Brazil

Assessing the hardness of each instance in a problem is an important
meta-knowledge which may leverage advances in Machine Learning.
In classification problems, an instance can be regarded as difficult
if it gets systematically misclassified by a diverse set of classification
techniques with different biases. The instance hardness measures were
proposed with the aim of relating data characteristics to this notion of
intrinsic difficulty of the instances. There are also in the literature a large
set of measures which are dedicated at describing the difficulty of a classification
problem from a dataset-level perspective. In this paper these
measures are decomposed at the instance-level, giving a perspective of
how each individual example in a dataset contributes to its overall complexity.
Experiments on synthetic and benchmark datasets demonstrate
the proposed measures can provide a complementary instance hardness
perspective when compared to those from related literature.

BRACIS 2020

Measuring instance hardness using data complexity measures

machine learning. instance hardness. data complexity

**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. **STIL** is the bi-annual Language Technology event.
All events are supported by the Brazilian Computer Society – Special Interest Groups: **CE-IA**, **CE-IC** and **CE-PLN**.

To access the 10th Brazilian Conference on Intelligent Systems (**BRACIS**), you need to register for the conference.  Please visit the Conference official site to register

[Register now](http://c4ai.inova.usp.br/bracis2021/)

Register now

BRACIS 2021

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. 
STIL is the bi-annual Language Technology event.
All events are supported by the Brazilian Computer Society – Special Interest Groups: CE-IA, CE-IC and CE-PLN

technical paper

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.

The increasingly relevance of data streams in the context of machine learning and artificial intelligence has motivated this paper which discusses and draws necessary relationships between the concepts of data streams and time series in attempt to build on theoretical foundations to support online learning in such scenarios. We unify the concepts of data streams and time series by assessing their definitions in the literature and discuss the major implications of this claim on the way that data streams research and practice is carried out, showing that many common assumptions are incorrect or unnecessary. We analyzed six data sources typically used in benchmark data-stream classification and found that none of those meet the requirements and assumptions qualifying them for online learning.

Data Streams are Time Series: Challenging Assumptions

Given the most recent events involving the fast spreading of COVID-19, policy makers around the world have been challenged with the difficult task of developing efficient strategies to contain the dissemination of the disease among the population, sometimes by taking severe measures to restrict local activities, both socially and economically. Within this context, models which can help on predicting the spread evolution of COVID-19 in a specific region would surely help the authorities on their planning. In this paper, we introduce a semi-supervised regression model which makes use of a correlations-based temporal network, by considering the evolution of COVID-19 in different world regions, in order to predict the evolution of new confirmed cases and deaths in 27 federal units of Brazil. In this approach, each node in the network represents the COVID-19 time series in a specific region, and the edges are created according to the variations similarity between each pair of nodes, at each new time step. The results obtained, by predicting the weekly new confirmed cases and deaths in each region, are promising, with a median and mean absolute percentage error of 21% and 24%, respectively, when predicting new cases, and a median and mean absolute percentage error of 16% and 23%, respectively, when predicting new deaths, for the considered period.

Predicting the Evolution of COVID-19 Cases and Deaths Through a Correlations-Based Temporal Network

The source code similarity problem consists in defining iftwo given distinct codes are the same program or not. This problem isvaluable for polymorphic malware detection, which can generate distinctversions of code applying rules of obfuscation that change the originalcode. A technique to measure the similarity between source codes isto model the codes as program dependency graphs (PDG) and find analignment between the graphs that maximizes a similarity function. Thiswork investigates the performance of distinct similarity metrics applied toa genetic algorithm (GA). To distinguish between similar codes and non-similar codes, we use the k-nearest-neighbors (KNN) algorithm based onthe similarity of the alignment found by the GA. The experiments areconducted with a database proposed by this work, where the source codeswere retrieved from the Codeforces website and obfuscated by the toolCXX-obfuscator.

KNN applied to PDG for source code similarityclassification

This presentation aims to provide a summary of the work done on the related paper, using media content to illustrate its approach and results. By its end, the audience should understand the steps taken to transform tweets into Time-Varying-Graphs, and then into a Time Series, to find changes in its users' behavior.

Quantifying Temporal Novelty in Social Networks using Time-Varying Graphs and Concept Drift Detection

Community detection in networks is a hard to solve combinatorial optimization problem with large-scale applications such as those based on flight networks. To approach this problem, a number of heuristics can be found in the literature, among them we highlight multilevel algorithms which are very efficient in detecting communities in larger networks. In this paper, we propose an improvement of a multilevel community detection algorithm that aims at finding communities so as to maximize the well-known modularity measure. The proposed improvement is the inclusion of a refinement phase to enhance the modularity value in flight networks we compiled. We presented a study case on the communities found for real flight networks. The vertices of the constructed networks correspond to airports in Europe, South America, United States of America and Canada and the arcs represent the flights whose weights indicate the number of flights between pairs of destinations. Experiments with these networks pointed to an improvement in the modularity of the communities found by the proposed algorithm when contrasted with its original version and the Lovain method. We visually identified consistent communities in the flight networks. Besides, in comparison to other reference algorithms on benchmark networks, ours was very competitive.

Improved Multilevel Algorithm to Detect Communities in Flight Networks

This work proposes a new approach to segmentation of hemorrhagic stroke (HS) based on morphological geodesic active contour method, with automatic initialization near to lesion region, without previous training, called Adaptive Morphological Geodesic Active Contour (Ada-MGAC). To evaluate the performance, we used 100 computed tomography images with HS of volunteers. These samples were compared against segmentation methods from specialized literature. A manual segmentation from a medical specialist was considered as the gold standard. The results indicate a competitive potential of Ada-MGAC and method showed a mean convergency time about 3 seconds, indicating a fast result to medical analysis. Thus, it is possible to conclude that the proposed approach can be used to aid medical diagnosis in the cerebral vascular accident.

New Adaptive Morphological Geodesic Active Contour Method for Segmentation of Hemorrhagic Stroke in Computed Tomography Image

Forecasting stock market prices is a hard task. The main reason for that is due to the fact that its environment is highly dynamic, intrinsically complex, and chaotic. Traditional economic theories suggest that trying to forecast short-term stock price movements is a wasted effort because the market is influenced by several external events and its behavior approximates a random walk. Recent studies that address the problem of stock market forecasting usually create specific prediction models for the price behavior of a single stock. In this work we propose a technique to predict price movements based on similar stock sets. Our goal is to build a model to identify whether the price tends to bullishness or bearishness in the near future, considering stock information from similar sets based on two sources of information: historical stock data and Google Trends news. Firstly, the proposed study applies a method to identify similar stock sets and then creates a forecasting model based on a LSTM (long short-term memory) for these sets. More specifically, two experiments were conducted: (1) using the K-Means algorithm to identify similar stock sets and then using a LSTM neural network to forecast stock price movements for these stock sets; (2) using the DBSCAN (Density-based spatial clustering) algorithm to identify similar stock sets and then using the same LSTM neural network to forecast stock price movements. The study was conducted over 51 stocks of the Brazilian stock market. The results show that the use of an algorithm to identify stock clusters yields an improvement of approximately 7% in accuracy and f1-score and 8% in recall and precision when compared to models for a single stock

Stocks Clustering Based on Textual Embeddings for Price Forecasting

The use of Semi-supervised Learning (SSL) methods have emerged as an efficient solution to smooth out the problem of availability of labelled instances. Several methods have been proposed in the literature and Self-training and Co-training are two well-known methods. The main aim is to use only a few labelled instances to define a model and to apply this model in a labelling process, in which unlabelled instances are labelled and included in the labelled set. However, the labelling process is always directly dependent on the selection of the unlabelled instances. Moreover, the selection criterion used to select and label new instances has an important effect in the performance of a semi-supervised method. In this paper, we propose a distance-weighted selection of unlabelled instances for Self-training and Co-training semi-supervised methods. In addition, we compare the standard Self-training and Co-training methods against the proposed versions of these two methods over 20 classification datasets.

A Distance-weighted Selection of Unlabelled Instances for Self-training and Co-training Semi-supervised Methods

Monte Carlo Tree Search (MCTS) parallelization is one of the many possible enhancements for MCTS algorithms. However, no work had been done on evaluating these methods in the rather new area of General Video Game Playing (GVGP), an area that challenges the creation of agents that can play any videogame even without prior knowledge about the video game they are going to play. 
In this presentation, we show how we have approached this research gap with the implementation of three different parallel MCTS agents for GVGP as well as the results we obtained from performing a set of experiments over them.

Parallel Monte Carlo Tree Search in General Video Game Playing

Despite the availability of benchmark machine learning (ML) repositories (e.g., UCI, OpenML), there is no standard evaluation strategy yet capable of pointing out which is the best set of datasets to serve as gold standard to test different ML algorithms. In recent studies, Item Response Theory (IRT) has emerged as a new approach to elucidate what should be a good ML benchmark. This work applied IRT to explore the well-known OpenML-CC18 benchmark to identify how suitable it is on the evaluation of classifiers. Several classifiers ranging from classical to ensembles ones were evaluated using IRT models, which could simultaneously estimate dataset difficulty and classifiers' ability. The Glicko-2 rating system was applied on the top of IRT to summarize the innate ability and aptitude of classifiers. It was observed that not all datasets from OpenML-CC18 are really useful to evaluate classifiers. Most datasets evaluated in this work (84%) contain easy instances in general (e.g., around 10% of difficult instances only). Also, 80% of the instances in half of this benchmark are very discriminating ones, which can be of great use for pairwise algorithm comparison, but not useful to push classifiers abilities. This paper presents this new evaluation methodology based on IRT as well as the tool decodIRT, developed to guide IRT estimation over ML benchmarks.

Measuring instance hardness using data complexity measures

Downloads

Next from BRACIS 2020

Data Streams are Time Series: Challenging Assumptions

Similar lecture

On the impact of MDP design for Reinforcement Learning agents in Resource Management

Measuring instance hardness using data complexity measures

.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

Data Streams are Time Series: Challenging Assumptions

Similar lecture

On the impact of MDP design for Reinforcement Learning agents in Resource Management

Downloads