http://moveon.lcc.uma.es), TIN2016-81766-REDT (http://cirti.es) y TIN2017-88213-R (http://6city.lcc.uma.es), y a la Universidad de Malaga. Campus de Excelencia Internacional Andalucia Tech.