PPGCC - Doutorado (Teses)
URI permanente para esta coleçãohttp://www.hml.repositorio.ufop.br/handle/123456789/9837
Navegar
5 resultados
Resultados da Pesquisa
Item Decision trees for the algorithm selection problem : integer programming based approaches.(2019) Vilas Boas, Matheus Guedes; Santos, Haroldo Gambini; Blum, Christian Clemens; Merschmann, Luiz Henrique de Campos; Silva, Rodrigo César Pedrosa; Toffolo, Túlio Ângelo MachadoEven though it is well known that for most relevant computational problems different algorithms may perform better on different classes of problem instances, most researchers still focus on determining a single best algorithmic configuration based on aggregate results such as the average. In this thesis, we propose Integer Programming based approaches to build decision trees for the Algorithm Selection Problem. These techniques allow the automation of three crucial decisions: (i) discerning the most important problem features to determine problem classes; (ii) grouping the problems into classes and (iii) select the best algorithm configuration for each class. We tested our approach from different perspectives: (i) univariate approach, where for each branch node, only one cutoff point of a feature is chosen and (ii) multivariate approach, where for each branch node, weights for multiple features are used (oblique decision trees). Considering the current scenario where the number of cores per machine has increased considerably, we also propose a new approach based on recommendation of concurrent algorithms. To evaluate our approaches, extensive computational experiments were executed using a dataset that considers the linear programming algorithms implemented in the COIN-OR Branch & Cut solver across a comprehensive set of instances, including all MIPLIB benchmark instances. We also conducted experiments with the scenarios/- datasets of the Open Algorithm Selection Challenge (OASC) held in 2017. Considering the first dataset and a 10-fold cross validation experiment, while selecting the single best solver across all instances decreased the total running time by 2%, our univariate approach decreased the total running time by 68% and using the multivariate approach, the total running time is decreased by 72%. An even greater performance gain can be obtained using concurrent algorithms, something not yet explored in the literature. For our experiments, using three algorithm configurations per leaf node, the total running time is decreased by 85%. These results indicate that our method generalizes quite well and does not overfit. Considering the results obtained using the scenarios of the OASC, the experimental results showed that our decision trees can produce better results than less interpretable models, such as random forest, which has been extensively used for algorithm recommendation.Item Mixed-integer linear programming based approaches for the resource constrained project scheduling problem.(2019) Araujo, Janniele Aparecida Soares; Santos, Haroldo Gambini; Santos, Haroldo Gambini; Barboza, Eduardo Uchoa; Souza, Marcone Jamilson Freitas; Jena, Sanjay Dominik; Toffolo, Túlio Ângelo MachadoResource Constrained Project Scheduling Problems (RCPSPs) without preemption are well-known NP-hard combinatorial optimization problems. A feasible RCPSP solution consists of a time-ordered schedule of jobs with corresponding execution modes, respecting precedence and resources constraints. First, in this thesis, we provide improved upper bounds for many hard instances from the literature by using methods based on Stochastic Local Search (SLS). As the most contribution part of this work, we propose a cutting plane algorithm to separate five different cut families, as well as a new preprocessing routine to strengthen resource-related constraints. New lifted versions of the well-known precedence and cover inequalities are employed. At each iteration, a dense conict graph is built considering feasibility and optimality conditions to separate cliques, odd-holes and strengthened Chvátal-Gomory cuts. The proposed strategies considerably improve the linear relaxation bounds, allowing a state-of-the-art mixed-integer linear programming solver to nd provably optimal solutions for 754 previously open instances of different variants of the RCPSPs, which was not possible using the original linear programming formulations.Item A proposal to improve wearables development time and performance : software and hardware approaches.(2019) Amorim, Vicente José Peixoto de; Oliveira, Ricardo Augusto Rabelo; Oliveira, Ricardo Augusto Rabelo; Aquino, André Luiz Lins de; Rodrigues, André Miguel; Ramos Filho, Heitor Soares; Martins, Henrique Resende; Nacif, José Augusto Miranda; Silva, Saul Emanuel DelabridaWearable devices are a trending topic in both commercial and academic areas. Increasing demand for innovation has raised the number of research and products, addressing brandnew challenges, and creating profitable opportunities. Current wearable devices can be employed in solving problems in a wide variety of areas. Such coverage generates a relevant number of requirements and variables that influences solutions performance. It is common to build specific wearable versions to fit each targeting application niche, what requires time and resources. Currently, the related literature does not present ways to treat the hardware/software in a generic way enough to allow both parts reuse. This manuscript presents the proposal of two components focused on hardware/software, respectively, allowing the reuse of di↵erent parts of a wearable solution. A platform for wearables development is outlined as a viable way to recycle an existing organization and architecture. The platform use was proven through the creation of a wearable device that was enabled to be used in di↵erent contexts of the mining industry. In the software side, a development and customization tool for specific operating systems is demonstrated. This tool aims not only to reuse standard software components but also to provide improved performance simultaneously. A real prototype was designed and created as a manner to validate the concepts. In the results, the comparison between the operating system generated by the tool versus a conventional operating system allows quantifying the improvement rate. The former operating system showed approximate performance gains of 100% in processing tasks, 150% in memory consumption and I/O operations, and approximately 20% of reduction in energy consumption. In the end, performance analysis allows inferring that the proposals presented here contribute to this area, easing the development and reuse of wearable solutions as a whole.Item Exploring deep learning representations for biometric multimodal systems.(2019) Luz, Eduardo José da Silva; Gomes, David Menotti; Moreira, Gladston Juliano Prates; Ferreira, Anderson Almeida; Moreira, Gladston Juliano Prates; Gomes, David Menotti; Cavalin, Paulo; Cámara Chávez, Guillermo; Santos, Thiago Oliveira dosBiometrics is an important area of research today. A complete biometric system comprises sensors, feature extraction, pattern matching algorithms, and decision making. Biometric systems demand high accuracy and robustness, and researchers are using a combination of several biometric sources, two or more algorithms for pattern matching and di↵erent decision-making systems. These systems are called multimodal biometric systems and today represent state-of-the-art for biometrics. However, the process of extracting features in multimodal biometric systems poses a major challenge today. Deep learning has been used by researchers in the machine learning field to automatize the feature extraction process and several advances were achieved, such as the case of face recognition problem. However, deep learning based methods require a large amount of data and with the exception of facial recognition, there are no databases large enough for the other biometric modalities, hindering the application of deep learning in multimodal methods. In this thesis, we propose a set of contributions to favor the use of deep learning in multimodal biometric systems. First of all, we explore data augmentation and transfer learning techniques for training deep convolution networks, in restricted biometric databases in terms of labeled images. Second, we propose a simple protocol, aiming at reproducibility, for the creation and evaluation of multimodal (or synthetic) multimodal databases. This protocol allows the investigation of multiple biometric modalities combination, even for less common and novel modalities. Finally, we investigate the impact of merging multimodal biometric systems in which all modalities are represented by means of deep descriptors. In this work, we show that it is possible to bring the expressive gains already obtained with the face modality, to other four biometric modalities, by exploring deep learning techniques. We also show that the fusion of modalities is a promising path, even when they are represented by means of deep learning. We advance state-of-the-art for important databases in the literature, such as FRGC (periocular region), NICE / UBIRIS.V2 (periocular region and iris), MobBio (periocular region and face), CYBHi (o↵-the-person ECG), UofTDB (o↵-the-person ECG) and Physionet (EEG signal). Our best multimodal approach, on the chimeric database, resulted in the impressive decidability of 9.15±0.16 and a perfect recognition in (i.e., EER of 0.00%±0.00) for the intra-session multimodal scenario. For inter-session scenario, we reported decidability of 7.91±0.19 and an EER of 0.03%±0.03, which represents a gain of more than 22% for the best inter-session unimodal case.Item Smart Wearable : let's think wearable from a new perspective.(2018) Silva, Saul Emanuel Delabrida; Rabelo, Ricardo Augusto Oliveira; Rabelo, Ricardo Augusto Oliveira; Loureiro, Antonio Alfredo Ferreira; Thomas, Bruce H.; Silva, Jorge Sá; Garcia, Luis Guilherme Uzeda; Billinghurst, MarkWearable computing is considered the next big thing. The market represents billions of dollars of investments. Samsung, Google, and other companies aggregate part of this market with the release of new gadgets such as smartwatches and smartglasses. Other startups emerged in the last years and made available devices such as Fitbit and ReconJet. On the other hand, users are losing the interest on wearable devices due to the low quality of applications and the restricted value o ered by them. We did a study about the role of wearable devices and its integration. In our ndings, the dependency and low processing capabilities of the devices contribute to the scenario. The goal of this work is to introduce a new way to build and design wearable devices and systems. We propose the de nition of the terms \Wearable Accessory" and \Smart Wearable" as a new classi cation of wearable devices, systems, and applications. Our characterization was done by the development of two smart wearable devices. In our methodology, we demonstrate the design until the user experience evaluation to build these devices. Our ndings consider some current limitation of these devices and users expectations. Two applications addressed to the ecological scenarios are the secondary contribution of this work. Also, challenges and future work to improve the proposed concept and user experience with wearable devices are discussed.