INESC-ID RCAAP Portal
Not a member yet
15883 research outputs found
Sort by
Back-end DAQ system prototype testing and integration on a full detector test system for the CMS HGCAL detector
No full textThis publication presents the development and testing of the Capture Block, an elementary unit in the back-end Data Acquisition System (DAQ) for the CMS High Granularity Calorimeter (HGCAL) detector at CERN's Large Hadron Collider. The Capture Block aggregates data from up to 12 ECON-Ds spread across 14 e-links, enabling the system to cope with various front-end electronic configuration possibilities. Through its integration in a full detector test system and evaluation of its performance, this work demonstrates the feasibility of the Capture Block in achieving an average data acquisition throughput of 750 k events per second at the detector's nominal rate
DeB3RTa: A Transformer-Based Model for the Portuguese Financial Domain
No full textThis study presents DeB3RTa, a transformer-based model specifically developed for Portuguese financial language processing, leveraging mixed-domain pretraining to capture nuanced linguistic and contextual information. The model outperforms baseline models, including BERTimbau and XLM-RoBERTa, across key financial NLP benchmarks, demonstrating the efficacy of mixed-domain pretraining in building high-performance language models for specialized applications. By integrating advanced fine-tuning techniques, DeB3RTa offers a robust tool for advancing NLP in the financial sector and supporting nuanced language processing needs in Portuguese-speaking contexts
Reconfigurable FPU with Precision Auto-Tuning for Next-Generation Transprecision Computing
No full textSVeritas: Benchmark for Robust Speaker Verification under Diverse Conditions
No full textThis research introduces SVeritas, a comprehensive speaker verification benchmark suite that systematically evaluates state-of-the-art models across diverse conditions, including content and style variations, acoustic and channel mismatches, demographic factors, and security threats. The authors use SVeritas to evaluate several leading architectures and identify systemic weaknesses in cross-language, age-mismatch, and codec-compressed trials, as well as fairness gaps that vary by gender and language background. By establishing a rigorous, reproducible foundation for diagnosing robustness and equity in speaker verification systems, SVeritas paves the way for developing more reliable, inclusive, and secure SV systems suitable for real-world applications
Evidências do Português Médio no Corpus de Textos Antigos
No full textThis study aims to analyze the evolution of Middle Portuguese in the Corpus of Ancient Texts (CTA), focusing on the syncopation of intervocalic -d- and the past participle ending forms -udo/-ido in verbs with Latin etymology. By applying Natural Language Processing (NLP) methods, the authors optimize the systematic collection and extraction of relevant data for analysis, contributing to a larger understanding of the Portuguese language's historical development. The study provides new insights into the linguistic characteristics of Middle Portuguese and sheds light on its transition towards Modern Portuguese
