A5011119333- Machine Learning and Data Classification
- Imbalanced Data Classification Techniques
- Explainable Artificial Intelligence (XAI)
- Adversarial Robustness in Machine Learning
- Sports Analytics and Performance
- Data Stream Mining Techniques
- Software Engineering Research
- Evolutionary Algorithms and Applications
- Advanced Statistical Methods and Models
- Online Learning and Analytics
- Face and Expression Recognition
- Data Analysis with R
- Neural Networks and Applications
- Artificial Intelligence in Healthcare and Education
- Machine Learning and Algorithms
Universidade Federal do Pará
2020-2023
Vale Technological Institute
2023
Universidade Federal do Maranhão
2021
In this paper we explore the reliability of contexts machine learning (ML) models. There are several evaluation procedures commonly used to validate a model (precision, F1 Score and others); However, these not linked itself, but only number correct answers presented by model. This characteristic makes it impossible assess whether was able learn through elements that make sense context in which is inserted. Therefore, could achieves good results training stage poor when needs be generalized....
Strategies based on Explainable Artificial Intelligence - XAI have emerged in computing to promote a better understanding of predictions made by black box models. Most measures used today explain these types models, generating attribute rankings aimed at explaining the model, that is, analysis Attribute Importance Model. There is no consensus which measure generates an overall explainability rank. For this reason, several proposals for tools (Ciu, Dalex, Eli5, Lofo, Shap and Skater). An...
Black box models are increasingly being used in the daily lives of human beings living society. Along with this increase, there has been emergence Explainable Artificial Intelligence (XAI) methods aimed at generating additional explanations regarding how model makes certain predictions. In sense, such as Dalex, Eli5, eXirt, Lofo and Shap emerged different proposals methodologies for black an agnostic way. these methods, questions arise "How Reliable Stable XAI Methods?". With aim shedding...
It is increasingly common for sectors of society to use Machine Learning (ML) techniques make decisions and variations with the data generated. One most problems that a dataset can present imbalance. Under these conditions, tendency produce biased models, which favor majority class. To mitigate this problem, balancing algorithms be used, one oversampling. However, it not simple task define whether an oversampling technique really helps in model learning process. The experiments carried out...
Gerar modelos de aprendizado máquina automaticamente segue sendo uma área muita pesquisa recente e desafiadora. E ainda não há forma definitiva capaz gerar resultados satisfatórios que apresente complexidade baixa. Com isso, neste trabalho, são apresentados os metodologia proposta utiliza conceitos da Teoria Resposta ao Item Algoritmo Genético para criar um algoritmo AutoML do tipo NAS seja Rede Neural competitiva com já conhecidos na literatura. Nos obtidos, foi possível modelo competitivo...
Intelligent systems that use Machine Learning classification algorithms are increasingly common in everyday society. However, many black-box models do not have characteristics allow for self-explanation of their predictions. This situation leads researchers the field and society to following question: How can I trust prediction a model cannot understand? In this sense, XAI emerges as AI aims create techniques capable explaining decisions classifier end-user. As result, several emerged, such...
Gerar modelos de aprendizado máquina automaticamente segue sendo uma área muita pesquisa recente. Porém, ainda não há forma definitiva capaz gerar simples que melhor generalizem e sejam imunes a subespecificação. Neste trabalho, são apresentado os resultados preliminares metodologia proposta utiliza conceitos da Teoria Resposta ao Item Algoritmo Genético para criar um algoritmo AutoML do tipo NAS seja Rede Neural competitiva com já conhecidos na literatura. Nos iniciais, foi possível modelo...
The experiments covered by Machine Learning (ML) must consider two important aspects to assess the performance of a model: datasets and algorithms. Robust benchmarks are needed evaluate best classifiers. For this, one can adopt gold standard available in public repositories. However, it is common not complexity dataset when evaluating. This work proposes new assessment methodology based on combination Item Response Theory (IRT) Glicko-2, rating system mechanism generally adopted strength...