29 Maio 2025, 13:00
•
Eugénio Ribeiro
Title: Computational assessment of chronic pain based on patient language of pain
Abstract: Healthcare presents a uniquely complex environment for the application of Natural Language Processing (NLP) techniques, characterized by its diversity of stakeholders, clinical contexts, and high-stakes decision-making. My research, titled "Computational Assessment of Chronic Pain Based on Patient Language of Pain", explores the potential of NLP to support clinical understanding and assessment of chronic pain through patient narratives. The work integrates multiple NLP pipelines, combining psycholinguistic feature extraction, pretrained and fine-tuned language encodings, and large language model (LLM) prompting techniques. This presentation outlines the challenges of applying NLP in such a nuanced domain, including linguistic variability and data limitations, and offers a comparative systematization of the approaches used. It highlights key insights into their performance, advantages, and limitations, emphasizing the value and restrictions of computational methods in enriching the clinical practice.
Bio: Diogo A. P. Nunes holds a Master’s (2021) and a Bachelor’s (2018) degree in Informatics and Computer Engineering from Instituto Superior Técnico (IST), University of Lisbon. He is currently a PhD candidate in Informatics and Computer Engineering at IST, under the supervision of Prof. David Martins de Matos (IST) and Prof. Joana Ferreira Gomes (Faculty of Medicine, University of Porto). Since 2019, he has been an Early Stage Researcher at INESC-ID Lisbon, working within the Human Language Technology (HLT) Laboratory. His doctoral research focuses on the computational assessment of chronic pain through the analysis of patient language of pain. His work aims to bridge the gap between computational methods and clinical practice by developing NLP pipelines that capture the nuances of pain expression in healthcare and other contexts
22 Maio 2025, 13:00
•
Eugénio Ribeiro
Title: Assessment of Parkinson’s Disease Medication State through Automatic Speech Analysis
Abstract: Parkinson’s disease (PD) is a progressive degenerative disorder of the central nervous system characterized by motor and non-motor symptoms. The cardinal motor signs of PD include the characteristic clinical picture of resting tremor, rigidity, bradykinesia, and impairment of postural reflexes, while non-motor symptoms include cognitive disorders, and sleep and sensory abnormalities. Motor symptoms of PD influence also the speech production of language. Dysarthria, which is characterized by a weakness, paralysis, or lack of coordination in the motor-speech system, is typically observed in PD patients and affects respiration, phonation, articulation and prosody. As a consequence, the main deficits of PD speech are loss of intensity, monotony of pitch and loudness, reduced stress, inappropriate silences, short rushes of speech, variable rate, imprecise consonant articulation and harsh and breathy voice. Both motor symptoms and speech impairments slowly worsen during the disease with a nonlinear progression. Patients in advanced stages typically present more severe speech abnormalities, with voice disorders and articulatory deficits being the most prevalent symptoms. As the disease progresses, patients alternate periods in which motor symptoms are mitigated due to medication intake (ON state) and periods with motor complications (OFF state). The time that patients spend in the OFF condition is currently the main parameter employed to assess pharmacological interventions and to evaluate the efficacy of different active principles. In this talk, I will present past work in which automatic speech processing and deep learning techniques were employed to classify the medication state of PD patients by leveraging personal speech-based bio-markers. We devised a speaker dependent approach and investigated the relevance of different acoustic-prosodic feature sets. Results showed an accuracy of 90.54% in a test task with mixed speech and an accuracy of 95.27% in a semi-spontaneous speech task. Overall, the experimental assessment showed the potentials of this approach towards the development of reliable, remote daily monitoring and scheduling of medication intake of PD patients.
Bio: Rubén Solera Ureña is a Researcher at the Human Language Technology (HLT) Lab at INESC–ID, Lisbon. He received the M.Sc. degree in Telecommunications Engineering in 2004 and the Ph.D. degree in Multimedia and Communications in 2011, both from Universidad Carlos III de Madrid, Spain. From 2004 to 2014 he held several teaching and research positions at the Department of Signal Theory and Communications, Universidad Carlos III de Madrid, Spain. In May 2015, Rubén joined INESC-ID as a post-doctoral researcher responsible for the development of human-robotic agents speech interaction within project INSIDE (Intelligent Networked Robot Systems for Symbiotic Interaction with Children with Impaired Development), which explored symbiotic interactions between children with Autism Spectrum Disorder and robots used for therapeutical purposes. Since October 2023, he holds a position as an Auxiliary Researcher working on the Accelarat.AI project. His current research interests focus on the application of signal processing and machine learning methods to real-life problems in the fields of speech processing, automatic speech recognition, and computational paralinguistics. Rubén Solera Ureña has participated in 17 national and international projects and co-authored 20 publications in peer-reviewed journals and conferences.
15 Maio 2025, 14:30
•
Ricardo Daniel Santos Faro Marques Ribeiro
Introdução aos Transformadores
- Arquitetura
- Atenção
- Encoder
- Decoder
- Transformadores mais conhecidos
Grandes Modelos de Língua
- Modelos generativos
- Alguns exemplos
