A5017258586- Natural Language Processing Techniques
- Topic Modeling
- Semantic Web and Ontologies
- Text Readability and Simplification
- ECG Monitoring and Analysis
- Acute Myocardial Infarction Research
- Speech and dialogue systems
- Machine Learning in Healthcare
- Heart Rate Variability and Autonomic Control
- AI-based Problem Solving and Planning
- Rough Sets and Fuzzy Logic
- Neurosurgical Procedures and Complications
- Manufacturing Process and Optimization
- Trauma and Emergency Care Studies
- Emergency and Acute Care Studies
- Chronic Obstructive Pulmonary Disease (COPD) Research
- Venous Thromboembolism Diagnosis and Management
- Traumatic Brain Injury and Neurovascular Disturbances
- Heart Failure Treatment and Management
- Speech Recognition and Synthesis
- Chronic Disease Management Strategies
- Flexible and Reconfigurable Manufacturing Systems
- Atrial Fibrillation Management and Outcomes
- Software Engineering Research
- Biomedical Text Mining and Ontologies
Lund University
2009-2024
Skåne University Hospital
2021
University of Stuttgart
2012-2017
This paper describes our contribution to the semantic role labeling task (SRL-only) of CoNLL-2009 shared in closed challenge (Hajič et al., 2009). Our system consists a pipeline independent, local classifiers that identify predicate sense, arguments predicates, and argument labels. Using these models, we carried out beam search generate pool candidates. We then reranked candidates using joint learning approach combines models proposition features.
We investigate different ways of learning structured perceptron models for coreference resolution when using non-local features and beam search. Our experimental results indicate that standard techniques such as early updates or Learning Search Optimization (LaSO) perform worse than a greedy baseline only uses local features. By modifying LaSO to delay until the end each instance we obtain significant improvements over baseline. model obtains best date on recent shared task data Arabic,...
Background Breathlessness is a troublesome and prevalent symptom in the population, but knowledge of related factors scarce. The aim this study was to identify most strongly associated with breathlessness general population describe shapes associations between main breathlessness. Methods A cross-sectional analysis carried out multicentre population-based Swedish CArdioPulmonary bioImage Study (SCAPIS) adults aged 50 64 years. defined as modified Medical Research Council rating ≥2. machine...
Robots used in manufacturing today are tailored to their tasks by system integration based on expert knowledge concerning both production and machine control. For upcoming new generations of even more flexible robot solutions, applications such as dexterous assembly, the setup programming gets challenging. Reuse solutions terms parameters, controls, process tuning, software modules general then increasingly important. There has been valuable progress within reuse automation when machines...
This paper describes the IMS-SZEGED-CIS contribution to SPMRL 2013 Shared Task. We participate in both constituency and dependency tracks, achieve state-of-theart for all languages. For tracks we make significant improvements through high quality preprocessing (re)ranking on top of strong baselines. Our system came out first tracks.
Abstract Aims In the present study, we aimed to evaluate performance of machine learning (ML) models for identification acute myocardial infarction (AMI) or death within 30 days among emergency department (ED) chest pain patients. Methods and results Using data from 9519 consecutive ED patients, created ML based on logistic regression artificial neural networks. Model inputs included sex, age, ECG first blood tests at patient presentation: High sensitivity TnT (hs-cTnT), glucose, creatinine,...
This paper presents the IMS contribution to CoNLL 2017 Shared Task. In preprocessing step we employed a CRF POS/morphological tagger and neural predicting supertags. On some languages, also applied word segmentation with sentence perceptron-based parser. For parsing took an ensemble approach by blending multiple instances of three parsers very different architectures. Our system achieved third place overall second for surprise languages.
We study non-deterministic oracles for training non-projective beam search parsers with swap transitions.We map out the spurious ambiguities of transition system and present two as well a static oracle that minimizes number swaps.An evaluation on 10 treebanks reveals difference between is generally insignificant but can improve accuracy greedy use transitions.
We present the ICARUS Coreference Explorer, an interactive tool to browse and search coreference-annotated data. It can display coreference annotations as a tree, entity grid, or in standard textbased mode, lets user switch freely between different modes. The compare two on same document, allowing system developers evaluate errors automatic predictions. features flexible engine, which enables graphically construct queries over sets of documents annotated with coreference.
Objective Computerized decision-support tools may improve diagnosis of acute myocardial infarction (AMI) among patients presenting with chest pain at the emergency department (ED). The primary aim was to assess predictive accuracy machine learning algorithms based on paired high-sensitivity cardiac troponin T (hs-cTnT) concentrations varying sampling times, age, and sex in order rule or out AMI. Methods In this register-based, cross-sectional diagnostic study conducted retrospectively 5695 2...
Despite diagnostic algorithms, identification of venous thromboembolism (VTE) in emergency departments (ED) remains a challenge. We evaluated symptoms, background, and laboratory data 27,647 ED patients presenting with pain, swelling, or other symptoms from the extremities, identified predictors VTE diagnosis within one year. Predictors clinical decision to perform phlebography, ultrasound, computer tomography (CT) angiography pelvic, lower, upper extremity veins, CT pulmonary arteries,...
Supertagging was recently proposed to provide syntactic features for statistical dependency parsing, contrary its traditional use as a disambiguation step.We conduct broad range of controlled experiments compare this specific application supertagging with another method providing features, namely stacking.We find that in context is form furthermore show (i) fast parser and sequence labeler are equally beneficial supertagging, (ii) supertagging/stacking improve parsing also cross-domain...
Health-related quality of life (HRQoL) is essential for human wellbeing, influenced by a complex interplay factors, and reported lower in women than men. We aimed to evaluate which factors were the most important HRQoL middle-aged general population.This was cross-sectional, multi-centre study 29,212 men (48%) (52%) aged 50-64 population Sweden. Physical mental (0-100) assessed using Short Form 12 questionnaire, association evaluated 356 variables including demographics, lifestyle, symptoms,...
Gregor Thiele, Wolfgang Seeker, Markus Gärtner, Anders Björkelund, Jonas Kuhn. Proceedings of the Demonstrations at 14th Conference European Chapter Association for Computational Linguistics. 2014.