Brazil

Reinforcement Learning (RL) is a well-known technique to train autonomous
agents through interactions with the environment. However, the learning process
has a high sample-complexity to infer an effective policy, especially when multiple agents are simultaneously actuating in the environment. We here propose to take advantage of previous knowledge, so as to accelerate learning in multiagent RL problems. Agents may reuse knowledge gathered from previously solved tasks, and they may also receive guidance from more experienced friendly agents to learn faster. However, specifying a framework to integrate knowledge reuse into the learning process requires answering challenging research questions, such as: How to abstract task solutions to reuse them later in similar yet different tasks? How to define when advice should be given? How to select the previous task most similar to the new one and map correspondences? and How to defined if received advice is trustworthy? Although many methods exist to reuse knowledge from a specific knowledge source, the literature is composed of methods very specialized in their own scenario that are not compatible. We propose in this thesis to reuse knowledge both from previously solved tasks and from communication with other agents. In order to accomplish our goal, we propose several flexible methods to enable each of those two types of knowledge reuse. Our proposed methods include: Ad Hoc Advising, an inter-agent advising framework, where agents can share knowledge among themselves through action suggestions; and an extension of the object-oriented representation to multiagent RL and methods to leverage it for reusing knowledge. Combined, our methods provide ways to reuse knowledge from both previously solved tasks and other agents with state-of-the-art performance. Our contributions are first steps towards more flexible and broadly applicable multiagent transfer learning methods, where agents will be able to consistently combine reused knowledge from multiple sources, including solved tasks and other learning agents.

BRACIS 2020

Methods and Algorithms for Knowledge Reuse in Multiagent Reinforcement Learning

machine learning

transfer learning

reinforcement learning

# Welcome everyone to ACL 2022!

The 60th Annual Meeting of the Association for Computational Linguistics is taking place May 22-27, 2022 as a hybrid event, in Dublin and online. We are happy to welcome all of you to this anniversary edition with an almost 50-50 in-person and virtual participation. 
The main conference program features oral presentations, in-person and virtual posters and demo sessions, a plenary session for our best paper presentations and awards, three amazing keynote events and two new initiatives of invited talks: Spotlight Talks for Young Rising Stars (STIRS) and The Next Big Idea Talks. Posters (including Findings of ACL 2022) and demos are grouped by areas for both the in-person and the virtual sessions. For the virtual component, the talks will be on Zoom and the posters and the demos will be in GatherTown. The Student Research Workshop will have an oral session and a poster session as part of Poster Session 1. The program also features eight Tutorials and 28 Workshops. 

 
We wish you a wonderful conference! 
[**The ACL 2022 Organizing Committee**](https://www.2022.aclweb.org/organisers)
 
[**Conference Handbook**](https://drive.google.com/file/d/1_BUCMfhMVrjG9E2e71aHdHeE28KSje0l/view?usp=sharing) 
[**Mini Handbook**](https://drive.google.com/file/d/1qlBKl0wzmlVF1oCeMQl3BahLd9nLP5Ce/view?usp=sharing) 
[**Posters and Demo guides**](https://drive.google.com/file/d/1UucMAoCNncIOaH1rMMDa0owuG9qgvJTG/view?usp=sharing)

ACL 2022

The Association for Computational Linguistics (ACL) is the premier international scientific and professional society for people working on computational problems involving human language, a field often referred to as either computational linguistics or natural language processing (NLP). 

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.

Network-based representations can model different types of relationships between texts, they are capable of capturing patterns that are hardly captured by vector space model, and network-based clustering algorithms, such as label propagation, have linear complexity. However, network-based clustering has not been explored (i) specifically in clustering texts; and (ii) with different possibilities of text representations on networks. Thus, this article’s objective is to explore and analyze clustering techniques applied to different types of network representations. Experiments were performed using 30 collections from different domains, represented in the bag-of-words format, similarity networks of the type k-Nearest Neighbors, and bipartite networks. The label propagation approach applied in similarity networks presented the best results for most evaluation measures and most text collections.

Unsupervised Machine Learning Based on Heterogeneous Networks for Text Clustering

Ferramentas de geração de bases de dados sintéticos ganharam atenção como uma alternativa para a avaliação eficaz de classificadores. Como são altamente configuráveis, as ferramentas possibilitam a criação de bases com características específicas, permitindo a avaliação do classificador em diferentes cenários. Este trabalho propõe a CbDGen, uma ferramenta para geração de bases sintéticas baseada em complexidade de dados. Com uma interface simples e parametrizável, a CbDGen possibilita a inserção de informações básicas (i.e., número de classes, atributos), a seleção de diferentes distribuições e nível de ruído. O seu grande diferencial é permitir que a escolha do tipo de complexidade desejado (i.e., balanceamento, separabilidade), bem como o nível desta complexidade. Os resultados mostraram que a CbDGen é capaz de produzir bases sintéticas aderentes às escolhas realizadas pelo usuário, se mostrando uma plataforma propícia para a avaliação de classificadores.

CbDGen: A Complexity-based Synthetic Dataset Generation Tool

Brazil has the third-largest prison population globally, and it has been growing steadily for more than two decades. Constant growth and low jail investment generated significant problems, such as overcrowding and widespread diseases. This study proposes the construction of a Random Forest classifier to predict the occurrence of deaths in prisons. We extracted data from the National Survey of Penitentiary Information for the years 2015 to 2016. The best-fitted classifier achieved accuracy equals 87% being able to identify correctly up to 84% of deaths occurrences. In the present work, it was possible to establish a relationship between prisons' reality and the data mined, determining areas in need of investment in the penitentiary system.

Death Registry Prediction in Brazilian Male Prisons with a Random Forest Ensemble

The definition of policies to control the spread of infectious diseases, such as COVID19, is an important issue for society and for the government agents, responsable for the controlling measures. The majority of the vast number of recent research on this topic, uses past data to estimate the disease behavior in the future, considering an arbitrary policy. However, the use of automated planning techniques based on Markov Decision Processes (MDPs) has been shown to be more efficient to compute optimized control policies. One of the most common model used to control infectious diseases is the "SIR with vaccination model" in which the dynamics of the susceptible, infected and recovered individuals amount is controled by vaccination policies. But since this is a problem of saving lives, it is also necessary to take into account the attitude towards risk of a decision maker. Thus, the purpose of this work is to use risk-sensitive MDPs to find optimized policies to control the spread of the considered infectious diseases, taking into account the parameters of the SIR with risk model. The results show that vaccination policies depend on both the basic reproductive rate R0, as usual, and the attitude towards the risk of the government agents.

Risk Sensitive Policies for the Control of the Spread of Infectious Diseases

Inspection of weld radiographs is an essential task for preventing leaks in oil industry pipelines. Automatic inspection is important to assist specialists whose performance can be influenced by fatigue, visual acuity, and experience. A challenging issue in automatic inspection is the availability of laudable images that can be used for training classifiers. Data augmentation is an alternative however it can produce images with little or even negative impact on the training. This article presents a method for filtering images, aiming to favor the classifier relearning process, which is done by eliminating images with low performance from the training set and further retraining the segmentation model, with a positive impact on its generalization capabilities. Experimental results of the method show that data filtering accelerates the learning process and produces an improvement in the segmentation performance.


Training Data Filtering for Deep Learning Applied toInspection of Welded Joints in Oil Pipelines

A previsão de receita tributária é uma tarefa desafiadora às secretarias de fazenda. Por conta disso, neste trabalho se inicia um projeto de investigação do uso do modelo LSTM para a previsão de receita tributária, em especial o tributo ICMS. As previsões obtidas pelo modelo LSTM são comparadas às realizadas pela SEFAZ e pelo modelo de rede neural MLP. A rede LSTM apresentou resultados promissores para a previsão do tributo ICMS.

Previsão de Arrecadação do Tributo ICMS do Estado do Rio de Janeiro Utilizando a Rede Neural LSTM

Deep Learning consists of modern techniques that approach one or more methods of artificial intelligence. An approach is to use Convolutional Neural Networks together with traditional Neural Networks for digital image processing. In this work, a research is conducted to evaluate a deep learning technique in the classification of potassium (k) macronutrient deficiency by the soybean leaf. This research presents an own dataset with distinct treatments of the potassium macronutrient. Several deep learning scenarios are evaluated with different metrics. The results are compared with the literature and show a great potential of Convolutional Neural Networks, with precision above 99% in this type of classification.

Classifying the Macronutrient Deficiency in Soybean Leaf with Deep Learning

In this work was proposed an approach to classify the organizational structure of a group of moving target agents that are continuously monitored by a smaller group of moving observer agents. The approach considers that the group of target agents can organize themselves according to eight different paradigms. The agents communicate through the exchange of messages whose contents are speech-act performative. We evaluate the approach considering seven techniques to solve the problem of classifying the group of target agents' organizational structure.

Classifying Organizational Structures on Targets in the Cooperative Target Observation

Currently, keyboards are the most common means of communicating with computers. Despite being the most commonly used keyboard layout, QWERTY has had various issues raised concerning its effectiveness, as it is not efficient in English (target language) or in fact other languages. Therefore, this paper presents the development process of a Genetic Algorithm with the intention of generating a more adequate and coherent layout proposal for Brazilian Portuguese, which has its focus on ergonomics and user productivity. Using five ergonomic criteria and a statistical analysis of the characters and sequences of most frequently used pairs in Brazilian Portuguese, a layout approximately 53% better than QWERTY was obtained.

Using Genetic Algorithms to Design an Optimized Keyboard Layout for Brazilian Portuguese

Convolutional Neural Networks (CNNs) is the main solution for image classification tasks in different applications. However, selecting the most suitable CNN and its parameters for a given image dataset is usually performed by trial and error, which may take much time and computational cost. This paper proposes the Dataset2Vec method and employs Meta-Learning (MtL) to select CNN architectures for image classification. Dataset2Vec adopts a deep neural network to extract features from images datasets, embedding them in a single feature vector. To evaluate the proposed solution, it was adopted to select among six CNN algorithms for 45 two-classes image datasets. The results showed that the MtL using Dataset2Vec outperformed by different baseline methods in all performance measures evaluated, indicating the proposal was able to extract representative features from datasets of images for CNN selection.

Methods and Algorithms for Knowledge Reuse in Multiagent Reinforcement Learning

Downloads

Next from BRACIS 2020

Unsupervised Machine Learning Based on Heterogeneous Networks for Text Clustering

Similar lecture

When to Use Multi-Task Learning vs Intermediate Fine-Tuning for Pre-Trained Encoder Transfer Learning

Methods and Algorithms for Knowledge Reuse in Multiagent Reinforcement Learning

.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

Unsupervised Machine Learning Based on Heterogeneous Networks for Text Clustering

Similar lecture

When to Use Multi-Task Learning vs Intermediate Fine-Tuning for Pre-Trained Encoder Transfer Learning

Downloads