Papers by Maria-cristina Marinescu
Object detection annotations on images from the Rijksmuseum
Zenodo (CERN European Organization for Nuclear Research), Feb 28, 2023
EpiGraph Internal Structure
Page 1. Carlos III University of Madrid Higher Polytechnic School Computer Science Department Com... more Page 1. Carlos III University of Madrid Higher Polytechnic School Computer Science Department Computer Architecture, Communications and Systems Area Technical Report EpiGraph Internal Structure Gonzalo Martín, Maria ...
Lecture Notes in Computer Science, 2023
Large datasets that were made publicly available to the research community over the last 20 years... more Large datasets that were made publicly available to the research community over the last 20 years have been a key enabling factor for the advances in deep learning algorithms for NLP or computer vision. These datasets are generally pairs of aligned image / manually annotated metadata, where images are photographs of everyday life. Scholarly and historical content, on the other hand, treat subjects that are not necessarily popular to a general audience, they may not always contain a large number of data points, and new data may be difficult or impossible to collect. Some exceptions do exist, for instance, scientific or health data, but this is not the case for cultural heritage (CH). The poor performance of the best models in computer vision-when tested over artworks-coupled with the lack of extensively annotated datasets for CH, and the fact that artwork images depict objects and actions not captured by photographs, indicate that a CH-specific dataset would be highly valuable for this community. We propose DEArt, at this point primarily an object detection and pose classification dataset meant to be a reference for paintings between the XIIth and the XVIIIth centuries. It contains more than 15000 images, about 80% non-iconic, aligned with manual annotations for the bounding boxes identifying all instances of 69 classes as well as 12 possible poses for boxes identifying human-like objects. Of these, more than 50 classes are CH-specific and thus do not appear in other datasets; these reflect imaginary beings, symbolic entities and other categories related to art. Additionally, existing datasets do not include pose annotations. Our results show that object detectors for the cultural heritage domain can achieve a level of precision comparable to state-of-art models for generic images via transfer learning.
Research Square (Research Square), Mar 19, 2020
medRxiv (Cold Spring Harbor Laboratory), Jul 7, 2023
Background. Collaborative comparisons and combinations of multiple epidemic models are used as po... more Background. Collaborative comparisons and combinations of multiple epidemic models are used as policy-relevant evidence during epidemic outbreaks. Typically, each modeller summarises their own distribution of simulated trajectories using descriptive statistics at each modelled time step. We explored information losses compared to directly collecting a sample of the simulated trajectories, in terms of key epidemic quantities, ensemble uncertainty, and performance against data. Methods. We compared July 2022 projections from the European COVID-19 Scenario Modelling Hub. Using shared scenario assumptions, five modelling teams contributed up to 100 simulated trajectories projecting incidence in Belgium, the Netherlands, and Spain. First, we compared epidemic characteristics including incidence, peaks, and cumulative totals. Second, we drew a set of quantiles from the sampled trajectories for each model at each time step. We created an ensemble as the median across models at each quantile, and compared this to an ensemble of quantiles drawn from all available trajectories at each time step. Third, we compared each .
International Journal of Modeling and Optimization, 2012
This work investigates how opera singers manipulate timing in order to produce expressive perform... more This work investigates how opera singers manipulate timing in order to produce expressive performances that have common features but also bear a distinguishable personal style. We characterize performances not only relative to the score, but also consider the contribution of features extracted from the libretto. Our approach is based on applying machine learning to extract singer-specific patterns of expressive singing from performances by Josep Carreras and Placido Domingo. We compare and contrast some of these rules, and we draw some analogies between them and some of the general expressive performance rules existing in the literature. Index Terms-Expressive performance, machine learning, timing model I. INTRODUCTION In an interview that Charlie Rose took of the ``three tenors'' back in 1994, Placido Domingo was explaining how he tries to color the notes to give a song the feel that he's looking for. Josep Carreras instead talks about building each note with precision until it transmits the right emotion in the context, but gladly sacrificing precision for expressiveness. While opera singers may conceptualize the interpretation process very differently and possibly at different abstraction levels, the modifications they apply to the score may be similar. With certainty, nevertheless, there are expressive changes that they consistently apply that create their personal mark. This work focuses on how these two specific singers manipulate timing to create expressive interpretations that have a well-defined personal style. We start with a benchmark suite consisting of CD recordings of a cappella fragments from different tenor arias-seven performed by Josep Carreras and six by Placido Domingo. Using sound analysis techniques based on spectral models we extract acoustic high-level descriptors representing properties of each note, as well as of its context. A note is characterized by its pitch and duration. The context information for a given note consists of the relative pitch and duration of the neighboring notes, as well as the Narmour [1] structures to which the note belongs. Given that the libretto is an important part of an operatic performance which may reinforce-but may also change the expressive quality of the music-we also consider it when characterizing the notes. Each note has a syllableoccasionally a couple of syllables-associated with it. Every syllable is naturally strongly or weakly stressed. A performer
Statecharts are probably the most popular mechanism for behavior modeling of embedded system comp... more Statecharts are probably the most popular mechanism for behavior modeling of embedded system components. Modeling a component involves using a mainstream language for features that statecharts cannot express: detailed behavior of conditions and actions, object-orientation and distributed computing features. Debugging is done at the level of the generated native code. Rather than treating statecharts as a separate programming model from the native programming model, we extend a (Java-like) language with support for key concepts of statecharts: (1) explicit states, (2) asynchronous events, and (3) conditional execution. This paper presents ESP * , a language that supports statecharts and a set of other advanced programming concepts to make programming embedded systems easier. The paper also shows how to translate statecharts to ESP * .
BMJ Open, Dec 1, 2022
Evaluation of vaccination strategies for the metropolitan area of Madrid via agentbased simulatio... more Evaluation of vaccination strategies for the metropolitan area of Madrid via agentbased simulation. BMJ Open 2022;12:e065937.
arXiv (Cornell University), Nov 2, 2022
Large datasets that were made publicly available to the research community over the last 20 years... more Large datasets that were made publicly available to the research community over the last 20 years have been a key enabling factor for the advances in deep learning algorithms for NLP or computer vision. These datasets are generally pairs of aligned image / manually annotated metadata, where images are photographs of everyday life. Scholarly and historical content, on the other hand, treat subjects that are not necessarily popular to a general audience, they may not always contain a large number of data points, and new data may be difficult or impossible to collect. Some exceptions do exist, for instance, scientific or health data, but this is not the case for cultural heritage (CH). The poor performance of the best models in computer vision-when tested over artworks-coupled with the lack of extensively annotated datasets for CH, and the fact that artwork images depict objects and actions not captured by photographs, indicate that a CH-specific dataset would be highly valuable for this community. We propose DEArt, at this point primarily an object detection and pose classification dataset meant to be a reference for paintings between the XIIth and the XVIIIth centuries. It contains more than 15000 images, about 80% non-iconic, aligned with manual annotations for the bounding boxes identifying all instances of 69 classes as well as 12 possible poses for boxes identifying human-like objects. Of these, more than 50 classes are CH-specific and thus do not appear in other datasets; these reflect imaginary beings, symbolic entities and other categories related to art. Additionally, existing datasets do not include pose annotations. Our results show that object detectors for the cultural heritage domain can achieve a level of precision comparable to state-of-art models for generic images via transfer learning.
Research Square (Research Square), Jan 13, 2020
Background: Predicting the details of how an epidemic evolves is highly valuable as health instit... more Background: Predicting the details of how an epidemic evolves is highly valuable as health institutions need to better plan towards limiting the infection propagation effects and optimizing their prediction and response capabilities. Simulation is a cost-and time-effective way of predicting the evolution of the infection as the joint influence of many different factors: interaction patterns, personal characteristics, travel patterns, meteorological conditions, previous vaccination, etc. The work presented in this paper extends EpiGraph, our influenza epidemic simulator, by introducing a meteorological model as a modular component that interacts with the rest of EpiGraph's modules to refine our previous simulation results. Our goal is to estimate the effects of changes in temperature and relative humidity on the patterns of epidemic influenza based on data provided by the Spanish Influenza Sentinel Surveillance System (SISSS) and the Spanish Meteorological Agency (AEMET). Methods: Our meteorological model is based on the regression model developed by AB and JS, and it is tuned with influenza surveillance data obtained from SISSS. After pre-processing this data to clean it and reconstruct missing samples, we obtain new values for the reproduction number of each urban region in Spain, every 10 minutes during 2011. We simulate the propagation of the influenza by setting the date of the epidemic onset and the initial influenza-illness rates for each urban region. 1 Results: We show that the simulation results have the same propagation shape as the weekly influenza rates as recorded by SISSS. We perform experiments for a realistic scenario based on actual meteorological data from 2010-2011, and for synthetic values assumed under simplified predicted climate change conditions. Results show that a diminishing relative humidity of 10% produces an increment of about 1.6% in the final infection rate. The effect of temperature changes on the infection spread is also noticeable, with a decrease of 1.1% per extra degree. Conclusions: Using a tool like ours could help predict the shape of developing epidemics and its peaks, and would permit to quickly run scenarios to determine the evolution of the epidemic under different conditions. We make EpiGraph source code and epidemic data publicly available.
Based on our positive, but limited experience with Jigsaw at the university level, half a year ag... more Based on our positive, but limited experience with Jigsaw at the university level, half a year ago we initiated a more extensive experiment with a larger sample of students, and incorporating changes that relate back to some negative comments we have received during the previous course. Jigsaw is a collaborative inquiry-based learning technique that works by dividing the learning material into different tasks and the class into different groups. What set out to be a controlled experiment in increasing motivation and participation through collaboration, turned into a much more complex scenario due to the arrival of the Covid-19 pandemic, which gave us some interesting results to report. We have seen more positive results this year than the last: the number of students that felt that Jigsaw requires more effort than traditional methods has fallen, they consistently thought that Jigsaw improved teamwork, and they felt they have learned more from their expert peers as the experiment advanced. Some of the results may be due to the confinement forcing people stay indoors, with no social outings and fewer distractionsso more time to study. Another factor that may be relevant are the implicit expectations that were set by the confinement about distance learning and the need to cooperate.
Lecture Notes in Computer Science, 2022
EasyChair preprints are intended for rapid dissemination of research results and are integrated w... more EasyChair preprints are intended for rapid dissemination of research results and are integrated with the rest of EasyChair.
Evaluating the spread of Omicron COVID-19 variant in Spain
Future Generation Computer Systems
Frontiers in Public Health
Automated metadata annotation: What is and is not possible with machine learning
Data Intelligence
Automated metadata annotation is only as good as the training set, or rules that are available fo... more Automated metadata annotation is only as good as the training set, or rules that are available for the domain. It's important to learn what type of content a pre-trained machine learning algorithm has been trained on to understand its limitations and potential biases. Consider what type of content is readily available to train an algorithm—what's popular and what's available. However, scholarly and historical content is often not available in consumable, homogenized, and interoperable formats at the large volume that is required for machine learning. There are exceptions such as science and medicine, where large, well documented collections are available. This paper presents the current state of automated metadata annotation in cultural heritage and research data, discusses challenges identified from use cases, and proposes solutions.
Lindaview: an OBDA-based tool for self-sufficiency assessment
Barcelona Supercomputing Center, May 1, 2021
Evaluating the spread of Omicron COVID-19 variant in Spain
2022 22nd IEEE International Symposium on Cluster, Cloud and Internet Computing (CCGrid)
Evolution of Epidemics
Leveraging social networks for understanding the
ACM Transactions on Embedded Computing Systems, 2013
This article presents FUSE, an approach for modeling and implementing embedded software component... more This article presents FUSE, an approach for modeling and implementing embedded software components which starts from a main-stream programming language and brings some of the key concepts of Statecharts as first-class elements within this language. Our approach provides a unified programming environment which not only preserves some of the advantages of Statecharts' formal foundation but also directly supports features of object-orientation and strong typing. By specifying Statecharts directly in FUSE we eliminate the out-of-synch between the model and the generated code and we allow the tuning and debugging to be done within the same programming model. This article describes the main language constructs of FUSE and presents its semantics by translation into the Java programming language. We conclude by discussing extensions to the base language which enable the efficient static checking of program properties.
Uploads
Papers by Maria-cristina Marinescu