Brazil

O objetivo do presente trabalho é investigar o desempenho de agentes jogadores baseados em aprendizado por reforço - mais especificamente, nos algoritmos Q-Learning e Deep Q-Networks (DQN) - por meio da plataforma Unity. Para tanto, os autores implementam nela agentes jogadores de Basic e GridWorld que são treinados segundo tais algoritmos. A fim de avaliar o desempenho desses agentes, foi efetuada uma análise comparativa entre eles e os melhores agentes jogadores desses jogos disponibilizados na plataforma Unity, os quais são baseados nas técnicas de aprendizagem Proximal Policy Optimization (PPO) e Soft Actor-Critic (SAC).

BRACIS 2021

Técnicas de Aprendizado por Reforço Aplicadas em Jogos Eletrônicos na Plataforma Geral do Unity

aprendizado por reforço

unity

The 2020 AAPT Virtual Summer Meeting brought together physics educators from around the globe to learn about the latest in physics theory, research methodology and education from leaders in their field. The program featured plenary speakers, sessions, workshops and many special activities.

For more information please visit the 2020 AAPT Virtual Summer Meeting page at: https://www.aapt.org/Conferences/SM2020/index.cfm.


AAPT SM 20

The 2020 AAPT Virtual Summer Meeting brought together physics educators from around the globe to learn about the latest in physics theory, research methodology and education from leaders in their field. 

technical paper

**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

Esta apresentação demonstra uma nova metodologia de incorporação de preferências em problemas multiobjetivo através da definição de uma região de interesse na forma de um hiper-elipsoide. A adição dessa camada ao algoritmo de otimização visa auxiliar a tomada de decisão, selecionando soluções de acordo com as preferências definidas pelo tomador de decisão. Além disso, a forma hiper-elipsoide permite que o usuário altere a quantidade de elementos selecionados ao se modificar o foco do hiper-elipsoide, ampliando ou reduzindo a área de cobertura da fronteira Pareto.

Definição de regiões de interesse em problemas multiobjetivo utilizando hiper-elipses

Um conjunto de estratégias automatizadas de negociação em mercados de capitais pode ser disposto em um portfólio buscando maximização de rendimentos e minimização de perdas. A melhor combinação para o portfólio requer a atribuição de pesos ótimos para cada estratégia, considerando diferentes indicadores financeiros. Neste artigo é proposta a aplicação de um algoritmo evolutivo com abordagem lexicográfica para otimizar um portfólio de estratégias automatizadas aplicadas ao mercado futuro brasileiro. Os experimentos consideram diferentes funções-objetivo de indicadores financeiros, com diferentes ordenações, além de condições de otimização e de variações temporais, aplicando dados históricos de minicontratos do índice futuro do Ibovespa e do dólar.

Otimização evolutiva lexicográfica de um portfólio de estratégias automatizadas no mercado futuro brasileiro

Emotions are one of the pillars of human communication, especially spoken language. In emotional speech, they can be identified by inherent attributes of the voice, such as pitch, frequency, intensity etc. In this paper, a model based on artificial data augmentation and Deep Learning, more specifically a Convolutional Recurrent Neural Network, was proposed to automate this emotion identification task by being trained on the RAVDESS database with cross-validation technique. Evaluated by the accuracy and F1-Score metrics, the model achieved in them 76.25% and 76% on average and a maximum of 83.33% and 80%, respectively, which are slightly better results than those presented in related research.

Identification of Emotions in Spoken Language Using Deep Learning

In Brazil, citizens can report irregularities in the Public Administration. The classification of these complaints needs information that is not in their texts. this article aims is to propose a methodology for extracting and enriching information identified in complaints texts. This methodology provides as output a set of structured data capable of characterizing the complaints. To validate the proposal, a case study was done. The study showed that structured data use enabled an improvement in the performance of the complaint’s classification.

Extraction and enrichment of features to improve complaint text classification performance

The purpose of this paper is to analyze the use of different machine learning approaches and algorithms to be integrated as an automated assistance on a tool to aid the creation of new annotated datasets. We evaluate how they scale in an environment without dedicated machine learning hardware.
In particular, we study the impact over a dataset with few examples and one that is being constructed.
We experiment using deep learning algorithms (Bert) and classical learning algorithms with a lower computational cost (W2V and Glove combined with RF and SVM).
Our experiments show that deep learning algorithms have a performance advantage over classical techniques.
However, deep learning algorithms have a high computational cost, making them inadequate to an environment with reduced hardware resources.
Simulations using Active and Iterative machine learning techniques to assist the creation of new datasets are conducted. For these simulations, we use the classical learning algorithms because of their computational cost.
The knowledge gathered with our experimental evaluation aims to support the creation of a tool for building new text datasets.

Iterative machine learning applied to annotation of text datasets

Subjects change frequently in moderated debates with several participants,
such as in parliamentary sessions, electoral debates, and trials. Partitioning a debate into blocks with the same subject is essential for understanding. Often a moderator is responsible for defining when a new block begins so that the task of automatically partitioning a moderated debate can focus solely on the moderator's behavior. 
In this paper, we (i) propose a new algorithm, DEBACER, which partitions moderated debates; (ii) carry out a comparative study between conventional and BERTimbau pipelines; and (iii) validate DEBACER applying it to the minutes of the Assembly of the Republic of Portugal. Our results show the effectiveness of DEBACER.

DEBACER: a method for slicing moderated debates

Predicting review helpfulness is an important task in Natural Language Processing. It is useful for dealing with the huge amount of online reviews on varied domains and languages, helping and guiding users on what to read and consider in their daily decisions. However, there are limited initiatives to investigate the nature of this task and how hard it is. This paper aims to fulfill this gap, providing a better understanding of it. Two complementary experiments are performed in order to uncover patterns of usefulness evaluation as performed by humans and relevant features for machine prediction. To assure our results, we run the experiments for two different domains: movies and apps. We show that humans agree on the process of assigning helpfulness to reviews, despite the difficulty of the task. More than this, people perform this process systematically and consistently. Finally, we empirically identify the most relevant content features for machine learning prediction of review helpfulness.

The Challenges of Modeling and Predicting Online Review Helpfulness

Abstract. The Brazilian Exclusive Economic Zone, or the “Blue Amazon”, with its extensive maritime area, is the primary means of transport for the country’s foreign trade and is important due to its oil reserves, gas and other mineral resources, in addition to the significant influence on the Brazilian climate. We have manually built a question answering (QA) dataset based on crawled arti- cles and have applied an off-the-shelf QA system based on a fine-tuned BERTim- bau Model, achieving an F1-score of 47.0. More importantly, we explored how the proper visualization of attention weights can support helpful interpretations of the system’s answers, which is critical in real environments.


Interpretability of Attention in Question and Answering System about the Blue Amazon in Portuguese.

RBF-based neural networks models (RBFNN) are classic machine
learning methods that contain a layer of Radial Basis Functions (RBF) which
act as the feature extractor for the final layer, which performs the pattern recognition. The estimation of RBF radius during the model training is one of the most
crucial tasks for this method and directly affects its generalization power and
accuracy. This works presents a new heuristic for radius estimation and employs
computational experiments to assess its performance against other approaches
using 14 classification problems. The proposed method is shown competitive
performance, winning the other approaches in 9 out of 14 problems.

A New Heuristic for Radius Estimation in RBF Neural Networks

This talk will present the result of a time-series classification task on the stock market, compare the results with well-known methods. We propose the
Trend Classification Trading Algorithm (TCTA) which is based on time series
classification and trend forecasting to perform trade. TCTA first employs the
K-means to cluster 5-days closing price segments and label them according to
its trend. A deep learning classification model is then trained with these label
sequences to estimate the next trend. Trading points are given by the alternation
on trend estimates. Results considering 20 shares from Ibovespa show TCTA
present higher profit than buy-and-hold and trading schemes based on Moving
Average Convergence Divergence (MACD) or Bollinger bands.

Premium content

Técnicas de Aprendizado por Reforço Aplicadas em Jogos Eletrônicos na Plataforma Geral do Unity

Downloads

Next from BRACIS 2021

Definição de regiões de interesse em problemas multiobjetivo utilizando hiper-elipses

Similar lecture

Utilizing Unity as a Classroom Tool for Physics Simulations