| Carrera D, Santana R and Lozano JA (2019), "Detection of sand dunes on Mars using a regular vine-based classification approach", Knowledge Based Systems. Vol. 163, pp. 858-874. |
| Abstract: This paper deals with the problem of detecting sand dunes from remotely sensed images of the surface of Mars. We build on previous approaches that propose methods to extract informative features for the classification of the images. The intricate correlation structure exhibited by these features motivates us to propose the use of probabilistic classifiers based on R-vine distributions to address this problem. R-vines are probabilistic graphical models that combine a set of nested trees with copula functions and are able to model a wide range of pairwise dependencies. We investigate different strategies for building R-vine classifiers and compare them with several state-of-the-art classification algorithms for the identification of Martian dunes. Experimental results show the adequacy of the R-vine-based approach to solve classification problems where the interactions between the variables are of a different nature between classes and play an important role in that the classifier can distinguish the different classes. |
BibTeX:
@article{Carrera_et_al:2019,
author = {Diana Carrera and Roberto Santana and Jose Antonio Lozano},
title = {Detection of sand dunes on Mars using a regular vine-based classification approach},
journal = {Knowledge Based Systems},
year = {2019},
volume = {163},
pages = {858--874},
url = {https://www.sciencedirect.com/science/article/pii/S0950705118304970}
}
|
| Garciarena U, Mendiburu A and Santana R (2019), "Towards automatic construction of multi-network models for heterogeneous multi-task learning", CoRR. Vol. abs/1903.09171 |
| Abstract: Multi-task learning, as it is understood nowadays, consists of using one single model to carry out several similar tasks. From classifying hand-written characters of different alphabets to figuring out how to play several Atari games using reinforcement learning, multi-task models have been able to widen their performance range across different tasks, although these tasks are usually of a similar nature. In this work, we attempt to widen this range even further, by including heterogeneous tasks in a single learning procedure. To do so, we firstly formally define a multi-network model, identifying the necessary components and characteristics to allow different adaptations of said model depending on the tasks it is required to fulfill. Secondly, employing the formal definition as a starting point, we develop an illustrative model example consisting of three different tasks (classification, regression and data sampling). The performance of this model implementation is then analyzed, showing its capabilities. Motivated by the results of the analysis, we enumerate a set of open challenges and future research lines over which the full potential of the proposed model definition can be exploited. |
BibTeX:
@article{Garciarena_et_al:2019,
author = {Garciarena, Unai and Mendiburu, Alexander and Santana, Roberto},
title = {Towards automatic construction of multi-network models for heterogeneous multi-task learning},
journal = {CoRR},
year = {2019},
volume = {abs/1903.09171},
url = {http://arxiv.org/abs/1903.09171}
}
|
| Lima RHR, Pozo ATR and Santana R (2019), "Automatic design of convolutional neural networks using grammatical evolution", In 2019 8th Brazilian Conference on Intelligent Systems (BRACIS). , pp. 329-334. |
| Abstract: The use of Convolutional Neural Networks (CNNs) has been demonstrated to be a solid approach for solving many machine learning problems, such as image classification and natural language processing tasks. Usual CNN architectures are composed of many convolutions, pooling and fully connected layers, from which the networks also learn a suitable representation for the data being processed. The manual design of CNNs is a complex task due to the high number of possible parameter configurations. Recent studies about automatic design of CNNs have shown positive results. Since it can be expressed as a hyperparameter optimization problem, in this study we propose to explore the design of CNN architectures through the use of Grammatical Evolution (GE). GE is a grammar based approach where a grammar is used to define the CNN components and structural rules. We performed a set of experiments using two well-known image classification datasets, the MNIST and CIFAR-10. The obtained results show that the presented approach achieved competitive results, while maintaining relatively small architectures, when compared with similar state-of-the-art approaches. |
BibTeX:
@inproceedings{Lima_et_al:2019a,
author = {R. H. R. Lima and A. T. R. Pozo and R. Santana},
title = {Automatic design of convolutional neural networks using grammatical evolution},
booktitle = {2019 8th Brazilian Conference on Intelligent Systems (BRACIS)},
year = {2019},
pages = {329--334},
url = {https://ieeexplore.ieee.org/abstract/document/8923816}
}
|
| Magalhães D, Pozo A and Santana R (2019), "An empirical comparison of distance/similarity measures for Natural Language Processing", In Anais do XVI Encontro Nacional de Inteligência Artificial e Computacional. Porto Alegre, RS, Brasil , pp. 717-728. SBC. |
| Abstract: Text Classification is one of the tasks of Natural Language Processing (NLP). In this area, Graph Convolutional Networks (GCN) has achieved values higher than CNN's and other related models. For GCN, the metric that defines the correlation between words in a vector space plays a crucial role in the classification because it determines the weight of the edges between two words (represented by nodes in the graph). In this study, we empirically investigated the impact of thirteen measures of distance/similarity. A representation was built for each document using word embedding from word2vec model. Also, a graph-based representation of five dataset was created for each measure analyzed, where each word is a node in the graph, and each edge is weighted by distance/similarity between words. Finally, each model was run in a simple graph neural network. The results show that, concerning text classification, there is no statistical difference between the analyzed metrics and the Graph Convolution Network. Even with the incorporation of external words or external knowledge, the results were similar to the methods without the incorporation of words. However, the results indicate that some distance metrics behave better than others in relation to context capture, with Euclidean distance reaching the best values or having statistical similarity with the best. |
BibTeX:
@inproceedings{Magalhaes_et_al:2019,
author = {Dimmy Magalhães and Aurora Pozo and Roberto Santana},
title = {An empirical comparison of distance/similarity measures for Natural Language Processing},
booktitle = {Anais do XVI Encontro Nacional de Inteligência Artificial e Computacional},
publisher = {SBC},
year = {2019},
pages = {717--728},
url = {https://sol.sbc.org.br/index.php/eniac/article/view/9328},
doi = {10.5753/eniac.2019.9328}
}
|
| Mei N, Sheikh U, Santana R and Soto D (2019), "How the brain encodes meaning: Comparing word embedding and computer vision models to predict fMRI data during visual word recognition", In Proceedings of the 2019 Conference on Cognitive Computational Neuroscience. Berlin, Germany , pp. 863-866. |
| Abstract: The brain representational spaces of conceptual knowledge remain unclear. We addressed this question in a functional MRI study in which 27 participants were required to either read visual words or think about the concepts that words represented. To examine the properties of the semantic representations in the brain, we tested different encoding models based on word embeddings models -FastText (Bojanowski, Grave, Joulin, & Mikolov, 2017), GloVe (Pennington, Socher, & Manning, 2014), word2vec (Mikolov, Sutskever, Chen, Corrado, & Dean, 2013)-, and, image vision models -VGG19 (Simonyan & Zisserman, 2014), MobileNetV2 (Howard et al., 2017), DenseNet121 (Huang, Liu, Van Der Maaten, & Weinberger, 2017)- fitted with the image referents of the words. These models were used to predict BOLD responses in putative substrates of the semantic network. We fitted and predicted the brain response using the feature representations extracted from the word embedding and computer vision models. Our results showed that computer vision models outperformed word embedding models in explaining brain responses during semantic processing tasks. Intriguingly, this pattern occurred independently of the task demand (reading vs thinking about the words). The results indicated that the abstract representations from the embedding layer of computer vision models provide a better semantic model of how the brain encodes word meaning. https://tinyurl.com/y5davcs6. |
BibTeX:
@inproceedings{Mei_et_al:2019,
author = {Mei, Ning and Sheikh, Usman and Santana, Roberto and Soto, David},
title = {How the brain encodes meaning: Comparing word embedding and computer vision models to predict fMRI data during visual word recognition},
booktitle = {Proceedings of the 2019 Conference on Cognitive Computational Neuroscience},
year = {2019},
pages = {863--866},
url = {https://ccneuro.org/2019/proceedings/0000863.pdf}
}
|
| Montenegro C, Lopez-Zorrilla A, Mikel-Olaso J, Santana R, Justo R, Lozano JA and Torres MI (2019), "A Dialogue-Act Taxonomy for a Virtual Coach Designed to Improve the Life of Elderly", Multimodal Technologies and Interaction. Vol. 3(3), pp. 52. MDPI. |
| Abstract: This paper presents a dialogue act taxonomy designed for the development of a conversational agent for elderly. The main goal of this conversational agent is to improve life quality of the user by means of coaching sessions in different topics. In contrast to other approaches such as task-oriented dialogue systems and chit-chat implementations, the agent should display a pro-active attitude, driving the conversation to reach a number of diverse coaching goals. Therefore, the main characteristic of the introduced dialogue act taxonomy is its capacity for supporting a communication based on the GROW model for coaching. In addition, the taxonomy has a hierarchical structure between the tags and it is multimodal. We use the taxonomy to annotate a Spanish dialogue corpus collected from a group of elder people. We also present a preliminary examination of the annotated corpus and discuss on the multiple possibilities it presents for further research. |
BibTeX:
@article{Montenegro_et_al:2019,
author = {C. Montenegro and A. Lopez-Zorrilla and J. Mikel-Olaso and R. Santana and R. Justo and J. A. Lozano and M. I. Torres},
title = {A Dialogue-Act Taxonomy for a Virtual Coach Designed to Improve the Life of Elderly},
journal = {Multimodal Technologies and Interaction},
publisher = {MDPI},
year = {2019},
volume = {3},
number = {3},
pages = {52},
url = {https://www.mdpi.com/2414-4088/3/3/52}
}
|
| Montenegro C, Santana R and Lozano JA (2019), "Data generation approaches for topic classification in multilingual spoken dialog system", In Proceedings of the 12th Conference on PErvasive Technologies Related to Assistive Environments Conference (PETRA-19). Rhodes, Greece , pp. 211-217. ACM. |
| Abstract: The conception of spoken-dialog systems (SDS) usually faces the problem of extending or adapting the system to multiple languages. This implies the creation of modules specifically for the new languages, which is a time consuming process. In this paper, we propose two methods to reduce the time needed to extend the SDS to other languages. Our methods are particularly oriented to the topic classification and semantic tagging tasks and we evaluate their effectiveness on topic classification for three languages: English, Spanish, French. |
BibTeX:
@inproceedings{Montenegro_et_al:2019a,
author = {C. Montenegro and R. Santana and J. A. Lozano},
title = {Data generation approaches for topic classification in multilingual spoken dialog system},
booktitle = {Proceedings of the 12th Conference on PErvasive Technologies Related to Assistive Environments Conference (PETRA-19)},
publisher = {ACM},
year = {2019},
pages = {211-217},
url = {https://dl.acm.org/doi/10.1145/3316782.3316792}
}
|
| Roman I, Santana R, Mendiburu A and Lozano JA (2019), "Sentiment analysis with genetically evolved Gaussian kernels", In Proceedings of the 2019 on Genetic and Evolutionary Computation Conference. Prague, Czech Republic , pp. 1328-1336. ACM. |
| Abstract: Sentiment analysis consists of evaluating opinions or statements based on text analysis. Among the methods used to estimate the degree to which a text expresses a certain sentiment are those based on Gaussian Processes. However, traditional Gaussian Processes methods use a predefined kernels with hyperparameters that can be tuned but whose structure can not be adapted. In this paper, we propose the application of Genetic Programming for the evolution of Gaussian Process kernels that are more precise for sentiment analysis. We use use a very flexible representation of kernels combined with a multi-objective approach that considers simultaneously two quality metrics and the computational time required to evaluate those kernels. Our results show that the algorithm can outperform Gaussian Processes with traditional kernels for some of the sentiment analysis tasks considered. |
BibTeX:
@inproceedings{Roman_et_al:2019,
author = {Ibai Roman and Roberto Santana and Alexander Mendiburu and Jose Antonio Lozano},
title = {Sentiment analysis with genetically evolved Gaussian kernels},
booktitle = {Proceedings of the 2019 on Genetic and Evolutionary Computation Conference},
publisher = {ACM},
year = {2019},
pages = {1328--1336},
url = {https://dl.acm.org/doi/10.1145/3321707.3321779}
}
|
| Roman I, Santana R, Mendiburu A and Lozano JA (2019), "Evolving Gaussian Process kernels for translation editing effort estimation", In Proceedings of the Learning and Intelligent Optimization Conference (LION). Chania, Greece , pp. 304-318. ACM. |
| Abstract: In many Natural Language Processing problems the combination of machine learning and optimization techniques is essential. One of these problems is estimating the effort required to improve, under direct human supervision, a text that has been translated using a machine translation method. Recent developments in this area have shown that Gaussian Processes can be accurate for post-editing effort prediction. However, the Gaussian Process kernel has to be chosen in advance, and this choice influences the quality of the prediction. In this paper, we propose a Genetic Programming algorithm to evolve kernels for Gaussian Processes. We show that the combination of evolutionary optimization and Gaussian Processes removes the need for a-priori specification of the kernel choice, and achieves predictions that, in many cases, outperform those obtained with fixed kernels. |
BibTeX:
@inproceedings{Roman_et_al:2019a,
author = {Ibai Roman and Roberto Santana and Alexander Mendiburu and Jose Antonio Lozano},
title = {Evolving Gaussian Process kernels for translation editing effort estimation},
booktitle = {Proceedings of the Learning and Intelligent Optimization Conference (LION)},
publisher = {ACM},
year = {2019},
pages = {304--318},
url = {https://link.springer.com/chapter/10.1007/978-3-030-38629-0_25}
}
|
| Roman I, Mendiburu A, Santana R and Lozano JA (2019), "Bayesian Optimization Approaches for Massively Multi-modal Problems", In International Conference on Learning and Intelligent Optimization (LION). , pp. 383-397. |
| Abstract: The optimization of massively multi-modal functions is a challenging task, particularly for problems where the search space can lead the optimization process to local optima. While evolutionary algorithms have been extensively investigated for these optimization problems, Bayesian Optimization algorithms have not been explored to the same extent. In this paper, we study the behavior of Bayesian Optimization as part of a hybrid approach for solving several massively multi-modal functions. We use well-known benchmarks and metrics to evaluate how different variants of Bayesian Optimization deal with multi-modality. |
BibTeX:
@inproceedings{Roman_et_al:2019b,
author = {Roman, Ibai and Mendiburu, Alexander and Santana, Roberto and Lozano, Jose A},
title = {Bayesian Optimization Approaches for Massively Multi-modal Problems},
booktitle = {International Conference on Learning and Intelligent Optimization (LION)},
year = {2019},
pages = {383--397},
url = {https://link.springer.com/chapter/10.1007/978-3-030-38629-0_31}
}
|
| Roman I, Santana R, Mendiburu A and Lozano JA (2019), "An Experimental Study in Adaptive Kernel Selection for Bayesian Optimization", IEEE Access. Vol. 7(8936460), pp. 184394-184302. IEEE Press. |
| Abstract: Bayesian Optimization has been widely used along with Gaussian Processes for solving expensive-to-evaluate black-box optimization problems. Overall, this approach has shown good results, and particularly for parameter tuning of machine learning algorithms. Nonetheless, Bayesian Optimization has to be also configured to achieve the best possible performance, being the selection of the kernel function a crucial choice. This paper investigates the convenience of adaptively changing the kernel function during the optimization process, instead of fixing it a priori. Six adaptive kernel selection strategies are introduced and tested in well-known synthetic and real-world optimization problems. In order to provide a more complete evaluation of the proposed kernel selection variants, two major kernel parameter setting approaches have been tested. According to our results, apart from having the advantage of removing the selection of the kernel out of the equation, adaptive kernel selection criteria show a better performance than fixed-kernel approaches. |
BibTeX:
@article{Roman_et_al:2019c,
author = {Roman, Ibai and Santana, Roberto and Mendiburu, Alexander and Lozano, Jose A},
title = {An Experimental Study in Adaptive Kernel Selection for Bayesian Optimization},
journal = {IEEE Access},
publisher = {IEEE Press},
year = {2019},
volume = {7},
number = {8936460},
pages = {184394--184302},
url = {https://ieeexplore.ieee.org/document/8936460}
}
|
| Santana R, Marti L and Zhang M (2019), "GP-based methods for domain adaptation: Using brain decoding across subjects as a test-case", Genetic Programming and Evolvable Machines. Vol. 20(3), pp. 385-411. Springer. |
| Abstract: Research on classifier transferability intends that the information gathered in the solution of a given classification problem could be reused in the solution of similar or related problems. We propose the evolution of transferable classifiers based on the use of multi-objective genetic programming and new fitness-functions that evaluate the amount of transferability. We focus on the domain adaptation scenario in which the problem to be solved is the same in the source and target domains, but the distribution of data is different between domains. As a real-world test case we address the brain decoding problem, whose goal is to predict the stimulus presented to a subject from the analysis of his brain activity. Brain decoding across subjects attempts to reuse the classifiers learned from some subjects in the classification of the others. We evolved GP-based classifiers using different variants of the introduced approach to test their effectiveness on data obtained from a brain decoding experiment involving 16 subjects. Our results show that the GP-based classifiers evolved trying to maximize transferability are able to improve classification accuracy over other classical classifiers that incorporate domain adaptation methods. Moreover, after comparing our algorithm to importance-weighted cross validation (in conjunction with many ML methods), we conclude that our approach achieves state of the art results in terms of transferability. |
BibTeX:
@article{Santana_et_al:2019,
author = {R. Santana and L. Marti and M. Zhang},
title = {GP-based methods for domain adaptation: Using brain decoding across subjects as a test-case},
journal = {Genetic Programming and Evolvable Machines},
publisher = {Springer},
year = {2019},
volume = {20},
number = {3},
pages = {385--411},
url = {https://link.springer.com/article/10.1007/s10710-019-09352-6}
}
|
| Torres ML, Olaso JM, Montenegro C, Santana R, Vazquez A, Justo R, Lozano JA, Schloegl S, Chollet G, Dugan N, Irvine M, Glackin N, Pickard C, Esposito A, Cordasco G, Troncone A, Petrovska-Delacretaz D, Mtibaa A, Hmani MA, Korsnes MS, Martinussen LJ, Escalera S, Palmero-Cantarino C, Deroo O, Gordeeva O, Tenerio-Laranga J, Gonzalez-Fraile E, Fernandez-Ruanova B and Gonzalez-Pinto A (2019), "The EMPATHIC Project: Mid-term Achievements", In Proceedings of the 12th Conference on PErvasive Technologies Related to Assistive Environments Conference (PETRA-19). , pp. 629-638. |
| Abstract: The goal of active aging is to promote changes in the elderly community so as to maintain an active, independent and socially-engaged lifestyle. Technological advancements currently provide the necessary tools to foster and monitor such processes. This paper reports on mid-term achievements of the European H2020 EMPATHIC project, which aims to research, innovate, explore and validate new interaction paradigms and platforms for future generations of personalized virtual coaches to assist the elderly and their carers to reach the active aging goal, in the vicinity of their home. The project focuses on evidence-based, user-validated research and integration of intelligent technology, and context sensing methods through automatic voice, eye and facial analysis, integrated with visual and spoken dialogue system capabilities. In this paper, we describe the current status of the system, with a special emphasis on its components and their integration, the creation of a Wizard of Oz platform, and findings gained from user interaction studies conducted throughout the first 18 months of the project. |
BibTeX:
@inproceedings{Torres_et_al:2019,
author = {M. L. Torres and J. M. Olaso and C. Montenegro and R. Santana and A. Vazquez and R. Justo and J. A. Lozano and S. Schloegl and G. Chollet and N. Dugan and M. Irvine and N. Glackin and C. Pickard and A. Esposito and G. Cordasco and A. Troncone and D. Petrovska-Delacretaz and A. Mtibaa and M. A. Hmani and M. S. Korsnes and L. J. Martinussen and S. Escalera and C. Palmero-Cantarino and O. Deroo and O. Gordeeva and J. Tenerio-Laranga and E. Gonzalez-Fraile and B. Fernandez-Ruanova and A. Gonzalez-Pinto},
title = {The EMPATHIC Project: Mid-term Achievements},
booktitle = {Proceedings of the 12th Conference on PErvasive Technologies Related to Assistive Environments Conference (PETRA-19)},
year = {2019},
pages = {629-638},
url = {https://dl.acm.org/doi/abs/10.1145/3316782.3322764}
}
|
| Vadillo J and Santana R (2019), "Universal Adversarial Examples in Speech Command Classification", CoRR. Vol. abs/1911.10182 |
| Abstract: Adversarial examples are inputs intentionally perturbed with the aim of forcing a machine learning model to produce a wrong prediction, while the changes are not easily detectable by a human. Although this topic has been intensively studied in the image domain, classification tasks in the audio domain have received less attention. In this paper we address the existence of universal perturbations for speech command classification. We provide evidence that universal attacks can be generated for speech command classification tasks, which are able to generalize across different models to a significant extent. Additionally, a novel analytical framework is proposed for the evaluation of universal perturbations under different levels of universality, demonstrating that the feasibility of generating effective perturbations decreases as the universality level increases. Finally, we propose a more detailed and rigorous framework to measure the amount of distortion introduced by the perturbations, demonstrating that the methods employed by convention are not realistic in audio-based problems. |
BibTeX:
@article{Vadillo_and_Santana:2019,
author = {Jon Vadillo and Roberto Santana},
title = {Universal Adversarial Examples in Speech Command Classification},
journal = {CoRR},
year = {2019},
volume = {abs/1911.10182},
url = {http://arxiv.org/abs/1911.10182}
}
|