INESC-ID RCAAP Portal
Not a member yet
15883 research outputs found
Sort by
A Dynamically Reconfigurable Power Management System with Self-Monitoring Capabilities for Energy-Efficient SoCs
No full textThis research presents a dynamically reconfigurable power management unit (PMU) with self-monitoring capabilities, designed to enable Ultra-Low Power (ULP) operation in energy-efficient system-on-chip (SoC) designs. The proposed PMU leverages a novel Real-Time Clock (RTC) architecture that integrates temperature, voltage, and current monitoring, enabling autonomous power mode management without the need for external processors or complex state machines. By achieving a 30% reduction in power consumption and a 70% area reduction compared to state-of-the-art implementations, this work demonstrates a promising solution for biomedical wearables and Internet of Medical Things (IoMT) devices
Leveraging Large Language Models for Pharmaceutical Documents: A Case Study in Package Leaflet Generation with Plain Language Adaptation
No full text
On the Relevance of Clinical Assessment Tasks for the Automatic Detection of Parkinson’s Disease Medication State from Speech
No full textThis study presents a novel approach for automatically detecting Parkinson's Disease medication states from speech, leveraging self-supervised speech representations and speaker-independent models. The authors explore the combination of knowledge-based eGeMAPS features and SSL Wav2Vec2.0 speech embeddings with support vector machines (SVMs) and attention-based deep neural networks (A-DNNs), achieving an F1-score of 88.2% on the European Portuguese FraLusoPark corpus. By developing a speaker-independent approach, this work offers practical advantages in terms of robustness and deployment, making it suitable for real-world screening scenarios
StoryReasoning Dataset: Using Chain-of-Thought for Scene Understanding and Grounded Story Generation
No full textThis research proposes the StoryReasoning dataset, a novel multi-frame narrative generation framework that addresses the challenges of visual storytelling by grounding characters, objects, and entities on visual elements across frames. The authors develop Qwen Storyteller, a model leveraging Chain-of-Thought reasoning to generate coherent visual narratives, achieving significant reductions in hallucinations compared to non-fine-tuned models. By establishing a ground truth for object identities across frames and employing structured tabular representations, the dataset and model provide a foundation for improving visual storytelling systems' ability to maintain character identity and produce coherent narratives
