Intelligent Systems: 12th Brazilian Conference, BRACIS 2023, Belo Horizonte, Brazil, September 25–29, 2023, Proceedings, Part III (Lecture Notes in Computer Science, 14197)

The three-volume set LNAI 14195, 14196, and 14197 constitutes the refereed proceedings of the 12th Brazilian Conference on Intelligent Systems, BRACIS 2023, which took place in Belo Horizonte, Brazil, in September 2023. The 90 full papers included in the proceedings were carefully reviewed and selected from 242 submissions. They have been organized in topical sections as follows: Part I: Best papers; resource allocation and planning; rules and feature extraction; AI and education; agent systems; explainability; AI models; Part II: Transformer applications; convolutional neural networks; deep learning applications; reinforcement learning and GAN; classification; machine learning analysis; Part III: Evolutionary algorithms; optimization strategies; computer vision; language and models; graph neural networks; pattern recognition; AI applications. Preface Organization Contents – Part III Evolutionary Algorithms Multiobjective Evolutionary Algorithms Applied to the Optimization of Expanded Genetic Codes 1 Introduction 2 Proposed Genetic Algorithms 2.1 Codification and Operators 2.2 Objectives 2.3 Weighted Approach 2.4 Pareto Approach 3 Experiments 3.1 Experimental Design 3.2 Experimental Results 4 Conclusions References Genetic Algorithms with Optimality Cuts to the Max-Cut Problem 1 Introduction 2 Optimality Cuts for Max-Cut 3 Developed Algorithms 3.1 Genetic Algorithm with Optimality Cuts 3.2 Genetic Algorithm with Perturbation-Based on Tabu Search 4 Computational Results 5 Conclusion References Assessment of Robust Multi-objective Evolutionary Algorithms on Robust and Noisy Environments 1 Introduction 2 Preliminary Concepts 2.1 Types of Uncertainties 2.2 Robustness Measurements 3 Function Generator Extension - Robust Optimization And/or Noisy Optimization 4 Computational Experiment 4.1 Algorithms 4.2 Experimental Setup 4.3 Results 4.4 Discussion 5 Conclusion References Optimization Strategies Binary Flying Squirrel Optimizer for Feature Selection 1 Introduction 2 Flying Squirrel Optimizer 2.1 Algorithmic Analysis 3 Methodology 4 Experimental Results 5 Conclusions References Fitness Landscape Analysis of TPOT Using Local Optima Network 1 Introduction 2 Related Work 3 Methodology 3.1 TPOT: Tree-Based Pipeline Optimization Tool 3.2 Construction of the Fitness Landscape 3.3 Compressing Neutral Nodes 3.4 Local Optima Networks (LON) 3.5 Nodes 3.6 Edges 3.7 Network Statistics 4 Experimental Setup 4.1 Characterization of the Fitness Landscapes 5 Results 6 Conclusion References Optimization Strategies for BERT-Based Named Entity Recognition 1 Introduction 2 Related Work 3 Methodology and Experiments 3.1 Model Soups Experiments 3.2 Domain Adaptation Experiments 3.3 Causal Language Modeling - Few/zero-Shot Learning 4 Results and Discussions 4.1 Model Soup Results Analysis 4.2 Domain Adaptation Results Analysis 4.3 Causal Language Modeling - Few/zero-Shot Learning 5 Conclusions References FlexCon-CE: A Semi-supervised Method with an Ensemble-Based Adaptive Confidence 1 Introduction 2 Theoretical Aspects 2.1 Semi-supervised Learning 2.2 Classifier Ensembles 3 Related Work 4 The Proposed Method: FlexCon-CE 5 Experimental Methodology 5.1 Datasets 5.2 Methods and Materials 6 Experimental Results 6.1 Accuracy Analysis 6.2 Statistical Analysis 7 Conclusion and Future Works References Computer Vision Single Image Super-Resolution Based on Capsule Neural Networks 1 Introduction 2 Background 3 Proposed Method 4 Experimental Results 5 Conclusions References Development of a Deep Learning Model for the Classification of Mosquito Larvae Images 1 Introduction 2 Background: Morphology of Mosquito Larvae 3 Related Work 4 Research Methodology 5 Development of the Image Classification Model 5.1 Requirements Analysis 5.2 Data Preparation 5.3 Model Training, Evaluation and Comparison 6 Prediction Test 6.1 Test Preparation 6.2 Test Results 7 Discussion 8 Conclusion References A Simple and Low-Cost Method for Leaf Surface Dimension Estimation Based on Digital Images 1 Introduction 2 Related Work 3 Proposed Method 3.1 Capture 3.2 Thresholding 3.3 Obtaining Contours 3.4 Filtering 3.5 Detection 3.6 Dimensions Calculation 4 Results and Discussion 4.1 Metrics 4.2 Controlled Tests 4.3 Leaf Tests 5 Conclusions References Crop Row Line Detection with Auxiliary Segmentation Task 1 Introduction 2 Related Work 3 Approach 3.1 Data Annotation and Preprocessing 3.2 Network Model 4 Simulation and Test Setup 5 Results 5.1 Training Phase 5.2 Simulation Runs 6 Discussion 7 Conclusion References Multiple Object Tracking in Native Bee Hives: A Case Study with Jataí in the Field 1 Introduction 2 Related Works 3 Methods 3.1 Custom Dataset – Tetragonisca angustula (Jataí) 3.2 Detection and Classification of the Native Bee 3.3 Tracking and Optical Flow of Jataís 4 Results and Discussion 4.1 Detection and Classification of Jataí 4.2 Tracking and Optical Flow of Detected Bees 5 Conclusion References An Open Source Eye Gaze Tracker System to Perform Remote User Testing Evaluations 1 Introduction 2 Context 2.1 User Testing and Usability 2.2 Eye Tracker 3 Related Work 4 Prototype Results 4.1 Gaze Tracker System 4.2 Web System 4.3 API 5 Conclusions and Future Work References Language and Models Who Killed the Winograd Schema Challenge? 1 Introduction 2 The Winograd Schema Challenge 3 LLMs, RoBERTa and WinoGrande: Main Suspects 4 BERT and Cloze Procedures 5 Textual Entailment, and Schemas as Cloze Units 6 A Second Look at RoBERTa and WinoGrande 7 Closing (Clozing?) Arguments References 5554721En15FigaPrint.eps Sabiá: Portuguese Large Language Models 1 Introduction 2 Related Work 3 Methodology 3.1 Pretraining Data 3.2 Sabiá Models 3.3 Sabiá-J 4 Evaluation on Poeta 5 Results 5.1 Results per Dataset 5.2 Data Contamination 5.3 Ablation: English Datasets 6 Limitations 7 Conclusion References Disambiguation of Universal Dependencies Part-of-Speech Tags of Closed Class Words in Portuguese 1 Introduction 2 Related Work 3 Problem Definition and Baseline Approaches 4 Prediction Methods 4.1 Prediction Through Markovian Models 4.2 Prediction Through UDPipe 2.0 4.3 Prediction Through BERTimbau-Based Model 5 Experiments 6 Conclusion References Bete: A Brazilian Portuguese Dataset for Named Entity Recognition and Relation Extraction in the Diabetes Healthcare Domain 1 Introduction 1.1 Related Work 2 Methodology 2.1 Corpus 2.2 Entity and Relation Extraction 3 Dataset Information 4 Experiments Setup 5 Results 6 Discussion 7 Conclusion References LegalBert-pt: A Pretrained Language Model for the Brazilian Portuguese Legal Domain 1 Introduction 2 Related Work 3 LegalBert-pt: A BERT Model for the Brazilian Legal Domain 3.1 Pretraining Data 3.2 Vocabulary Generation 3.3 Variations of the LegalBert-pt Model 4 Evaluation 4.1 Perplexity 4.2 Named Entity Recognition 4.3 Text Classification 5 Experimental Results 6 Conclusion and Future Works References A Framework for Controversial Political Topics Identification Using Twitter Data 1 Introduction 2 Related Works 3 Methodology 3.1 Data Collection 3.2 Tweet Pre-processing 3.3 Clustering Tweets with HDBSCAN 3.4 Parameter Calibration 3.5 Sentiment Analysis 3.6 Controversial Topic Identification 3.7 Cluster Analysis 3.8 Evaluation 4 Results and Discussions 4.1 Clustering and Sentiment Analysis 4.2 Controversial Topic Discussion 4.3 Evaluating a Simpler Approach 4.4 Other Discoveries 5 Conclusions References Leveraging Sign Language Processing with Formal SignWriting and Deep Learning Architectures 1 Introduction 2 Background 2.1 Sign Language 2.2 Writing Systems for Sign Language and the SignWriting 2.3 Deep Learning Basics 3 Related Work 4 Method 4.1 Datasets 4.2 Data Mapping 4.3 Implementation Details 4.4 Classifier Evaluation 5 Experiments and Results 6 Final Remarks References A Clustering Validation Index Based on Semantic Description 1 Introduction 2 Related Works 3 Validation Based on the Semantic Description of the Clusters 3.1 Proposed Index 4 Synthetic Data 5 Set of Experiments 6 Results and Discussion 7 Conclusion References Graph Neural Networks Detecting Multiple Epidemic Sources in Network Epidemics Using Graph Neural Networks 1 Introduction 2 Network Epidemics and Problem Statement 2.1 Epidemic Observation 2.2 Problem Statement 3 Related Work 4 Proposed Framework 4.1 Metrics and Attributes 4.2 Graph Neural Network Model 4.3 Loss Function 4.4 Datasets and Training 5 Evaluation 5.1 Network Models and Real Networks 5.2 Evaluation Metrics 5.3 Results 5.4 Results with Mixed Datasets 5.5 Comparison with Baselines 6 Conclusion References Prediction of Cancer-Related miRNA Targets Using an Integrative Heterogeneous Graph Neural Network-Based Method 1 Introduction 2 Related Works 3 Methodology 3.1 Data Collection, Preprocessing and Integration 3.2 Model Training and Evaluation 4 Experiments 5 Results and Discussion 5.1 S1: Exploring the Predictive Performance of HinSAGE 5.2 S2: Comparison of HinSAGE Model with Related Works 6 Conclusion References Time Series Forecasting of COVID-19 Cases in Brazil with GNN and Mobility Networks 1 Introduction 2 Related Works 3 Methodology 3.1 Mathematical Preliminaries 3.2 GCN 3.3 GCN Models for Time Series Prediction 3.4 Prophet and LSTM 3.5 Datasets: COVID-19 Cases and the Brazilian Mobility Network 3.6 Experimental Setup 4 Results 4.1 GCRN 4.2 GCLSTM 4.3 Comparing Forecast Models Results 5 Discussions and Conclusions References Pattern Recognition Federated Learning and Mel-Spectrograms for Physical Violence Detection in Audio 1 Introduction 2 Background 2.1 Convolutional Neural Networks (CNN) 2.2 Federated Learning (FL) 2.3 Mel-spectrogram 3 Related Works 4 Materials and Methods 4.1 Dataset 4.2 Compared Methods 4.3 Flower Framework 4.4 Experimental Setup 5 Results 6 Conclusion and Future Work References Police Report Similarity Search: A Case Study 1 Introduction 2 Backgound and Related Works 3 Data and Methods 4 Experimental Results 5 Conclusion and Future Works References Evaluating Contextualized Embeddings for Topic Modeling in Public Bidding Domain 1 Introduction 2 Related Work 3 Methodology 3.1 Preprocessing 3.2 Topic Modeling 4 Experimental Design 4.1 Dataset 4.2 Sentence Embedding Models 4.3 Evaluation 5 Experimental Results 5.1 Internal Evaluation 5.2 External Evaluation 6 Conclusion References AI Applications A Tool for Measuring Energy Consumption in Data Stream Mining 1 Introduction 2 Data Stream Mining and Energy Consumption 2.1 Classifiers 2.2 Requirements 3 Related Works 4 Proposal 5 Experiments 5.1 Experiment 1 - Validation Against a Hardware Solution 5.2 Experiment 2 - Analyzing Different Classifiers in Stationary and Non-Stationary Environments 6 Conclusion References Improved Fuzzy Decision System for Energy Bill Reduction in the Context of the Brazilian White Tariff Scenario 1 Introduction 2 Literature Review 3 Features of the Proposed System 4 Development of the Fuzzy Algorithm 5 Simulation Specification 6 Research Results and Discussion 7 Conclusion References Exploring Artificial Intelligence Methods for the Automatic Measurement of a New Biomarker Aiming at Glaucoma Diagnosis 1 Introduction 2 Related Work 3 Proposed Approach 3.1 Data Annotation 3.2 CNN Architecture 3.3 Data Augmentation 3.4 Measurement of Cup Portion on ONH Structure 4 Conversion to Micron Meter 5 Results 5.1 Evaluation of NN Segmentation 5.2 Results over Segmentation Using NN 5.3 Discovery of the Hypotenuse in the Excavation Region 6 Conclusion References Investigation of Deep Active Self-learning Algorithms Applied to Named Entity Recognition 1 Introduction 2 Related Works 3 Deep Active Self-learning Algorithm 4 Sentence-Level Active Self-learning Algorithm 5 Token-Level Active-Self Learning Algorithm 6 Experimental Design 6.1 CNN-CNN-LSTM Model 6.2 CNN-biLSTM-CRF Model 7 Results 7.1 Results of Sentence-Level Strategy 7.2 Results of Token-Level Strategy 8 Conclusion References Author Index