This work analyzed the performance of the transferred learning deformable detection transformer (DeTR) architecture in the breast mass detection task on mammography images. Our experiment focused on optimizing the number of queries used by this architecture, given that this hyperparameter significantly influences the detection quality. We found that the deformable DeTR architecture with 50 queries outperformed the remaining models in terms of mAP 50 = 0.68 and mAP 50 : 95 = 0.41 metrics, demonstrating its ability to detect medium-large breast masses in non-high-density mammography images accurately. In contrast, some situations, such as tiny, small, and overlapped mass lesions and high-density mammogra-phy images, can limit the model's performance. However, it was also evidenced that these limitations are related to fine-tuning training on a small and unrepresentative mass lesion dataset such as INBreast. The model did not make false positive detections but did experience false negatives. Compared with three state-of-the-art YOLOv8 models, the proposed model outperformed but still produced competitive detection results while training significantly more parameters than the three YOLOv8 models.

This paper analyzes the impact of the integration of the Modular Static Synchronous Series Compensator (M-SSSC) solution that will be deployed in the Santa Marta 220 kV (Colombian substation) by using real-time simulations with Hardware in the Loop (HIL) technique, over the protection coordination schemes in the area of influence. The results show that the M-SSSC devices can be successfully coordinated with the protection schemes, mitigating overloads through the transmission lines during steady state and contingencies, while avoiding potential harmful interactions with the grid and, most importantly, with protection devices. This behavior is achieved by using overcurrent detection logics implemented on M-SSSC devices that trip fast-acting Silicon Controlled Rectifiers (SCRs) that cease the voltage injection in less than 5 ms

The recent development of large language models (LLMs) with multi-billion parameters, coupled with the creation of user-friendly application programming interfaces (APIs), has paved the way for automatically generating and executing code in response to straightforward human queries. This paper explores how these emerging capabilities can be harnessed to facilitate complex scientific workflows, eliminating the need for traditional coding methods. We present initial findings from our attempt to integrate Phyloflow with OpenAI's function-calling API, and outline a strategy for developing a comprehensive workflow management system based on these concepts.

Machine learning is a branch of artificial intelligence that provides computers the ability to create or improve algorithms without being explicitly programmed by directly learning from data. It is widely used in automation or decision-making tasks in fields such as image or speech recognition, sentiment analysis, or self-driving cars. However, its application in the field of communication networks is limited by the lack of appropriate research resources, such as rich datasets for training or the definition of a standard set of features. In this context, a standard latent space dimension is proposed by performing an autoencoder-based dimensionality reduction process. Different network security datasets are projected onto a lower-dimensional space to determine a standard or convergent dimension. The convergent dimension is determined by identifying the threshold above which diminishing returns begin to occur in the autoencoder loss as the latent space dimension increases. The experimental validation showed that four machine learning classification models, trained with a standard latent space of ten dimensions, performed as well as the models that used the non-reduced versions of the datasets in terms of F1-score and accuracy. Furthermore, a Wilcoxon statistical test showed that the mean accuracy of all classification models trained with the standard latent space dimension had a difference of less than 0.0235 in comparison to the models trained with the original inputs. A negligible difference in accuracy is a significant outcome because researchers can use only the latent space to perform experiments with certainty that the performance of ML models will not be constrained.

The increasing reliance on telematic applications has made networks attractive targets for cybercriminals. To address this concern, securing sensitive data by including a Network Intrusion Detection System (NIDS) is effective and feasible. Several approaches based on machine learning have been proposed to improve the detection accuracy of these systems and overcome associated drawbacks. However, machine learning is poorly adopted in the network domain, mainly because most datasets are outdated and guidance on selecting features that significantly contribute to a more efficient and effective learning process is needed. In this light, this paper proposes a standard set of features by applying feature selection techniques. A search on the feature space is conducted on five datasets to determine the best subset per dataset. The commonalities among the best subsets are then identified to define a standard set. Validation shows that, beyond improving the learning process, feature selection provides encouraging results for standardization of network datasets.

The annual Workshop on Workflows in Support of Large-Scale Science (WORKS) is a premier venue for the scientific workflow community to present the latest advances in research and development on the many facets of scientific workflows throughout their life-cycle. The Lightning Talks at WORKS focus on describing a novel tool, scientific workflow, or concept, which are work-in-progress and address emerging technologies and frameworks to foster discussion in the community. This paper summarizes the lightning talks at the 2023 edition of WORKS, covering five topics: leveraging large language models to build and execute workflows; developing a common workflow scheduler interface; scaling uncertainty workflow applications on exascale computing systems; evaluating a transcriptomics workflow for cloud vs. HPC systems; and best practices in migrating legacy workflows to workflow management systems.

Early detection and classification of mass lesion in mammograms constitute an essential step to decrease patient mortality caused by breast cancer, because it is possible to analyze the initial stages of cancer before it appears clinically. A well-performed segmentation task allows the lesion to be separated from the background to improve its shape-based classification. However, it is a challenging task because of its similarity to surrounding tissue. Therefore, we propose exploring two active contour models, Geodesic and Chan-Vese, to maximize the performance of mass segmentation in mammography images. Both models were optimized in terms of initialization radius and number of iterations used and validated on an experimental dataset containing 115 images with mass lesions. The best-selected Chan-Vese model, with a radius of 50 pixels and 436 iterations, outperformed the best Geodesic model, attaining a mean Dice score of 0.812 versus 0.558. This result highlighted the successful performance of the Chan-Vese model in segmenting mass lesions from different images. It also demonstrated the Geodesic model's tendency to get stuck in local minimums. The Median and CLAHE filters were crucial to improving the boundary quality of the mass lesion prior to the segmentation step. Also, the proposed method was able to successfully segment complex and irregular mass shapes, which is considered an essential result for cancer classification with respect to the degree of malignancy.