Collin Lynch

Automated feather condition assessment is essential for monitoring poultry welfare and promoting precision farming. Feathering is also related to fertility in broiler breeders. However, manual feather scoring (FS) is time-consuming, subjective, and difficult to apply reliably in large-scale commercial production systems. To address these issues, this study presents a novel deep learning (DL) framework, Attention-Gated Fusion Network (AGF-Net), for automated multiclass FS classification using RGB data. The proposed architecture integrates multiscale feature extraction, a Convolutional Block Attention Module (CBAM), and adaptive gated fusion to capture discriminative FS characteristics across feature hierarchies. The model was trained from scratch on a large dataset of 67,231 back feathering images with bird-ID grouped partitioning to minimize data leakage. A multidimensional evaluation protocol was conducted, including independent testset analysis, bird-level evaluation, calibration assessment, robustness testing, ablation analysis, computational profiling, and statistical validation. On the independent test set, AGF-Net achieved 97.97% image-level accuracy, 97.98% weighted F1-score, 0.9925 quadratic weighted kappa (QWK), and 99.97% macro-AUC. At the bird level, AGF-Net achieved 97.41% accuracy, 97.41% weighted F1-score, and 0.9926 QWK, confirming strong generalization to unseen birds. AGF-Net also produced the best calibration performance, with the lowest expected calibration error (ECE) at image-level (0.0022) and bird-level (0.0122). Robustness analysis showed strong stability under clean and illumination-varied conditions. The Wilcoxon signed-rank test confirmed significant improvements over all baseline models and other state-of-the-art (SOTA) techniques. Additionally, AGF-Net achieved real-time inference at 20.37 FPS, supporting its potential for scalable poultry welfare monitoring and decision-support systems. These findings demonstrate a reliable RGBbased framework for objective, real-time FS assessment in broiler breeder production environments and precision poultry systems.

The increasing demand for efficiency, biosecurity, and reduced labor dependency in poultry farming has accelerated the adoption of automation and intelligent technologies. These systems are essential for improving animal health management and production efficiency. Among these technologies, deep learning (DL) has emerged as an effective solution, outperforming conventional image processing and machine learning approaches for poultry health monitoring and identification. This study is among the first to integrate a systematic review with a bibliometric mapping of DL applications in precision poultry farming. This study systematically reviews DLbased methods to identify major application areas and the models, datasets, and evaluation practices commonly used. An initial search retrieved 1,028 studies from five academic databases: IEEE Xplore, ScienceDirect, Wiley, SpringerLink, and Google Scholar. After applying predefined inclusion and exclusion criteria, 55 studies were selected for detailed analysis. These studies were used to answer the defined review questions. Two main application domains were identified: health-related and identification tasks. Poultry health monitoring was the dominant focus, accounting for 65% of the studies. The remaining studies focused on identification tasks. Among the eighteen reported DL models, Convolutional Neural Networks (CNNs) were the most widely used. Other commonly used models included YOLO variants, Long Short-Term Memory networks, and Mask-RCNN. Nineteen networks were identified, with ResNet-50, ResNet-18, ResNet-34, and ResNet-101 used most frequently. Twenty evaluation metrics were identified, although Matthew's Correlation Coefficient, One-off Accuracy, and Area Under the Curve appeared in fewer than five studies each. Bibliometric mapping further revealed major research themes, keyword relationships, and collaboration patterns, providing insight into the intellectual structure and development of this field. The findings highlight the growing importance of highquality datasets, lightweight architectures, and multimodal DL systems for future research and practical deployment. This study offers a structured reference for researchers and practitioners seeking to advance DL-based solutions in intelligent poultry health and identification systems.

Office hours are a core feature of CS courses. They provide a crucial vector for personalized instruction and individual support. Despite their importance, we the impact of these personal events has not been well assessed each office hour interaction is often hidden from instructors. We do not know if students actually understand the solution or the advice given during the interaction. This lack of knowledge presents a challenge to efforts geared at improving instructional practices and student outcomes. In this poster we report on the results of a survey analysis of office hour participants. Our results show that students and instructional staff had divergent assessments of the interactions with teaching staff overestimating the benefits more than 18.9% of the time. This work highlights detailed analyses of these results along with implications for teaching strategies.

Coding assignments and code-based problem-solving are central to CS education. They are also often the least visible to instructors. Students working on larger problems typically do so individually or in small teams with most of their effort taking place on their systems and in their own time. This lack of visibility can make it difficult for us to understand how students approach coding problems; when they develop and modify their plans (if any); how they respond to errors; and what specific events trigger them to seek help. While we can monitor students' overall progress and general struggles based on commits and help requests this data can give us an incomplete picture of their effort. However by tracking students' coding actions we can evaluate how they approach both high- and low-level coding challenges, evaluate their overall coding processes, and we can better identify opportunities for tailored feedback. We describe the development of EclipseMonitor, an open-source tool for the analysis of students' real-time coding activities and we discuss insights that we have obtained from its evaluation in a large CS2 course. The tool is publicly available at https://github.com/ArgLab/EclipseEvents

Promptly and properly addressing students' help requests during office hours is a critical challenge for large CS courses. With a large number of help requests, the queue gets longer and students have to endure long wait times. To address this problem, we try to quantify students' patience in the queue through survival analysis. Our results show that half of the students are willing to stay in the queue after waiting for 142.5 minutes. Moreover, we find that female students, morning requests, returning students, and requests about test failures are more likely to stay in the queue for a longer time.