Septentrio Academic Publishing
Not a member yet
6415 research outputs found
Sort by
Indefinite readings of referential null subjects and null objects in Spanish
No full textIt has been claimed in the literature that referential null subjects receive definite readings in consistent null-subject languages, like Spanish. That is why they are said to display the typical behavior of definite pronouns (pro). However, referential null subjects can receive indefinite readings under the following conditions: i) the antecedent must be a bare plural, ii) the indefinite null argument is interpreted as a bare noun, typically as an internal argument, and iii) identity of sense anaphora is involved. Crucially, referential null objects appear under the same conditions, and receive indefinite readings in Peninsular Spanish. This suggests that the gap could be analyzed in the same way. In this paper, I argue that both null arguments should not be analyzed as empty pronominals (pro), given the syntactic behavior that they show. By contrast, I will argue that an argument ellipsis analysis should be assumed, given that the internal argument position is occupied by an indefinite, non-specific bare noun, the null gap can receive sloppy readings, and the omission is not as ‘big’ as the one at verb-stranding VP-ellipsis, nor as ‘small’ as the one at nominal ellipsis.En la bibliografía se ha dicho que los sujetos nulos anafóricos reciben lecturas definidas en lenguas de sujeto nulo como el español. Esta es la razón por la que se ha dicho que se comportan como pronombres definidos (pro). Sin embargo, los sujetos nulos anafóricos pueden recibir lecturas indefinidas bajo las siguientes condiciones: i) el antecedente debe ser un plural escueto, ii) el argumento nulo indefinido se interpreta como un nombre escueto, normalmente como un argumento interno, y iii) se trata de casos de anáfora de identidad de sentido. Crucialmente, los objetos nulos anafóricos apareccen bajo las mismas condiciones, y reciben lecturas indefinidas en español peninsular. Ello sugiere que el hueco podría analizarse de la misma manera. En este artículo argumento que ambos argumentos nulos no deben analizarse mediante pronombres vacíos (pro), dado su comportamiento sintáctico. Por el contrario, deben analizarse mediante elipsis argumental, dado que la posición de argumento interno está ocupada por un nombre escueto indefinido e inespecífico, el hueco nulo puede recibir lecturas descuidadas, y la omisión no es tan amplia como la que tiene lugar en la elipsis del SV, ni tan reducida como la que tiene lugar en la elipsis nominal
Elevers språkbruk: Hvordan forbedre relasjoner i klasserommet: Anne Birgitta Nilsen (2021)
No full textExplainability in subgraphs-enhanced Graph Neural Networks
No full textRecently, subgraphs-enhanced Graph Neural Networks (SGNNs) have been introduced to enhance the expressive power of Graph Neural Networks (GNNs), which was proved to be not higher than the 1-dimensional Weisfeiler-Leman isomorphism test. The new paradigm suggests using subgraphs extracted from the input graph to improve the model\u27s expressiveness, but the additional complexity exacerbates an already challenging problem in GNNs: explaining their predictions. In this work, we adapt PGExplainer, one of the most recent explainers for GNNs, to SGNNs. The proposed explainer accounts for the contribution of all the different subgraphs and can produce a meaningful explanation that humans can interpret. The experiments that we performed both on real and synthetic datasets show that our framework is successful in explaining the decision process of an SGNN on graph classification tasks
Learning to solve arithmetic problems with a virtual abacus
No full textAcquiring mathematical skills is considered a key challenge for modern Artificial Intelligence systems. Inspired by the way humans discover numerical knowledge, here we introduce a deep reinforcement learning framework that allows to simulate how cognitive agents could gradually learn to solve arithmetic problems by interacting with a virtual abacus. The proposed model successfully learn to perform multi-digit additions and subtractions, achieving an error rate below 1% even when operands are much longer than those observed during training. We also compare the performance of learning agents receiving a different amount of explicit supervision, and we analyze the most common error patterns to better understand the limitations and biases resulting from our design choices
Improving Wind Speed Uncertainty Forecasts Using Recurrent Neural Networks
No full textFor integration of growing amounts of volatile renewable energy in the European electricity system, reliable weather prognosis gains importance. But, depending on weather conditions, forecast reliability of wind speed for predicting wind power can vary drastically with time. Thus, relevance of risk-aware system operation strategies is increasing based on wind speed uncertainty a measure of which is provided by the standard deviations of ensemble forecasts of the German Weather Service. However, lacking validity of this measure is known as a long-standing problem.Therefore, this work investigates how machine learning based on a suitably selected set of physical quantities of weather ensemble data as well as historic wind data allows for a more realistic uncertainty quantification. A recurrent neural network (RNN) based sequence-to-sequence architecture is implemented and probabilistic wind speed forecasts are generated for a region in northern Germany.The results are evaluated and compared with the forecasts of the German Weather Service thereby revealing improved validity of such deep-learning based uncertainty measures
Contrastive learning for unsupervised medical image clustering and reconstruction
No full textThe lack of large labeled medical imaging datasets, along with significant inter-individual variability compared to clinically established disease classes, poses significant challenges in exploiting medical imaging information in a precision medicine paradigm, where in principle dense patient-specific data can be employed to formulate individual predictions and/or stratify patients into finer-grained groups which may follow more homogeneous trajectories and therefore empower clinical trials. In order to efficiently explore the effective degrees of freedom underlying variability in medical images in an unsupervised manner, in this work we propose an unsupervised autoencoder framework which is augmented with a contrastive loss to encourage high separability in the latent space. The model is validated on (medical) benchmark datasets. As cluster labels are assigned to each example according to cluster assignments, we compare performance with a supervised transfer learning baseline. Our methods achieves similar performance to the supervised architecture, indicating that separation in the latent space reproduces expert medical observer-assigned labels. The proposed method could be beneficial for patient stratification, exploring new subdivision of larger classes or pathological continua or, due to its sampling abilities in a variation setting, data augmentation in medical image processing
Automatic Postoperative Brain Tumor Segmentation with Limited Data using Transfer Learning and Triplet Attention
No full textAccurate brain tumor segmentation is clinically important for diagnosis and treatment planning. Convolutional neural networks (CNNs) have achieved promising performance in various visual recognition tasks. Training such networks usually requires large amount of labeled data, which is often challenging for medical applications. In this work, we address the segmentation problem by applying transfer learning to downstream segmentation tasks. Specifically, we explore how knowledge acquired from a large preoperative dataset can be transferred to postoperative tumor segmentation on a smaller dataset. To this end, we have developed a 3D CNN for brain tumor segmentation, and fine-tuned the pretrained models on the target domain data. To better exploit the inter-channel and spatial information, triplet attention has been incorporated and extended into existing segmentation network. Extensive experiments on our dataset demonstrate the effectiveness of transfer learning and attention modules for improved postoperative tumor segmentation performance when only limited amount of annotated data is available
Breaking up with Elsevier
No full textJanine Bijsterbosch, member of the editorial team of Imaging Neuroscience, informs about their recent break with publishing giant Elsevier. Collectively, the entire team of editors of Neuroimage left Elsevier to form a new journal, Imaging Neuroscience, at MIT Press. While Neuroimage already was an open access journal, it charged 3,450 dollars in Article Processing Charge (APC). At MIT Press, Imaging Neuroscience will charge 1,600 dollars in APC, with waivers for authors from low- and middle-income countries. Bijsterbosch explains why the editors collectively resigned from Neuroimage and what they hope to achieve with the move to a less costly model.
At the time when the editors left Neuroimage, they published around 1,000 peer reviewed articles per year, with an Impact Factor of 7.4. With the non-profit Imaging Neuroscience, they hope to bring authors and peer reviewers with them in their effort to build the (new) leading publishing outlet for researchers in the field. The reception has been quite positive and large amounts of manuscripts are already being submitted to the new journal.
Besides the economics and ethics of sustainability, Bijsterbosch is concerned with other aspects of open and transparent science, such as Open Code and Open Data. She sees Imaging Neuroscience as a player in this field as well, with Author’s Instructions including statements about the sharing of code and data whenever possible
Monitoring Open Science beyond publications: Datasets and software as research products to be shared
No full textWatch VIDEO.
Since 2018, the French Open Science Monitor (BSO) has assessed the effectiveness of the national public policy in open science. This steering tool, developed by the French Ministry of Higher Education and Research, the University of Lorraine and Inria, measures the evolution of open science in France using reliable, open and controlled data updated every year. The result is a website presenting different dashboards, tracking for example the ratio of open access scientific publications by year, discipline or publisher.
Since its last release in March 2023, the BSO also tracks the production and openness of research datasets and software mentioned in scientific publications on a national scale. To ensure a realistic coverage, our platform relies on large-scale open source Deep Learning techniques applied to the full texts of publications with at least one co-author with a French affiliation.
DataStet identifies every mention of datasets in scholarly publications, including implicit mentions of datasets and explicitly named datasets. SoftCite recognizes any software mentions in scientific publications, using as training data the Softcite Dataset. Dataset and software mentions are then characterized automatically as used, created and shared by the research work described in the scientific document. These characterizations can be cumulative. Among 1,608,839 publications from our corpus, we were able to analyze 655,954 of them with our tool DataStet. For this subset, we found 6,511,998 mentions of datasets characterized as used, 330,062 mentions characterized as created, and 78,178 mentions characterized as shared.
With this methodology, the BSO can offer new indicators about the proportion of French publications mentioning the usage, creation and sharing of data, as well as the proportion of publications in France that include a "Data Availability Statement". Similar indicators are dedicated to code and software. In addition, these indicators are further broken down into disciplines, publishers and institutions.
The project is addressing major technical and organizational challenges: to identify French datasets and software without reference registries as for publications, thanks to artificial intelligence; to produce relevant indicators for the different scientific communities. As an enabling technology to identify research datasets and software, deep learning plays a crucial role. This presentation will be an opportunity to present the latest results of the project, to detail the methodology, and finally to underline the reusability of the project results
Recognition and Assessment of Digital Scholarly Outputs in the Humanities
No full textWatch VIDEO.
In recent years we have observed an increase in digital practices and outputs in scholarship, which should be understood as a standard evolution of scholarly practices to take advantage of digital technologies. And although written genres, such as the monograph or essay, remain dominant in the humanities, the range of technological possibilities allow scholars to redefine those forms of expression and enrich them with other media or genres. However, as the opening example showed, this innovation is not supported by the assessment system, or even sometimes takes place in spite of it. A change in attitude requires recognition of three key aspects of digital humanities work: (1) its interdisciplinarity in borrowing tools and methods from ICT or social sciences; (2) the new research practices which should be recognised as valid scholarly work; (3) innovative scholarly outputs that go beyond the traditional genres but provide valid research results.
This presentation discusses the recommendations of the ALLEA E-Humanities Working Group with regards to the assessment of novel scholarly communication genres in the humanities. The work is based on the group’s previous report, Sustainable and FAIR Data Sharing in the Humanities, which provided recommendations on data practices in the humanities. The current focus is on attuning institutional policies to emerging scholarly needs in connection to current research assessment reform (CoARA). The recommendations are prepared in close cooperation with stakeholders and the research community. It underwent an open consultation whereby we collected more than 200 comments from the public. The final draft is under preparation and will be published in fall 2023.
The ALLEA E-Humanities Working Group recommendations are meant to serve as guidance for institutions and evaluators to embrace innovative outputs in the humanities and thus create space for their development. The Working Group has prepared tailored recommendations which could be divided into two main groups. First, the group focuses on the cross-cutting issues pertinent to digital practices in the humanities, which are (1) linking studies with underlying data, (2) updating and versioning of the outputs, (3) collaboration and authorship, (4) training and competence building, and (5) reviewing. Next, we discuss particular case studies of innovative outputs where cross-cutting issues manifest themselves, such as digital scholarly editions, extended publications, databases, visualisations, code and blogs. The overall conclusions provide some general remarks on recognising and evaluating digital practices in the humanities