Brazil

DevOps practices have been increasingly used by software engineering teams in order to improve development stages. In processes that involve machine learning (ML), DevOps can also be applied to deploy machine learning models into production - a practice also known as MLOps. This systematic mapping aims to understand how DevOps has been applied to machine learning processes and what challenges are faced. Most of the 15 articles selected make use of CI/CD practices and propose architectures for the implementation of ML models. The biggest challenges are the characteristics inherent to the ML models and resistance to change.

BRACIS 2021

Applying DevOps to Machine Learning Processes: A Systematic Mapping

systematic mapping; ci/cd

devops

machine learning

**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 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

Events are phenomena that occur at a specific time and place. Its detection can bring benefits to society since it is possible to extract knowledge from these events. Event detection is a multimodal task since these events have textual, geographical, and temporal components. Most multimodal research in the literature uses the concatenation of the components to represent the events. These approaches use multi-class or binary learning to detect events of interest which intensifies the user's labeling effort, where the user should label event classes even if there is no interest in detecting them. In this paper, we present the Triple-VAE approach that learns a unified representation from textual, spatial, and density modalities through a variational autoencoder, which is one of the state-of-the-art in representation learning. Our proposed Triple-VAE obtains suitable event representations for one-class classification, where users provide labels only for events of interest, thereby reducing the labeling effort. We carried out an experimental evaluation with ten real-world event datasets, four multimodal representation methods, and five evaluation metrics. Triple-VAE outperforms and presents a statistically significant difference considering the other three representation methods in all datasets. Therefore, Triple-VAE proved to be promising to represent the events in the one-class event detection scenario.

Triple-VAE: A Triple Variational Autoencoder to Represent Events in One-Class Event Detection

Wikipedia is an important free source of intelligible knowledge. Despite that, Brazilian Portuguese Wikipedia still lacks descriptions for many subjects. In an effort to expand the Brazilian Wikipedia, we contribute PLSum, a framework for generating wiki-like abstractive summaries from multiple descriptive websites. The framework has an extractive stage followed by an abstractive one. In particular, for the abstractive stage, we fine-tune and compare two recent variations of the Transformer neural network, PTT5, and Longformer. To fine-tune and evaluate the model, we created a dataset with thousands of examples, linking reference websites to Wikipedia. Our results show that it is possible to generate meaningful abstractive summaries from Brazilian Portuguese web content.

PLSUM: Generating PT-BR Wikipedia by Summarizing Multiple Websites

In this work we propose a selection Hyper-Heuristic (HH) based onthe Thompson Sampling (TS) approach to the solution of Quadratic AssignmentProblem (QAP). The QAP aims to allocate facilities in a set of possible knownlocations, in order to minimize the total cost of all movements between facili-ties. The proposed HH is applied in the automatic configuration of a memeticalgorithm, acting on the selection of a combination of low-level heuristics. Eachcombination involves the selection of a recombination heuristic, a local searchstrategy and a mutation heuristic. The algorithm was tested in 15 instances ofNug benchmark and the performance of HH is superior to that obtained by anycombination of heuristics applied in isolation, demonstrating its efficiency inthe automatic configuration of the algorithm. Experiments show that TS per-formance is affected by the quality of the set of low-level heuristics. The bestversion of HH obtains the optimal solution in 9 instances and the mean percen-tage deviation of the optimal solution (gap), considering all 15 instances, was 8.6%, with the largest gaps found for the three largest instances.

Thompson Sampling in Heuristic Selection for the Quadratic Assignment Problem

The universal approximation theorem ensures that any continuous real-valued function defined on a compact subset can be approximated with arbitrary precision by a single hidden layer neural network. Similarly, Arena et al. proved in 1996 that any continuous quaternion-valued function on a compact can be approximated with arbitrary precision by a single hidden layer quaternion-valued neural network whose hidden units are equipped with the so-called split activation function. Quaternions are four-dimensional hypercomplex numbers widely used to describe spatial rotations. Like quaternions, tessarines are four-dimensional hypercomplex numbers. However, in contrast quaternion algebra, the product of tessarines is commutative. The commutativity of tessarine's product may be advantageous for the design of neural networks because synaptic weights and neuron inputs can interchange their roles. In this paper, we show that the universal approximation theorem also holds for tessarine-valued neural networks. Precisely, we show that that any continuous tessarine-valued function can be approximated with arbitrary precision by a single hidden layer tessarine-valued neural network with split activation functions in the hidden layer. A simple numerical example, confirming the theoretical result and revealing the superior performance of a tessarine-valued neural network over a real-valued model for interpolating a vector-valued function, is presented in the paper.

Universal Approximation Theorem for Tessarine-Valued Neural Networks

Many image processing and analysis tasks are performed with deep neural networks. Although the vast majority of advances have been made with real numbers, recent works have shown that complex and hypercomplex-valued networks may achieve better results. In this paper, we address quaternion-valued and introduce tessarine-valued deep neural networks, including tessarine-valued 2D convolutions. We also address initialization schemes and hypercomplex batch normalization. Finally, a tessarine-valued ResNet model with hypercomplex batch normalization outperformed the corresponding real and quaternion-valued networks on the CIFAR dataset.

Tessarine and Quaternion-Valued Deep Neural Networks for Image Classification

Process mining explores event logs to offer valuable insights to business process managers. Some types of business processes are hard to mine, including unstructured and knowledge-intensive processes. Then, trace clustering is usually applied to event logs aiming to break it into sublogs, making it more amenable to the typical process mining task. However, applying clustering algorithms involves decisions, such as how traces are represented, that can lead to better results. In this paper, we compare four vector space models for trace clustering, using them with an agglomerative clustering algorithm in synthetic and real-world event logs. Our analyses suggest the embeddings-based vector space model can properly handle trace clustering in unstructured processes.

Vector space models for trace clustering: a comparative study

Morphological neural networks represent a class of artificial neural networks whose neurons perform an operation from mathematical morphology followed by the application of an activation function. This paper provides a comparative study of different approaches that use morphological neural networks. Specifically, according to the training rule, we review incremental approaches, approaches based on maximum descent methods, extreme learning machines, and convex-concave optimization procedures. Computational experiments showed that, on average, the reduced dilation-erosion perceptron with bagging and ensemble strategies had better results in several binary classification problems.

A Comparative Study on Morphological Neural Networks for Binary Classification

O objetivo desse trabalho é analisar conjuntos de variáveis que podem ter mais influência na previsão do resultado de uma partida de futebol, utilizando técnicas como cálculos de probabilidade e algoritmos de previsão, com a intenção de obter lucros em apostas. As técnicas utilizadas para o desenvolvimento foram redes bayesianas, com o algoritmo Naïve Bayes, e a de probabilidade, baseada no cálculo de Poisson. Os dados utilizados para treinamento foram do Campeonato Brasileiro, entre 2010 a 2017, sendo considerados os dados dos anos de 2018 e 2019 para testes. Os principais resultados atingidos foram de 53% de acerto do resultado de uma partida e as principais variáveis envolvidas foram força de ataque e força de defesa.

Análise de Variáveis em Partidas de Futebol: Previsão de Resultados com Naïve Bayes e Poisson

The Sustainable Development Goals (SDGs) are part of a global effort to reduce the impacts of climate change, promoting social justice and economic growth. 
The United Nations provides a database with hundreds of indicators to track the SDGs since 2016 for a total of 302 regions. 
This work aims to assess which countries are in a similar situation regarding sustainable development. 
Principal Component Analysis was used to reduce the dimension of the dataset and k-means algorithm was used to cluster countries according to their SDGs indicators. For the years of 2016, 2017 and 2018 were obtained 11, 13 and 11 groups, respectively. This paper also analyses clusters changes throughout the years.

A Comparative Analysis of Countries' Performance According to SDG Indicators based on Machine Learning

Companies created for money-laundering or as a means for tax-evasion are harmful to the country’s economy and society. This problem isusually tackled by governmental agencies by having officials to pore over com-panies’ financial data and to single out those that exhibit fraudulent behavior. Such work tends to be slow-paced and tedious. This paper proposes a machinelearning-based system capable of classifying whether a company is likely to beinvolved in fraud or not. Based on financial and tax data from various compa-nies, four different classifiers –k-Nearest Neighbors, Random Forest, SupportVector Machine (SVM), and a Neural Network – were trained and then used toindicate fraud. The best-performing model achieved a macro-averaged F1-scoreof 92.98% with the Random Forest.

Premium content

Applying DevOps to Machine Learning Processes: A Systematic Mapping

Downloads

Next from BRACIS 2021

Triple-VAE: A Triple Variational Autoencoder to Represent Events in One-Class Event Detection

Similar lecture

Iris-CV: Classifying Iris Flowers Is Not as Easy as You Thought

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES