https://jamaltoutouh.github.io/NeCOL2021-03-12T12:51:06+00:00This website presents the main outcomes of <strong>NeCOL project</strong> under the <strong>Marie Sklodowska-Curie grant agreement No 799078</strong>.Jamal Toutouhtoutouh@mit.eduhttp://www.jamal.es/Jekyllhttps://jamaltoutouh.github.io/publications/gecco2020-presentation/Effective ensmebles of GANs in GECCO 20202020-07-10T00:00:00+00:00Jamal Toutouhtoutouh@mit.eduhttp://www.jamal.es/During this on-line GECCO 2020, we have presented our research work about using Evolutionary Ensemble Learning to address mode collapse in GANs.<hr />
<p>This year COVID-19 has not allowed us to meet in person with our colleagues of the evolutionary computing community, but this has not prevented us from organizing and assisting the amazing sessions and workshops that GECCO offers us every year.</p>
<p>During this on-line GECCO 2020, we have presented one of our last research works related to GANs. In this case, we have presented our paper entitled <a href="https://arxiv.org/abs/2003.13532"><strong>Re-purposing Heterogeneous Generative Ensembles with Evolutionary Computation</strong></a> in which we define a new way to use Evolutionary Ensemble Learning (EEL) in such kind of special machine learning approach.</p>
<p><img src="https://jamaltoutouh.github.io/images/blog/gecco2020_slide3.PNG" alt="GANs" title="GANs" /></p>
<p>The main idea behind this paper is to take advantage of already pre-trained heterogeneous generators according to a given loss function (optimization problem) to combine them (create mixtures of generators) in such a way that they are able to address a new optimization problem. In this case, we took 3,000 generators optimized according to the sample generation quality (FID score) and we optimized the diversity of the generated samples to address mode collapse.
In order to do so, we defined the <strong>Ensemble Optimization Problem</strong> and we addressed by using two different evolutionary approaches based on <strong>Genetic Algorithms</strong>.
The <em>presentation</em> can be downloaded from <a href="https://jamaltoutouh.github.io/downloads/presentations/gecco2020_repurposing_gans.pdf"><em>here</em></a></p>
<p><img src="https://jamaltoutouh.github.io/images/blog/gecco2020_slide19.PNG" alt="GANs" title="GANs" /></p>
<p>As we are working in this interesting project for Deep Learning community, we will be able to report new, stimulating, and remarkable results in short period of time.</p>
<hr />
<h3 id="re-purposing-heterogeneous-generative-ensembles-with-evolutionary-computation">Re-purposing Heterogeneous Generative Ensembles with Evolutionary Computation</h3>
<h4 id="abstract">Abstract</h4>
<p>Generative Adversarial Networks (GANs) are popular tools for generative modeling. The dynamics of their adversarial learning give rise to convergence pathologies during training such as mode and discriminator collapse. In machine learning, ensembles of predictors demonstrate better results than a single predictor for many tasks. In this study, we apply two evolutionary algorithms (EAs) to create ensembles to re-purpose generative models, i.e., given a set of heterogeneous generators that were optimized for one objective (e.g., minimize Frechet Inception Distance), create ensembles of them for optimizing a different objective (e.g., maximize the diversity of the generated samples). The first method is restricted by the exact size of the ensemble and the second method only restricts the upper bound of the ensemble size. Experimental analysis on the MNIST image benchmark demonstrates that both EA ensembles creation methods can re-purpose the models, without reducing their original functionality. The EA-based demonstrate significantly better performance compared to other heuristic-based methods. When comparing both evolutionary, the one with only an upper size bound on the ensemble size is the best.</p>
2020-07-10T00:00:00+00:00https://jamaltoutouh.github.io/publications/pdco2020-presentation/Parallel/distributed GAN training by using MPI at IPDPS2020/PDCO20202020-05-18T00:00:00+00:00Jamal Toutouhtoutouh@mit.eduhttp://www.jamal.es/In order to allow the use of our distributed coevolutionary GAN training method, we have developed an MPI version of Lipizzaner/Mustangs, which is presented at IPDPS2020/PDCO2020.<hr />
<p>This year <a href="http://www.ipdps.org/" title="IPDPS 2020">IPDPS 2020</a> activities are conducted in a virtual environment.
Thus, we are not going to New Orleans to present our work entitled <strong>Parallel/distributed implementation of cellular training for generative adversarial neural networks</strong>.
The main idea behind this study is to propose a <strong>distributed memory parallel implementation</strong> of our coevolutionar GAN training framework (<strong>Lipizzaner/Mustangs</strong>) for execution in high performance/supercomputing centers.
The results show that the proposed implementation is able to reduce the training times and scale properly.</p>
<ul>
<li>The paper can be seen <a href="https://arxiv.org/abs/2004.04633"><em>here</em></a></li>
<li>The presentation can be seen <a href="https://jamaltoutouh.github.io/downloads/PDCO2020.pdf"><em>here</em></a></li>
</ul>
<hr />
<h3 id="paralleldistributed-implementation-of-cellular-training-for-generative-adversarial-neural-networks">Parallel/distributed implementation of cellular training for generative adversarial neural networks</h3>
<h4 id="abstract">Abstract</h4>
<p>Generative adversarial networks (GANs) are widely used to learn generative models. GANs consist of two networks, a generator and a discriminator, that apply adversarial learning to optimize their parameters. This article presents a parallel/distributed implementation of a cellular competitive coevolutionary method to train two populations of GANs. A distributed memory parallel implementation is proposed for execution in high performance/supercomputing centers. Efficient results are reported on addressing the generation of handwritten digits (MNIST dataset samples). Moreover, the proposed implementation is able to reduce the training times and scale properly when considering different grid sizes for training.</p>
<p><img src="https://jamaltoutouh.github.io/images//PDCO2020-slide.jpg" alt="PDCO 2020 slide" /></p>
2020-05-18T00:00:00+00:00https://jamaltoutouh.github.io/communication/spainai-gan/GANs Introduction Webinar2020-04-15T00:00:00+00:00Jamal Toutouhtoutouh@mit.eduhttp://www.jamal.es/SpainAI invited me to present my research in webinar during lockdown #StayAtHome ...<p>Here you can find the webinar to introduce one of the main topics about my current research <strong>Generative Adversarial Networks</strong></p>
<p>This video, in Spanish, has two main sections with their source code and presentation.</p>
<ol>
<li><strong>Introduction to Neural Networks and Deep Learning</strong> <a href="https://jamaltoutouh.github.io/communication/introduction-neural-networks/">(find here the code and the presentation)</a></li>
<li><strong>Generative Adversarial Networks</strong></li>
</ol>
<h4 id="see-the-video">See the video</h4>
<iframe width="560" height="315" src="https://www.youtube.com/embed/dzWhocfvMpI" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen=""></iframe>
2020-04-15T00:00:00+00:00https://jamaltoutouh.github.io/communication/introduction-neural-networks/Introduction to neural networks2020-04-14T00:00:00+00:00Jamal Toutouhtoutouh@mit.eduhttp://www.jamal.es/This post contains a presentation that I use to introduce the concept of Artificial Intelligence and neural networks to the general public.<p><strong>Artificial neural networks</strong> (<strong>ANN</strong>) or, basically, <strong>neural networks</strong> (<strong>NN</strong>) are computing systems inspired by the biological neural networks that constitute animal brains. They have become one of the main models applied in machine learning and deep learning to address the most challenging tasks.</p>
<p>In this post, I am sharing a presentation that is part of my seminar about <strong>Generative Adversarial Networks</strong>. This presentation contains a set of different examples in <a href="https://github.com/generative-adversarial-networks/introduction-neural-networks">python of neural networks</a> developed from their basis. The idea is to allow the learner to understand the main foundations and operation of neural networks and back-propagation.</p>
<ul>
<li><a href="https://jamaltoutouh.github.io/downloads/presentations/introduction_to_dl.pdf"><strong>Slides</strong></a></li>
<li><a href="https://github.com/generative-adversarial-networks/introduction-neural-networks"><strong>GitHub Code:</strong></a></li>
</ul>
<h5 id="github-code">GitHub Code</h5>
<ul>
<li>
<p><code class="language-plaintext highlighter-rouge">neuron.py</code> includes the code to create a basic neuron that uses different activation functions.</p>
</li>
<li>
<p><code class="language-plaintext highlighter-rouge">basic-neuron.py</code> contains the code to create a basic neuron that is able to simulate logic functions: AND and OR
by using step function.</p>
</li>
<li><code class="language-plaintext highlighter-rouge">basic_two_layer_neural_network.py</code> contains the code to create a basic neural network with:
<ul>
<li>two inputs: x1, and x2</li>
<li>a hidden layer with two neurons: h1 and h2</li>
<li>an output layer with a neuron: o1</li>
</ul>
</li>
<li><code class="language-plaintext highlighter-rouge">two_layer_neural_network.py</code> contains the code to create a neural network with:
<ul>
<li>input_layer_size inputs</li>
<li>a hidden layer with hidden_layer_size neurons</li>
<li>an output layer with one neuron</li>
</ul>
</li>
<li><code class="language-plaintext highlighter-rouge">train-two-layer-neural-network.py</code> contains the code to create and train a basic neural network with:
<ul>
<li>two inputs: x1, and x2</li>
<li>a hidden layer with two neurons: h1 and h2</li>
<li>an output layer with a neuron: o1</li>
</ul>
</li>
</ul>
<p><a href="https://jamaltoutouh.github.io/downloads/presentations/introduction_to_dl.pdf"><img src="https://jamaltoutouh.github.io/images/introduction-nn-main.png" alt="Fist slide" /></a></p>
2020-04-14T00:00:00+00:00https://jamaltoutouh.github.io/publications/waste-management-journal/Published our work about Smart Waste Management in the Waste Management journal2020-03-02T00:00:00+00:00Jamal Toutouhtoutouh@mit.eduhttp://www.jamal.es/The research carried out about optimizing the garbage collection points' localization, and configuration provided competitive results. Waste Management journal has published our research work on this topic.<hr />
<p>Find the article <a href="https://jamaltoutouh.github.io/downloads/wastemanagement2020.pdf"><strong>here</strong></a> or <a href="https://zenodo.org/record/4598742"><strong>here (Zenodo repository)</strong></a> and cite as:
Rossit, D. G., Toutouh, J., & Nesmachnow, S. (2020). <strong>Exact and heuristic approaches for multi-objective garbage accumulation points location in real scenarios</strong>. <em>Waste Management</em>, 105, 467-481.
https://doi.org/10.1016/j.wasman.2020.02.016</p>
<hr />
<h3 id="exact-and-heuristic-approaches-for-multi-objective-garbage-accumulation-points-location-in-real-scenarios">Exact and heuristic approaches for multi-objective garbage accumulation points location in real scenarios</h3>
<h4 id="abstract">Abstract</h4>
<p>Municipal solid waste management is a major challenge for nowadays urban societies, because it accounts for a large proportion of public budget and, when mishandled, it can lead to environmental and social problems. This work focuses on the problem of locating waste bins in an urban area, which is considered to have a strong influence in the overall efficiency of the reverse logistic chain. This article contributes with an exact multiobjective approach to solve the waste bin location in which the optimization criteria that are considered are: the accessibility to the system (as quality of service measure), the investment cost, and the required frequency of waste removal from the bins (as a proxy of the posterior routing costs). In this approach, different methods to obtain the objectives ideal and nadir values over the Pareto front are proposed and compared. Then, a family of heuristic methods based on the PageRank algorithm is proposed which aims to optimize the accessibility to the system, the amount of collected waste and the installation cost. The experimental evaluation was performed on real-world scenarios of the cities of Montevideo, Uruguay, and Bahía Blanca, Argentina. The obtained results show the competitiveness of the proposed approaches for constructing a set of candidate solutions that considers the different trade-offs between the optimization criteria.</p>
<p><img src="https://github.com/jamaltoutouh/jamaltoutouh.github.io/blob/master/images/blog/content/waste-management-instance.jpg?raw=true" alt="Illustration of the problem representation" title="Illustration of the problem representation" /></p>
<h4 id="highlights">Highlights</h4>
<ul>
<li>Suitable location of waste bins is crucial for waste management.</li>
<li>Different criteria need to be considered for optimizing bins location.</li>
<li>Mathematical programming methods need to be adapted for dealing with this problem.</li>
<li>PageRank-based heuristics are fast in obtaining locations of waste bins.</li>
<li>Proposed approaches were able to solve real scenarios of two different cities.</li>
</ul>
2020-03-02T00:00:00+00:00https://jamaltoutouh.github.io/publications/smart-waste-management-01/Where to Locate Garbage Collection Points (Bins)?2019-11-25T00:00:00+00:00Jamal Toutouhtoutouh@mit.eduhttp://www.jamal.es/We are researching about optimization problem of locating and configuring the garbage collection points. The main idea is to provide the best service to the city inhabitants while minimizing the costs. Some results have been already published.<hr />
<p>Enhancing efficiency in Municipal Solid Waste Management (MSWM) Systems is crucial for local governments and inhabitants. This post presents two research works published in different journals addressing MSWM. This research has been carried out in collaboration with two other researchers:</p>
<ul>
<li><a href="https://www.fing.edu.uy/inco/grupos/cecal/hpc/pmwiki/pmwiki.php?n=Main.SergioNesmachnow"><strong>Sergio Nesmachnow, Ph.D.</strong></a> Full Professor in the Numerical Center (CeCal) at Computer Science Institute, Engineering Faculty <strong>Universidad de la Republica</strong> (Uruguay).</li>
<li><a href="https://www.conicet.gov.ar/new_scp/detalle.php?keywords=&id=44156&datos_academicos=yes"><strong>Diego Gabriel Rossit, Ph.D.</strong></a> Researcher at the Department of Engineering at Universidad Nacional del Sur and at <strong>CONICET</strong> (Argentina).</li>
</ul>
<p>The main goal of this research is to provide an efficient waste collection infrastructure that maximizes the quality of the service provided to the citizens, but taking into account the costs. These costs includes both the economical costs and the required number of visits of the collection vehicles. These studies have been applied in real regions of Uruguay and Argentina.
During <em>LION 12</em> (2018) we presented a preliminary approach <a href="https://jamaltoutouh.github.io/downloads/Toutouh_LION2018.pdf"><strong>(see the slides)</strong></a>.</p>
<p><img src="https://github.com/jamaltoutouh/jamaltoutouh.github.io/blob/master/images/blog/waste-management-representation.jpg?raw=true" alt="Illustration of the problem representation" title="Illustration of the problem representation" /></p>
<p>In order to address this MSWM problem, we have proposed different methods based on <em>Linear Programming</em>, different <em>heuristics</em> (such as PageRank), and <em>Evolutionary Algorithms</em>.</p>
<p>We have already published the results in the following journals:</p>
<ol>
<li>
<p>J. Toutouh, D. G. Rossit, and S. Nesmachnow (2019). <strong>Soft computing methods for multiobjective location of garbage accumulation points in smart cities</strong>. <em>Annals of Mathematics and Artificial Intelligence</em> <em>(In Pres)</em> <a href="https://link.springer.com/article/10.1007/s10472-019-09647-5">doi: 10.1007/s10472-019-09647-5</a>.</p>
</li>
<li>
<p>D. G. Rossit, S. Nesmachnow, and J. Toutouh (2019). <strong>A bi-objective integer programming model for locating garbage accumulation points: a case study</strong>. <em>Revista de Ingeniería</em> <em>(In Pres)</em> <a href="https://doi.org/10.17533/udea.redin.20190509">doi: 10.17533/udea.redin.20190509</a>.</p>
</li>
</ol>
<p>And the following conference:</p>
<ol>
<li>J. Toutouh, D. Rossit, S. Nesmachnow (2018). <strong>Computational intelligence for locating garbage accumulation points in urban scenarios</strong>. <em>International Conference on Learning and Intelligent Optimization, LION 12</em>, pp. 1-15, 2018. <em>Lecture Notes in Computer Science, vol 11353.</em> <a href="https://doi.org/10.1007/978-3-030-05348-2_34">doi: 10.1007/978-3-030-05348-2_34</a>.</li>
</ol>
<p>See the abstracts bellow:</p>
<h3 id="soft-computing-methods-for-multiobjective-location-of-garbage-accumulation-points-in-smart-cities">Soft computing methods for multiobjective location of garbage accumulation points in smart cities</h3>
<p>This article describes the application of soft computing methods for solving the problem of locating garbage accumulation points in urban scenarios. This is a relevant problem in modern smart cities, in order to reduce negative environmental and social impacts in the waste management process, and also to optimize the available budget from the city administration to install waste bins. A specific problem model is presented, which accounts for reducing the investment costs, enhance the number of citizens served by the installed bins, and the accessibility to the system. A family of single- and multi-objective heuristics based on the PageRank method and two mutiobjective evolutionary algorithms are proposed. Experimental evaluation performed on real scenarios on the cities of Montevideo (Uruguay) and Bahía Blanca (Argentina) demonstrates the effectiveness of the proposed approaches. The methods allow computing plannings with different trade-off between the problem objectives. The computed results improve over the current planning in Montevideo and provide a reasonable budget cost and quality of service for Bahía Blanca.</p>
<h3 id="a-bi-objective-integer-programming-model-for-locating-garbage-accumulation-points-a-case-study">A bi-objective integer programming model for locating garbage accumulation points: a case study</h3>
<p>Enhancing efficiency in Municipal Solid Waste (MSW) management is crucial for local governments, which are generally in charge of collection, since this activity explains a large proportion of their budgetary expenses. The incorporation of decision support tools can contribute to improve the MSW system, specially by reducing the required investment of funds. This article proposes a mathematical formulation, based on integer programming, to determine the location of garbage accumulation points while minimizing the expenses of the system, i.e., the installment cost of bins and the required number of visits the collection vehicle which is related with the routing cost of the collection. The model was tested in some scenarios of an important Argentinian city that stills has a door-to-door system, including instances with unsorted waste, which is the current situation of the city, and also instances with source classified waste. Although the scenarios with classified waste evidenced to be more challenging for the proposed resolution approach, a set of solutions was provided in all scenarios. These solutions can be used as a starting point for migrating from the current door-to-door system to a community bins system.</p>
<h3 id="computational-intelligence-for-locating-garbage-accumulation-points-in-urban-scenarios">Computational Intelligence for Locating Garbage Accumulation Points in Urban Scenarios</h3>
<p>This article presents computational intelligence methods for solving the problem of locating garbage accumulation points in urban scenarios, which is a relevant problem in nowadays smart cities to optimize budget and reduce negative environmental and social impacts. The problem model considers reducing the investment costs, enhance the proportion of citizens served by bins, and the accessibility to the system. A family of heuristics based on the generic PageRank schema and a mutiobjective evolutionary algorithm are proposed. Experimental evaluation performed on real scenarios on the city of Montevideo, Uruguay, demonstrates the effectiveness of the proposed approaches. The methods allow computing plannings with different trade-off between the problem objectives and improving over the current planning.</p>
2019-11-25T00:00:00+00:00https://jamaltoutouh.github.io/publications/madrid-central-cities/What does science really say about Madrid Central?2019-09-23T00:00:00+00:00Jamal Toutouhtoutouh@mit.eduhttp://www.jamal.es/During ICSC-CITIES 2019 we will present our analysis of the environmental impact (air and noise pollution) of this measure. You can consult here the complete outcomes of this work.<hr />
<p>Briefly, this post highlights the main outcomes of our research work entitled <strong>Assessing the environmental impact of car restrictions policies: Madrid Central case</strong>.</p>
<p>Authors:</p>
<ul>
<li><a href="https://www.researchgate.net/profile/Irene_Lebrusan_Murillo"><strong>Irene Lebrusán, Ph.D.</strong></a> in Sociology and RCC Postdoctoral Researcher at <strong>Harvard University</strong>.</li>
<li><a href="https://www.jamal.es"><strong>Jamal Toutouh, Ph.D.</strong></a> in Computer Science and Marie-Curie Postdoctoral Fellow at <strong>Massachusetts Institute of Technology</strong>.</li>
</ul>
<hr />
<h4 id="motivation">Motivation</h4>
<p>Air pollution is the top health hazard in the European Union (EU): it reduces life expectancy, provokes loss of years of a healthy life, and diminishes the quality of life. Noise pollution is also a major environmental concern, as it critically impacts the quality of life of the city inhabitants. Accordingly, and under the risk of fines, Member States of the EU are required to reduce noise and air pollutants.
The European Union specifically demanded to Spain, under the risk of important economic sanctions, to reduce air pollutants (mainly nitrogen dioxide, NO2). Thanks to measures like Madrid Central, focused on the traffic reduction, Spain managed to stop the disciplinary action in May 2018.
However, this measure has been controversial, as different social and political groups consider it to be bothersome and ineffective. In fact, and after the elections (held on May, 26th 2019) the new government reversed the driving ban stated by Madrid Central, applying a moratorium to override the measure. The question here is: Was Madrid Central an effective measure to reduce air pollutants and noise?</p>
<p>We (<a href="https://www.researchgate.net/profile/Irene_Lebrusan_Murillo"><em>Irene Lebrusán, Ph.D.</em></a>, a postdoctoral researcher at Harvard and Jamal Toutouh, Ph.D., postdoctoral researcher at MIT) decided to apply Data Science to analyse the impact of Madrid Central over air quality and noise. We analyzed data gathered by the sensors installed in Madrid Central and shared though the <a href="https://datos.madrid.es">Madrid Open Data Portal</a>. The temporal range of this study starts in December of 2016 and finishes in may of 2019, i.e., 30 moths groped in two periods: 24 months previous Madrid Central (Pre-MC) and six months with the car restrictions (Post-MC). The complete work, entitled <a href="https://jamaltoutouh.github.io/downloads/lebrusan2019.pdf"><em>Assessing the environmental impact of car restrictions policies: Madrid Central case</em></a>, and results will presented during the <a href="http://icsc-cities2019.com/EN_index.html"><em>II Ibero-American Congress of Smart Cities (ICSC-CITIES 2019)</em></a>.</p>
<h4 id="results">Results</h4>
<p>This post highlights the results in terms of NO2, as this is the pollutant that almost leads Spain to the European Court. If you are interested in the results of other air pollutants, you can check it in our <a href="https://jamaltoutouh.github.io/downloads/lebrusan2019.pdf"><em>paper</em></a>.</p>
<p>The following table reports the minimum (min), mean, and maximum (max) values of the concentration of NO2 in the air sensed during the evaluated periods in terms of micro-grams per cubic meter.</p>
<table>
<thead>
<tr>
<th style="text-align: right"> </th>
<th style="text-align: right">Pre-MC</th>
<th style="text-align: right"> </th>
<th style="text-align: center">-</th>
<th style="text-align: right"> </th>
<th style="text-align: right">Post-MC</th>
<th style="text-align: right"> </th>
</tr>
<tr>
<th style="text-align: right">min</th>
<th style="text-align: right">mean</th>
<th style="text-align: right">max</th>
<th style="text-align: center">-</th>
<th style="text-align: right">min</th>
<th style="text-align: right">mean</th>
<th style="text-align: right">max</th>
</tr>
</thead>
<tbody>
<tr>
<td style="text-align: right">15.00</td>
<td style="text-align: right">46.92</td>
<td style="text-align: right">96.00</td>
<td style="text-align: center">-</td>
<td style="text-align: right">8.00</td>
<td style="text-align: right">39.60</td>
<td style="text-align: right">96.00</td>
</tr>
</tbody>
</table>
<p>The following figure shows the mean and standard deviation of the concentration of NO2 grouped by months. When we compare the same months with the car restrictions or without them (e.g., May of 2019 vs May of 2018 and 2017) we observe that the NO2 concentration is significantly reduced in more than one third (35.65% of reduction).
<img src="https://jamaltoutouh.github.io/images/blog/air_NO2.png" alt="NO2 mean and standard deviation" title="Mean and standard deviation" /></p>
<p>Finally, we carried out a temporal analysis of the data. The next figure shows the NO2 concentration during the 30 months evaluated in this research. According to the regressions computed (black lines in the figure), there is a declined trend over time on the air concentration of NO2. That is, Madrid Central has caused the NO2 concentration to decrease over time.
<img src="https://jamaltoutouh.github.io/images/blog/air_NO2_regression.png" alt="NO2 regression" title="NO2 regression" /></p>
<p>The lowering of NO2 is a very positive result. This is the component of pollution that affects health the most, increasing bronchitis asthma and lung problems, especially among the children and the older people. Besides, this is the component on which the lowering was specifically required to Spain by the European Commission. Accordingly, the reduction of this pollutant is extremely positive, not just having a positive effect on health but fulfilling so the international directions and so, avoiding the risk of fine.</p>
<p>A similar positive impact on noise pollution is shown when we apply Data Science to evaluate the data. <a href="https://jamaltoutouh.github.io/downloads/lebrusan2019.pdf"><em>Our paper</em></a> further illustrates the analysis and results.</p>
<p>The main draft of our paper can be downloaded from <a href="https://jamaltoutouh.github.io/downloads/lebrusan2019.pdf"><em>here</em></a></p>
<hr />
<h3 id="assessing-the-environmental-impact-of-car-restrictions-policies-madrid-central-case">Assessing the environmental impact of car restrictions policies: Madrid Central case</h3>
<h4 id="abstract">Abstract</h4>
<p>With the increase of population living in urban areas, many transportation-related problems have grown very rapidly. Pollution causes many inhabitants health problems. A major concern for the International Community is pollution, which causes many inhabitants health problems. Accordingly, and under the risk of fines, countries are required to reduce noise and air pollutants. As a way to do so, road restrictions policies are applied in urban areas. Evaluating objectively the benefits of this type of measures is important to asses their real impact. In this work, we analyze the application of Madrid Central (MC), which is a set of road traffic limitation measures applied in the downtown of Madrid (Spain), by using smart city tools. According to our results, MC significantly reduces the nitrogen dioxide (NO2) concentration in the air and the levels of noise in Madrid, while not arising any border effect.</p>
2019-09-23T00:00:00+00:00https://jamaltoutouh.github.io/communication/ola2019/Special Session on Computational Intelligence for Smart Cities2019-08-09T00:00:00+00:00Jamal Toutouhtoutouh@mit.eduhttp://www.jamal.es/We ar organizing a Special Session on Computational Intelligence for Smart Cities that will be carried out during International Conference on Optimization and Learning (Cadiz, Spain, February 17-19, 2020)<h3 id="special-session-on-computational-intelligence-for-smart-cities">Special Session on Computational Intelligence for Smart Cities</h3>
<p><a href="https://www.fing.edu.uy/inco/grupos/cecal/hpc/si_ola2020">www.fing.edu.uy/inco/grupos/cecal/hpc/si_ola2020</a></p>
<h3 id="international-conference-on-optimization-and-learning">International Conference on Optimization and Learning</h3>
<p>OLA 2020, Cádiz, Spain, February 17-19, 2020
<a href="ola2020.sciencesconf.org">ola2020.sciencesconf.org</a></p>
<p>Smart cities are based on the synergistic application of communication and information technologies, interconnected devices and information from sensors to improve the quality of life of their citizens.</p>
<p>Many real-life problems arise in modern smart cities, including those related to smart transportation systems, smart buildings, smart communications, and smart grid/energy networks. Innovative resolution approaches have been proposed in the literature in recent years, including those dealing with computational intelligence, learning, optimization, and other novel problem-solving strategies.</p>
<p>The Special Session on Computational Intelligence for Smart Cities aims at discussing recent advances and exploring future directions on the application of computational methods to solve a wide range of problems arising in smart cities.</p>
<p>The topics of interest include, but are not limited to:</p>
<ul>
<li>Learning and data science for smart cities</li>
<li>Computational intelligence for smart energy, energy efficiency and sustainability (environmental, social, economic)</li>
<li>Computational intelligence for logistics</li>
<li>Novel resolution approaches for infrastructure, energy and environmental problems</li>
<li>Optimization and management for smart mobility</li>
<li>Computational intelligence in smart homes and Internet of Things</li>
<li>Computational methods to improve governance and citizenship</li>
<li>Computational intelligence in smart healthcare systems</li>
<li>Computational intelligence in tourism and entertainment</li>
<li>Computational intelligence in circular economy</li>
<li>Cyberphysical systems and Internet of Things</li>
<li>Computational intelligence for security, big data, open data, and software</li>
</ul>
<p><strong>SUBMISSION OF PAPERS</strong></p>
<p>Submissions in two different formats will be accepted:</p>
<ul>
<li>S1: Extended abstracts of work-in-progress and position papers of a maximum of 3 pages</li>
<li>S2: Original research contributions of a maximum of 10 pages</li>
</ul>
<p><strong>IMPORTANT DATES</strong></p>
<ul>
<li>Submission deadline: <strong>Ocober 4, 2019</strong></li>
<li>Notification of acceptance: <strong>November 16, 2019</strong></li>
</ul>
<p><strong>PROCEEDINGS</strong></p>
<p>Accepted papers in categories S1 and S2 will be published in the proceedings that will be available at the conference. In addition, a post-conference indexed Springer book will be published. Participants will be invited to submit updated versions of their work for consideration. Special issues of indexed journals are planned too.</p>
<p><strong>SPECIAL SESSION ORGANIZERS</strong></p>
<ul>
<li>Sergio Nesmachnow, Universidad de la República, Uruguay</li>
<li><strong>Jamal Toutouh</strong>, Massachusetts Institute of Technology, USA</li>
<li>Luis Hernández, Universidad de Valladolid, Spain</li>
<li>Renzo Massobrio, Universidad de la República, Uruguay and Universidad de Cádiz, Spain</li>
</ul>
2019-08-09T00:00:00+00:00https://jamaltoutouh.github.io/publications/gecco-presentation/Mustangs in GECCO 20192019-07-17T00:00:00+00:00Jamal Toutouhtoutouh@mit.eduhttp://www.jamal.es/My colleagues were at GECCO 2019 presenting our last research about evolutionary computing, including of course, Mustangs.<hr />
<p>Last week <a href="https://gecco-2019.sigevo.org/" title="GECCO 2019">GECCO 2019</a> was held in Prague and my colleagues from ALFA went there to present some of our exiting research about evolutionary computation.</p>
<p>A paper about our <strong>Mustangs</strong> framework, which is one of the pillars on which NeCOL is based, was accepted as full paper to be presented in this event.
Mustangs is about fostering diversity when training generative adversarial networks applying a coevolutionary method to improve the training process. <br />
The paper is entitled <em>Spatial Evolutionary Generative Adversarial Networks</em> and it can be downloaded from <a href="http://alfagroup.csail.mit.edu/sites/default/files/documents/2019Spatial_Evolutionary_Generative_Adversarial_Networks.pdf" title="FULL PAPER"><em>here</em></a>.</p>
<p><a href="https://alfagroup.csail.mit.edu/erik" title="Erik">Erik Hemberg</a> was in charge to present this work to the audience, that showed a lot of interest during the session. We had very fruitful feedback, which would open new and excited research lines.</p>
<p>The <em>presentation</em> can be downloaded from <a href="https://jamaltoutouh.github.io/downloads/GECCO-2019-Mustangs.pdf"><em>here</em></a></p>
<p>As we are working in this interesting project for Deep Learning community, we will be able to report new, stimulating, and remarkable results in short period of time.</p>
<hr />
<h3 id="spatial-evolutionary-generative-adversarial-networks">Spatial Evolutionary Generative Adversarial Networks</h3>
<h4 id="abstract">Abstract</h4>
<p>Generative adversary networks (GANs) suffer from training pathologies such as instability and mode collapse. These pathologies mainly arise from a lack of diversity in their adversarial interactions. Evolutionary generative adversarial networks apply the principles of evolutionary computation to mitigate these problems. We hybridize two of these approaches that promote training diversity. One, E-GAN, at each batch, injects mutation diversity by training the (replicated) generator with three independent objective functions then selecting the resulting best performing generator for the next batch. The other, Lipizzaner, injects population diversity by training a two-dimensional grid of GANs with a distributed evolutionary algorithm that includes neighbor exchanges of additional training adversaries, performance based selection and population-based hyper-parameter tuning. We propose to combine mutation and population approaches to diversity improvement. We contribute a superior evolutionary GANs training method, Mustangs, that eliminates the single loss function used across Lipizzaner ’s grid. Instead, each training round, a loss function is selected with equiprobability, from among the three E-GAN uses. Experimental analyses on a standard benchmark, MNIST, demonstrate that Mustangs provides a statistically faster training method resulting in more accurate networks.</p>
2019-07-17T00:00:00+00:00https://jamaltoutouh.github.io/communication/brussels/Discussing AI in the REA2019-07-01T00:00:00+00:00Jamal Toutouhtoutouh@mit.eduhttp://www.jamal.es/We are glad to have participated in the "Artificial intelligence a way forward for Europe" meeting, in which outstanding researchers presented their current work and discussed AI and how its application may change the European society.<p>During June 17-18 (2019), the <a href="https://europa.eu/european-union/about-eu/agencies/rea">Research Executive Agency (REA)</a> of the European Commission
held the <strong>Artificial intelligence a way forward for Europe</strong> meeting.
The main idea behind this event was providing an environment of debate in which outstanding AI European funded researchers presented their
current work and discuss the future of AI in Europe and to point possible future directions of the new policies.</p>
<p>We were invited to present our work regarding the application of <strong>co-evolutionary algorithms</strong> to address
<strong>efficient deep learning</strong>, specifically, to deal with generative adversarial networks (GAN) training.
Besides, we also presented the use of such novel approach in two real world use cases for our society: <strong>cybersecurity</strong> and <strong>smart cities</strong>.</p>
<p><a href="https://jamaltoutouh.github.io/downloads/NeCOL-Brussels.pdf"><em>Download the presentation</em></a>.</p>
<p><a href="https://twitter.com/MSCActions/status/1141048188796919808"><em>Tweet launched by the ofical MSCA acount</em></a>.</p>
2019-07-01T00:00:00+00:00