A5000356136- Machine Learning and Data Classification
- Advanced Text Analysis Techniques
- Explainable Artificial Intelligence (XAI)
- Advanced Multi-Objective Optimization Algorithms
- Plant and animal studies
- Advanced Optimization Algorithms Research
- Topic Modeling
- Imbalanced Data Classification Techniques
- Face and Expression Recognition
- Insect and Arachnid Ecology and Behavior
- Optimal Experimental Design Methods
- Date Palm Research Studies
- Psychometric Methodologies and Testing
- Adversarial Robustness in Machine Learning
- Private Equity and Venture Capital
- Epilepsy research and treatment
- Text and Document Classification Technologies
- EEG and Brain-Computer Interfaces
- Gambling Behavior and Treatments
- Data Visualization and Analytics
- Medical Image Segmentation Techniques
- Crime Patterns and Interventions
- Data Mining Algorithms and Applications
- Multi-Criteria Decision Making
- Electricity Theft Detection Techniques
Universidade Estadual de Campinas (UNICAMP)
2018-2024
Fundação Getulio Vargas
2020-2023
Works requiring taxonomic knowledge face several challenges, such as arduous identification of many taxa and an insufficient number taxonomists to identify a great deal collected organisms. Machine learning tools, particularly convolutional neural networks (CNNs), are then welcome automatically generate high-performance classifiers from available data. Supported by the image datasets at largest online database on ant biology, AntWeb (www.antweb.org), we propose here ensemble CNNs genera...
When dealing with a multi-objective optimization problem, obtaining comprehensive representation of the Pareto front can be computationally expensive. Furthermore, identifying most representative solutions difficult and sometimes ambiguous. A popular selection are so-called knee solutions, where small improvement in any objective leads to large deterioration at least one other objective. In this paper, using sensitivity, we show how compute according their verbal definition maximal change....
Extracting and analyzing crime patterns in big cities is a challenging spatiotemporal problem. The hardness of the problem linked to two main factors, sparse nature activity its spread large spatial areas. Sparseness hampers most time series (crime series) comparison methods from working properly, while handling urban areas tends render computational costs such impractical. Visualizing different hidden data another issue this context, mainly due number that can show up analysis. In article,...
Saliency post-hoc explainability methods are important tools for understanding increasingly complex NLP models. While these can reflect the model's reasoning, they may not align with human intuition, making explanations plausible. In this work, we present a methodology incorporating rationales, which text annotations explaining decisions, into classification This incorporation enhances plausibility of while preserving their faithfulness. Our approach is agnostic to model architectures and...
Saliency post-hoc explainability methods are important tools for understanding increasingly complex NLP models.While these can reflect the model's reasoning, they may not align with human intuition, making explanations plausible.In this work, we present a methodology incorporating rationales, which text annotations explaining decisions, into classification models.This incorporation enhances plausibility of while preserving their faithfulness.Our approach is agnostic to model architectures...
The widespread use of machine learning in credit scoring has brought significant advancements risk assessment and decision-making. However, it also raised concerns about potential biases, discrimination, lack transparency these automated systems. This tutorial paper performed a non-systematic literature review to guide best practices for developing responsible models scoring, focusing on fairness, reject inference, explainability. We discuss definitions, metrics, techniques mitigating biases...
LLMs trained to detect hate speech have a significant challenge on identifying directed toward new or less common target groups. This happens because the models are primarily data focused more prevalent forms of hate, targeting groups that historically been subjected speech. Not only way defamation evolves through time, but targets may emerge, presenting were previously non-existent in datasets. work presents analyses influence targeted model prediction. We evaluate training strategies...
Regularized multitask learning is explicitly interpreted hereas a many-objective optimization problem, dealt with deterministic solver that properly controls the sampling of Pareto frontier. Each objective function corresponds to loss task, so we have as many objectives tasks. The obtained Pareto-optimal models are then explored implement distinct sharing strategies: (1) by considering single parameter vector for all tasks, simplest model could been conceived in learning, trade-offs along...
On binary classification, the goal of minimizing false positive and negative rates creates a conflict, being impossible to optimize both simultaneously. This challenge is even more significant on imbalanced classification datasets since an incorrect choice relative relevance each objective optimization can lead ignoring, or poorly learning minority class. The proposal this work takes into account existing conflict among losses classes, use deterministic multi-objective method, called MONISE,...
Abstract Mining counterfactual antecedents became a valuable tool to discover knowledge and explain machine learning models. It consists of generating synthetic samples from an original sample achieve the desired outcome in model thus helping understand prediction. An insightful methodology would explore broader set reveal multiple possibilities while operating on any classifier. Thus, we create tree-based search that requires monotonicity objective functions (a.k.a. cost functions); it...
Works requiring taxonomic knowledge face several challenges, such as arduous identification of many taxa and an insufficient number taxonomists to identify a great deal collected organisms. Machine learning tools, particularly convolutional neural networks (CNNs), are then welcome automatically generate high-performance classifiers from available data. We propose ensemble CNNs ant genera directly the head, profile dorsal perspectives images. Transfer is also considered improve individual...