Brazil

In probabilistic planning, an usual problem is the Stochastic Shortest Path (SSP) in a Markov Decision Process (MDP), where the objective is to find policies that reach a goal with the minimum expected accumulated cost. In the presence of states from where it is not possible to reach a goal, this criterion is not well-defined in these states. Because of that, there are extensions and other formulations for this particular case, like the GUBS criterion. Although algorithms to solve SSPs under this criterion exist, they need to make sweeps through the entire state space. With that in mind, this work proposes UCT-GUBS: an online planning algorithm based on UCT to solve SSPs under the GUBS criterion. The results of the experiments that were performed show empirically that the algorithm makes the desired trade-off between cost and probability to goal when its parameters are varied, with the ability of making a decision in less or more time depending on the time limit provided to the algorithm, an important characteristic of UCT.

BRACIS 2020

Monte Carlo Tree Search Algorithm for SSPs Under the GUBS Criterion

gubs

stochastic shortest path

probabilistic planning

markov decision processes

The 9th Brazilian Conference on Intelligent Systems (BRACIS), the 16h Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), and the 12th Concurso de Teses e Dissertações em Inteligência Artificial e Computacional (CTDIAC) was held from 20 to 23 October, 2020.

BRACIS originated from the combination of the two most important scientific events in Brazil in Artificial Intelligence (AI) and Computational Intelligence (CI): the Brazilian Symposium on Artificial Intelligence - SBIA (22 editions), and the Brazilian Symposium on Neural Networks - SBRN (13 editions). ENIAC is a national forum for researchers, practitioners, educators and students to present and discuss innovations, trends, experiences, developments and work in progress in the fields of Artificial and Computational Intelligence. CTDIAC is a satellite event whose goals are to disseminate and award the best dissertations and theses (MSc and PhDs concluded in Brazilian universities from June 2018 to May 2020) in Artificial and Computational Intelligence areas.

In this year, the Symposium on Knowledge Discovery, Mining and Learning (KDMiLe) will be held in conjunction with BRACIS 2020. This sysposium aims at integrating researchers, practitioners, developers, students and users to present their research results, to discuss ideas, and to exchange techniques, tools, and practical experiences – related to the Data Mining and Machine Learning areas.

technical paper

COVID-19 has caused thousands of deaths worldwide and undermined the economy of those countries which have been ravaged by the pandemic.
In this paper, we present an agent-based simulation to study the propagation of
COVID-19 in Ibirama (SC). The simulation uses real population, educational,
and business data, as well as the actual map of the city containing its traffic
network and location of educational institutions and businesses. The simulation
considers students and workers agents, which move around the environment and
interact with each other. Different social distancing scenarios of these agents
were simulated. The results obtained indicate that social distancing policies are
effective in reducing the spread of the virus.

An Agent-based Simulation to Study the Spread of COVID-19in Ibirama(SC)

K-Shortest Loopless Paths (KSLP) consiste em encontrar K | K > 1 caminhos, sem repetições de vértices, classificados em ordem crescente de distância, em um grafo G(V,E), no qual V representa o conjunto de vértices e E o conjunto dos arestas que os conecta. Dada a dificuldade de obter boa performance para solução do problema, este trabalho propõe a paralelização em GPUs CUDA do algoritmo mais veloz existente na literatura para o problema.

Proposta de paralelização em GPUs CUDA do algoritmo MPS para resolução do problema de encontrar os K-Caminhos Mais Curtos Sem Repetições de Vértices em grafos direcionados e não direcionados

This paper describes the application of supervised and unsupervised
machine learning in large volumes of open governmental data from INEP.
This work uses the following algorithms: K-Nearest Neighbors, Logistic
Regression, Decision Tree, Random Forest and K-means. The methodology is
based on the CRISP-DM and KDD processes, requiring the use of the
DataBricks cloud platform. In addition, the Hadoop and Apache Spark cluster
Technologies were also used. Such technologies provided high processing
power for the execution of the experiments. This enabled the performance
evaluation of the models and the discovery of knowledge about Brazilian basic
education.

Cloud Computing and Machine Learning for Analysis ofLarge Volumes of Educational Data

This paper presents a parallel strategy with a heuristic approach to
reduce the execution time bottleneck of a routing and wavelength assignment
problem in wavelength-division multiplexing networks of a previous work that
uses a sequential genetic algorithm. As the parallelization solution, the GPU
hardware processing on CUDA architecture and CUDA C programming lan-
guage were adopted. The results achieved were between 35 and 40 times faster
than the sequential version of the genetic algorithm.

A Parallel Strategy for a Genetic Algorithm in Routing Wavelength Assignment Problem Using GPU with CUDA

Multiple coin detection is a Computer Vision problem which aims at finding coins in an image and classifying their respective values. For Brazilian Real, literature typically consider feature extraction techniques followed by Machine Learning classifiers, specially Artificial Neural Networks. However, some recent results on coin detection with Deep Learning motivated us to propose a hybrid approach for the previously mentioned problem considering localization with traditional Computer Vision techniques and classification with Convolutional Neural Networks. Multiple architectures were tested in a realistic dataset and the proposed solution has mean average precision higher than 92%with classifier accuracy greater than 97% surpassing other work for the same dataset by 6%.


Hybrid Approach for Detecting Brazilian Real Coins with Localization Algorithms and Convolutional Neural Networks

Data classification is a major machine learning paradigm, which has been widely applied to solve a large number of real-world problems. Traditional data classification techniques consider only physical features (e.g., distance, similarity, or distribution) of the input data. For this reason, those are called \textit{low-level} classification. On the other hand, the human (animal) brain performs both low and high orders of learning and it has a facility in identifying patterns according to the semantic meaning of the input data. Data classification that considers not only physical attributes but also the pattern formation is referred to as \textit{high-level} classification. Several high-level classification techniques have been developed, which make use of complex networks to characterize data patterns and have obtained promising results.
In this paper, we propose a pure network-based high-level classification technique that uses the betweenness centrality measure.
We test this model in nine different real datasets and compare it with other nine traditional and well-known classification models. The results show us a competent classification performance.

A Network-Based High-Level Data Classification Algorithm Using Betweenness Centrality

Many works that apply Deep Neural Networks (DNNs) to Speech Emotion Recognition (SER) use single datasets or train and evaluate the models separately when using multiple datasets. Those datasets are constructed with specific guidelines and the subjective nature of the labels for SER makes it difficult to obtain robust and general models. We investigate how DNNs learn shared representations for different datasets in both multi-task and unified setups. We also analyse how each dataset benefits from others in different combinations of datasets and popular neural network architectures. We show that the long-standing belief of more data resulting in more general models doesn’t always hold for SER, as different dataset and meta-parameter combinations hold the best result for each of the analysed datasets.

Interaffection of Multiple Datasets with Neural Networks in Speech Emotion Recognition

Machine learning algorithms are widely used in credit card fraud detection systems due to their ability to distinguish between legitimate and fraudulent transactions. A known problem in this field is the high imbalance usually found in the classes, which can compromise the performance of the classifiers. The empirical studies found in the literature apply, at most, two sampling techniques. This article presents a comparative study of five classification models under five different methods of balancing the training sets. The best performance was obtained by random forest, which in addition to having the highest average F-score (0.867), proved to be considerably more robust than the other classifiers in relation to the choice of the balancing technique and attribute selection.

Empirical evaluation of classifiers and balancing methods for fraud detection in credit card transactions

Non-traumatic Vertebral Compression Fractures (VCFs) are generally caused by osteoporosis (benign VCFs) or metastatic cancer (malignant VCFs) and the success of the medical treatment strongly depends on a fast and correct classification of VCFs. Recently, methods for computer-aided diagnosis (CAD) based on machine learning have been proposed for classifying VCFs. In this work, we investigate the problem of clustering images of VCFs and the impact of feature selection by genetic algorithms on this task. The analysis of the clusters helps to understand the results of classifiers and difficulties of differentiating images of different classes by an expert. The results indicate that features selection improved the separability of clusters and purity. Feature selection also helps to understand which attributes are more important for analysing the images of vertebral bodies.

Impact of Feature Selection on Clustering Images of Vertebral Compression Fractures

This work presents an investigation on how to define Neural Networks (NN) architectures adopting a data-driven approach using clustering to create sub-labels to facilitate the learning process and to discover the number of neurons needed to compose the layers. We also increase the depth of the model aiming to represent the samples better, the more in-depth it flows into the model. We hypothesize that the clustering process identifies sub-regions in the feature space in which the samples belonging to the same cluster have strong similarities. We used seven benchmark datasets to validate our hypothesis using 10-fold cross validation 3 times. The proposed model increased the performance, while never decreased it, with statistical significance considering the p-value < 0.05 in comparison with a Multi-Layer Perceptron with a single hidden layer with approximately the same number of parameters of the architectures found by our approach.

Downloads

Next from BRACIS 2020

An Agent-based Simulation to Study the Spread of COVID-19in Ibirama(SC)

Similar lecture

Finding Feasible Policies for Extreme Risk-Averse Agents in Probabilistic Planning

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from BRACIS 2020

An Agent-based Simulation to Study the Spread of COVID-19in Ibirama(SC)

Similar lecture

Finding Feasible Policies for Extreme Risk-Averse Agents in Probabilistic Planning

Downloads