Contemporary education is a vast field that is concerned with the performance of education systems. In a formal e-learning context, student dropout is considered one of the main problems and has received much attention from the learning analytics research community, which has reported several approaches to the development of models for the early prediction of at-risk students. However, maximizing the results obtained by predictions is a considerable challenge. In this work, we developed a solution using only students’ interactions with the virtual learning environment and its derivative features for early predict at-risk students in a Brazilian distance technical high school course that is 103 weeks in duration. To maximize results, we developed an elitist genetic algorithm based on Darwin’s theory of natural selection for hyperparameter tuning. With the application of the proposed technique, we predicted the student at risk with an Area Under the Receiver Operating Characteristic Curve (AUROC) above 0.75 in the initial weeks of a course. The results demonstrate the viability of applying interaction count and derivative features to generate prediction models in contexts where access to demographic data is restricted. The application of a genetic algorithm to the tuning of hyperparameters classifiers can increase their performance in comparison with other techniques.
12 September 2019 in Journal of Intelligent & Robotic Systems
Reward functions define an agent's behavior in reinforcement learning by determining actions based on the feedback received. In the domain of path following, reward functions guide the agent toward successfully following the desired path. This paper presents a survey of reward functions focused on path-following tasks for mobile robots. The characteristics, strategies, and challenges associated with creating feedback mechanisms in various domains are highlighted. Components of the reward functions that can be controlled are explored and discussed. Furthermore, for each reward function, scenarios and mobile robots are indicated. Thus, this study provides insights into the reward components that influence reinforcement learning systems in robotic navigation.
Anthropomorphism plays a crucial role in Human-Robot Interaction (HRI), as nonverbal cues, such as eye gaze and facial expressions, significantly influence social perception. Among facial features, the eyes are particularly important, as they convey emotion and indicate the focus of attention. This work presents the development of a compact and functional animatronic eye system designed to improve an existing mechanism integrated into the robotic face of a domestic service robot. The current system, which mechanically links both eyes, limits expressiveness and restricts natural movement. To address these issues, two alternative mechanisms were explored: ABENICS and a system inspired by the HITSZ-Snakebot II robot. These were developed through CAD modeling and fabricated using both FDM and DLP 3D printing techniques. Three prototypes were created-two based on ABENICS and one on the snakebot design. The final prototype, based on the snakebot model, proved to be the most effective, offering a greater angular range, size reduction, and smoother, more natural movements. Its modular, compact structure makes it suitable for integration into expressive robotic faces, contributing to the development of more relatable and socially capable robots.
Identifying probiotic microorganisms requires linking genomic content to experimentally reported host-interaction traits. A machine-learning framework predicted seven probiotic phenotypes (acid resistance, bile resistance, adhesion, antimicrobial activity, immunomodulation, antioxidant activity, and antiproliferative potential) from functional COG categories and biosynthetic gene cluster annotations in 1183 genomes (780 probiotic and 403 non-probiotic). CatBoost, Random Forest, XGBoost, LightGBM, and Logistic Regression were evaluated with stratified cross-validation, leave-one-out cross-validation, and an 80:20 holdout split. Antioxidant and antiproliferative models performed best, with cross-validation F1-scores of 0.92 and 0.97 and holdout F1-scores of 0.70 and 0.72. SHAP analysis identified model-derived genomic associations rather than causal mechanisms, including an inverse association between T3PKS abundance and antioxidant classification and associations between lipid transport/metabolism features and antiproliferative classification. The framework supports high-throughput in silico prioritization of candidate probiotic strains for phenotype-specific validation. Interpretation remains limited by heterogeneous source annotations, label noise, and potential false negatives in the non-probiotic dataset.
Robust perception is paramount for autonomous systems that operate in real public environments. LiDAR sensors are known for their robustness, making them ideal for autonomous robots, which enable obstacle detection, reliability in adverse lighting conditions, and overall safety assurance. Most 3D LiDAR semantic segmentation networks are fine-tuned for outdoor and urban environments, overlooking other equally important scenes, such as industrial scenarios. In this work, we contribute to an open-source Convolutional Neural Network (CNN) pipeline for semantic segmentation of 3D LiDAR point clouds of indoor warehouse environments. Additionally, we introduce a dataset of simulated industrial point clouds for 3D semantic segmentation called WareSet. We evaluate the quality and viability of our network using well-defined quality metrics, including accuracy, Intersection over Union (IoU) per class, and mean Intersection over Union (mIoU), for both training and validation sets. The experiments were conducted in an unfamiliar warehouse environment, taking into account the proposed dataset and the specified quality metrics. The proposed approach yields satisfactory results of 98.49% and 90.59% for accuracy, and 90.64% and 49.49% for mIoU in training and validation, respectively, with good model convergence. Additionally, it achieves 89.7% accuracy and 50.4% mIoU on the unknown warehouse, emphasizing its generalization capability. The proposed dataset is open-sourced at https://github.com/DeskFanzin/WareSet.