A comparative analysis of preprocessing techniques of cardiac event series for the study of heart rhythm variability using simulated signals

In the present study, using noise-free simulated signals, we performed a comparative examination of several preprocessing techniques that are used to transform the cardiac event series in a regularly sampled time series, appropriate for spectral analysis of heart rhythm variability (HRV). First, a group of noise-free simulated point event series, which represents a time series of heartbeats, was generated by an integral pulse frequency modulation model. In order to evaluate the performance of the preprocessing methods, the differences between the spectra of the preprocessed simulated signals and the true spectrum (spectrum of the model input modulating signals) were surveyed by visual analysis and by contrasting merit indices. It is desired that estimated spectra match the true spectrum as close as possible, showing a minimum of harmonic components and other artifacts. The merit indices proposed to quantify these mismatches were the leakage rate, defined as a measure of leakage components (located outside some narrow windows centered at frequencies of model input modulating signals) with respect to the whole spectral components, and the numbers of leakage components with amplitudes greater than 1%, 5% and 10% of the total spectral components. Our data, obtained from a noise-free simulation, indicate that the utilization of heart rate values instead of heart period values in the derivation of signals representative of heart rhythm results in more accurate spectra. Furthermore, our data support the efficiency of the widely used preprocessing technique based on the convolution of inverse interval function values with a rectangular window, and suggest the preprocessing technique based on a cubic polynomial interpolation of inverse interval function values and succeeding spectral analysis as another efficient and fast method for the analysis of HRV signals

Improved image analysis based corrosion assessment using preprocessing techniques

The importance of non-destructive techniques (NDT) in structural health monitoring programmes is being critically felt in the recent times. The quality of the measured data, often affected by various environmental conditions can be a guiding factor in terms usefulness and prediction efficiencies of the various detection and monitoring methods used in this regard. Often, a preprocessing of the acquired data in relation to the affecting environmental parameters can improve the information quality and lead towards a significantly more efficient and correct prediction process. The improvement can be directly related to the final decision making policy about a structure or a network of structures and is compatible with general probabilistic frameworks of such assessment and decision making programmes. This paper considers a preprocessing technique employed for an image analysis based structural health monitoring methodology to identify sub-marine pitting corrosion in the presence of variable luminosity, contrast and noise affecting the quality of images. A preprocessing of the gray-level threshold of the various images is observed to bring about a significant improvement in terms of damage detection as compared to an automatically computed gray-level threshold. The case dependent adjustments of the threshold enable to obtain the best possible information from an existing image. The corresponding improvements are observed in a qualitative manner in the present study.

The application of preprocessing and cutting plane techniques for a class of production planning problems

This paper investigates properties of integer programming models for a class of production planning problems. The models are developed within a decision support system to advise a sales team of the products on which to focus their efforts in gaining new orders in the short term. The products generally require processing on several manufacturing cells and involve precedence relationships. The cells are already (partially) committed with products for stock and to satisfy existing orders and therefore only the residual capacities of each cell in each time period of the planning horizon are considered. The determination of production recommendations to the sales team that make use of residual capacities is a nontrivial optimization problem. Solving such models is computationally demanding and techniques for speeding up solution times are highly desirable. An integer programming model is developed and various preprocessing techniques are investigated and evaluated. In addition, a number of cutting plane approaches have been applied. The performance of these approaches which are both general and application specific is examined.

Arrow plot and CA maps on microarray preprocessing methods

Microarray allow to monitoring simultaneously thousands of genes, where the abundance of the transcripts under a same experimental condition at the same time can be quantified. Among various available array technologies, double channel cDNA microarray experiments have arisen in numerous technical protocols associated to genomic studies, which is the focus of this work. Microarray experiments involve many steps and each one can affect the quality of raw data. Background correction and normalization are preprocessing techniques to clean and correct the raw data when undesirable fluctuations arise from technical factors. Several recent studies showed that there is no preprocessing strategy that outperforms others in all circumstances and thus it seems difficult to provide general recommendations. In this work, it is proposed to use exploratory techniques to visualize the effects of preprocessing methods on statistical analysis of cancer two-channel microarray data sets, where the cancer types (classes) are known. For selecting differential expressed genes the arrow plot was used and the graph of profiles resultant from the correspondence analysis for visualizing the results. It was used 6 background methods and 6 normalization methods, performing 36 pre-processing methods and it was analyzed in a published cDNA microarray database (Liver) available at http://genome-www5.stanford.edu/ which microarrays were already classified by cancer type. All statistical analyses were performed using the R statistical software.

Forecasting Intraday Stock Price Trends with Text Mining Techniques

In this paper, we describe NewsCATS (news categorization and trading system), a system implemented to predict stock price trends for the time immediately after the publication of press releases. NewsCATS consists mainly of three components. The first component retrieves relevant information from press releases through the application of text preprocessing techniques. The second component sorts the press releases into predefined categories. Finally, appropriate trading strategies are derived by the third component by means of the earlier categorization. The findings indicate that a categorization of press releases is able to provide additional information that can be used to forecast stock price trends, but that an adequate trading strategy is essential for the results of the categorization to be fully exploited.

Signal-to-noise ratio of the MEG signal after preprocessing

Background Magnetoencephalography (MEG) provides a direct measure of brain activity with high combined spatiotemporal resolution. Preprocessing is necessary to reduce contributions from environmental interference and biological noise. New method The effect on the signal-to-noise ratio of different preprocessing techniques is evaluated. The signal-to-noise ratio (SNR) was defined as the ratio between the mean signal amplitude (evoked field) and the standard error of the mean over trials. Results Recordings from 26 subjects obtained during and event-related visual paradigm with an Elekta MEG scanner were employed. Two methods were considered as first-step noise reduction: Signal Space Separation and temporal Signal Space Separation, which decompose the signal into components with origin inside and outside the head. Both algorithm increased the SNR by approximately 100%. Epoch-based methods, aimed at identifying and rejecting epochs containing eye blinks, muscular artifacts and sensor jumps provided an SNR improvement of 5–10%. Decomposition methods evaluated were independent component analysis (ICA) and second-order blind identification (SOBI). The increase in SNR was of about 36% with ICA and 33% with SOBI. Comparison with existing methods No previous systematic evaluation of the effect of the typical preprocessing steps in the SNR of the MEG signal has been performed. Conclusions The application of either SSS or tSSS is mandatory in Elekta systems. No significant differences were found between the two. While epoch-based methods have been routinely applied the less often considered decomposition methods were clearly superior and therefore their use seems advisable.

Monitoring the ex-vivo expansion of human mesenchymal stem/stromal cells in xeno-free microcarrier-based reactor systems by MIR spectroscopy

Human mesenchymal stem/stromal cells (MSCs) have received considerable attention in the field of cell-based therapies due to their high differentiation potential and ability to modulate immune responses. However, since these cells can only be isolated in very low quantities, successful realization of these therapies requires MSCs ex-vivo expansion to achieve relevant cell doses. The metabolic activity is one of the parameters often monitored during MSCs cultivation by using expensive multi-analytical methods, some of them time-consuming. The present work evaluates the use of mid-infrared (MIR) spectroscopy, through rapid and economic high-throughput analyses associated to multivariate data analysis, to monitor three different MSCs cultivation runs conducted in spinner flasks, under xeno-free culture conditions, which differ in the type of microcarriers used and the culture feeding strategy applied. After evaluating diverse spectral preprocessing techniques, the optimized partial least square (PLS) regression models based on the MIR spectra to estimate the glucose, lactate and ammonia concentrations yielded high coefficients of determination (R2 ≥ 0.98, ≥0.98, and ≥0.94, respectively) and low prediction errors (RMSECV ≤ 4.7%, ≤4.4% and ≤5.7%, respectively). Besides PLS models valid for specific expansion protocols, a robust model simultaneously valid for the three processes was also built for predicting glucose, lactate and ammonia, yielding a R2 of 0.95, 0.97 and 0.86, and a RMSECV of 0.33, 0.57, and 0.09 mM, respectively. Therefore, MIR spectroscopy combined with multivariate data analysis represents a promising tool for both optimization and control of MSCs expansion processes.

Traduction statistique vers une langue à morphologie riche : combinaison d’algorithmes de segmentation morphologique et de modèles statistiques de traduction automatique

Les systèmes statistiques de traduction automatique ont pour tâche la traduction d’une langue source vers une langue cible. Dans la plupart des systèmes de traduction de référence, l'unité de base considérée dans l'analyse textuelle est la forme telle qu’observée dans un texte. Une telle conception permet d’obtenir une bonne performance quand il s'agit de traduire entre deux langues morphologiquement pauvres. Toutefois, ceci n'est plus vrai lorsqu’il s’agit de traduire vers une langue morphologiquement riche (ou complexe). Le but de notre travail est de développer un système statistique de traduction automatique comme solution pour relever les défis soulevés par la complexité morphologique. Dans ce mémoire, nous examinons, dans un premier temps, un certain nombre de méthodes considérées comme des extensions aux systèmes de traduction traditionnels et nous évaluons leurs performances. Cette évaluation est faite par rapport aux systèmes à l’état de l’art (système de référence) et ceci dans des tâches de traduction anglais-inuktitut et anglais-finnois. Nous développons ensuite un nouvel algorithme de segmentation qui prend en compte les informations provenant de la paire de langues objet de la traduction. Cet algorithme de segmentation est ensuite intégré dans le modèle de traduction à base d’unités lexicales « Phrase-Based Models » pour former notre système de traduction à base de séquences de segments. Enfin, nous combinons le système obtenu avec des algorithmes de post-traitement pour obtenir un système de traduction complet. Les résultats des expériences réalisées dans ce mémoire montrent que le système de traduction à base de séquences de segments proposé permet d’obtenir des améliorations significatives au niveau de la qualité de la traduction en terme de le métrique d’évaluation BLEU (Papineni et al., 2002) et qui sert à évaluer. Plus particulièrement, notre approche de segmentation réussie à améliorer légèrement la qualité de la traduction par rapport au système de référence et une amélioration significative de la qualité de la traduction est observée par rapport aux techniques de prétraitement de base (baseline).

Um sistema inteligente de classificação de sinais de EEG para Interface Cérebro-Computador

The Brain-Computer Interfaces (BCI) have as main purpose to establish a communication path with the central nervous system (CNS) independently from the standard pathway (nervous, muscles), aiming to control a device. The main objective of the current research is to develop an off-line BCI that separates the different EEG patterns resulting from strictly mental tasks performed by an experimental subject, comparing the effectiveness of different signal-preprocessing approaches. We also tested different classification approaches: all versus all, one versus one and a hierarchic classification approach. No preprocessing techniques were found able to improve the system performance. Furthermore, the hierarchic approach proved to be capable to produce results above the expected by literature

Estudo de algoritmos para classificação de séries temporais: uma aplicação em qualidade de energia elétrica

Aborda a classificação automática de faltas do tipo curto-circuito em linhas de transmissão. A maioria dos sistemas de transmissão possuem três fases (A, B e C). Por exemplo, um curto-circuito entre as fases A e B pode ser identicado como uma falta\AB". Considerando a possibilidade de um curto-circuito com a fase terra (T), a tarefa ao longo desse trabalho de classificar uma série temporal em uma das 11 faltas possíveis: AT, BT, CT, AB, AC, BC, ABC, ABT, ACT, BCT, ABCT. Estas faltas são responsáveis pela maioria dos distúrbios no sistema elétrico. Cada curto-circuito é representado por uma seqüência (série temporal) e ambos os tipos de classificação, on-line (para cada curto segmento extraído do sinal) e off-line (leva em consideração toda a seqüência), são investigados. Para evitar a atual falta de dados rotulados, o simulador Alternative Transient Program (ATP) é usado para criar uma base de dados rotulada e disponibilizada em domínio público. Alguns trabalhos na literatura não fazem distinção entre as faltas ABC e ABCT. Assim, resultados distinguindo esse dois tipos de faltas adotando técnicas de pré-processamento, diferentes front ends (por exemplo wavelets) e algoritmos de aprendizado (árvores de decisão e redes neurais) são apresentados. O custo computacional estimado durante o estágio de teste de alguns classificadores é investigado e a escolha dos parâmetros dos classificadores é feita a partir de uma seleção automática de modelo. Os resultados obtidos indicam que as árvores de decisão e as redes neurais apresentam melhores resultados quando comparados aos outros classificadores.

Exact Solutions to the Symmetric and Asymmetric Vehicle Routing Problem with Simultaneous Delivery and Pick-Up

In reverse logistics networks, products (e.g., bottles or containers) have to be transported from a depot to customer locations and, after use, from customer locations back to the depot. In order to operate economically beneficial, companies prefer a simultaneous delivery and pick-up service. The resulting Vehicle Routing Problem with Simultaneous Delivery and Pick-up (VRPSDP) is an operational problem, which has to be solved daily by many companies. We present two mixed-integer linear model formulations for the VRPSDP, namely a vehicle-flow and a commodity-flow model. In order to strengthen the models, domain-reducing preprocessing techniques, and effective cutting planes are outlined. Symmetric benchmark instances known from the literature as well as new asymmetric instances derived from real-world problems are solved to optimality using CPLEX 12.1.

Contributions to Speech Analytics based on Speech Recognition and Topic Identification

La última década ha sido testigo de importantes avances en el campo de la tecnología de reconocimiento de voz. Los sistemas comerciales existentes actualmente poseen la capacidad de reconocer habla continua de múltiples locutores, consiguiendo valores aceptables de error, y sin la necesidad de realizar procedimientos explícitos de adaptación. A pesar del buen momento que vive esta tecnología, el reconocimiento de voz dista de ser un problema resuelto. La mayoría de estos sistemas de reconocimiento se ajustan a dominios particulares y su eficacia depende de manera significativa, entre otros muchos aspectos, de la similitud que exista entre el modelo de lenguaje utilizado y la tarea específica para la cual se está empleando. Esta dependencia cobra aún más importancia en aquellos escenarios en los cuales las propiedades estadísticas del lenguaje varían a lo largo del tiempo, como por ejemplo, en dominios de aplicación que involucren habla espontánea y múltiples temáticas. En los últimos años se ha evidenciado un constante esfuerzo por mejorar los sistemas de reconocimiento para tales dominios. Esto se ha hecho, entre otros muchos enfoques, a través de técnicas automáticas de adaptación. Estas técnicas son aplicadas a sistemas ya existentes, dado que exportar el sistema a una nueva tarea o dominio puede requerir tiempo a la vez que resultar costoso. Las técnicas de adaptación requieren fuentes adicionales de información, y en este sentido, el lenguaje hablado puede aportar algunas de ellas. El habla no sólo transmite un mensaje, también transmite información acerca del contexto en el cual se desarrolla la comunicación hablada (e.g. acerca del tema sobre el cual se está hablando). Por tanto, cuando nos comunicamos a través del habla, es posible identificar los elementos del lenguaje que caracterizan el contexto, y al mismo tiempo, rastrear los cambios que ocurren en estos elementos a lo largo del tiempo. Esta información podría ser capturada y aprovechada por medio de técnicas de recuperación de información (information retrieval) y de aprendizaje de máquina (machine learning). Esto podría permitirnos, dentro del desarrollo de mejores sistemas automáticos de reconocimiento de voz, mejorar la adaptación de modelos del lenguaje a las condiciones del contexto, y por tanto, robustecer al sistema de reconocimiento en dominios con condiciones variables (tales como variaciones potenciales en el vocabulario, el estilo y la temática). En este sentido, la principal contribución de esta Tesis es la propuesta y evaluación de un marco de contextualización motivado por el análisis temático y basado en la adaptación dinámica y no supervisada de modelos de lenguaje para el robustecimiento de un sistema automático de reconocimiento de voz. Esta adaptación toma como base distintos enfoque de los sistemas mencionados (de recuperación de información y aprendizaje de máquina) mediante los cuales buscamos identificar las temáticas sobre las cuales se está hablando en una grabación de audio. Dicha identificación, por lo tanto, permite realizar una adaptación del modelo de lenguaje de acuerdo a las condiciones del contexto. El marco de contextualización propuesto se puede dividir en dos sistemas principales: un sistema de identificación de temática y un sistema de adaptación dinámica de modelos de lenguaje. Esta Tesis puede describirse en detalle desde la perspectiva de las contribuciones particulares realizadas en cada uno de los campos que componen el marco propuesto: _ En lo referente al sistema de identificación de temática, nos hemos enfocado en aportar mejoras a las técnicas de pre-procesamiento de documentos, asimismo en contribuir a la definición de criterios más robustos para la selección de index-terms. – La eficiencia de los sistemas basados tanto en técnicas de recuperación de información como en técnicas de aprendizaje de máquina, y específicamente de aquellos sistemas que particularizan en la tarea de identificación de temática, depende, en gran medida, de los mecanismos de preprocesamiento que se aplican a los documentos. Entre las múltiples operaciones que hacen parte de un esquema de preprocesamiento, la selección adecuada de los términos de indexado (index-terms) es crucial para establecer relaciones semánticas y conceptuales entre los términos y los documentos. Este proceso también puede verse afectado, o bien por una mala elección de stopwords, o bien por la falta de precisión en la definición de reglas de lematización. En este sentido, en este trabajo comparamos y evaluamos diferentes criterios para el preprocesamiento de los documentos, así como también distintas estrategias para la selección de los index-terms. Esto nos permite no sólo reducir el tamaño de la estructura de indexación, sino también mejorar el proceso de identificación de temática. – Uno de los aspectos más importantes en cuanto al rendimiento de los sistemas de identificación de temática es la asignación de diferentes pesos a los términos de acuerdo a su contribución al contenido del documento. En este trabajo evaluamos y proponemos enfoques alternativos a los esquemas tradicionales de ponderado de términos (tales como tf-idf ) que nos permitan mejorar la especificidad de los términos, así como también discriminar mejor las temáticas de los documentos. _ Respecto a la adaptación dinámica de modelos de lenguaje, hemos dividimos el proceso de contextualización en varios pasos. – Para la generación de modelos de lenguaje basados en temática, proponemos dos tipos de enfoques: un enfoque supervisado y un enfoque no supervisado. En el primero de ellos nos basamos en las etiquetas de temática que originalmente acompañan a los documentos del corpus que empleamos. A partir de estas, agrupamos los documentos que forman parte de la misma temática y generamos modelos de lenguaje a partir de dichos grupos. Sin embargo, uno de los objetivos que se persigue en esta Tesis es evaluar si el uso de estas etiquetas para la generación de modelos es óptimo en términos del rendimiento del reconocedor. Por esta razón, nosotros proponemos un segundo enfoque, un enfoque no supervisado, en el cual el objetivo es agrupar, automáticamente, los documentos en clusters temáticos, basándonos en la similaridad semántica existente entre los documentos. Por medio de enfoques de agrupamiento conseguimos mejorar la cohesión conceptual y semántica en cada uno de los clusters, lo que a su vez nos permitió refinar los modelos de lenguaje basados en temática y mejorar el rendimiento del sistema de reconocimiento. – Desarrollamos diversas estrategias para generar un modelo de lenguaje dependiente del contexto. Nuestro objetivo es que este modelo refleje el contexto semántico del habla, i.e. las temáticas más relevantes que se están discutiendo. Este modelo es generado por medio de la interpolación lineal entre aquellos modelos de lenguaje basados en temática que estén relacionados con las temáticas más relevantes. La estimación de los pesos de interpolación está basada principalmente en el resultado del proceso de identificación de temática. – Finalmente, proponemos una metodología para la adaptación dinámica de un modelo de lenguaje general. El proceso de adaptación tiene en cuenta no sólo al modelo dependiente del contexto sino también a la información entregada por el proceso de identificación de temática. El esquema usado para la adaptación es una interpolación lineal entre el modelo general y el modelo dependiente de contexto. Estudiamos también diferentes enfoques para determinar los pesos de interpolación entre ambos modelos. Una vez definida la base teórica de nuestro marco de contextualización, proponemos su aplicación dentro de un sistema automático de reconocimiento de voz. Para esto, nos enfocamos en dos aspectos: la contextualización de los modelos de lenguaje empleados por el sistema y la incorporación de información semántica en el proceso de adaptación basado en temática. En esta Tesis proponemos un marco experimental basado en una arquitectura de reconocimiento en ‘dos etapas’. En la primera etapa, empleamos sistemas basados en técnicas de recuperación de información y aprendizaje de máquina para identificar las temáticas sobre las cuales se habla en una transcripción de un segmento de audio. Esta transcripción es generada por el sistema de reconocimiento empleando un modelo de lenguaje general. De acuerdo con la relevancia de las temáticas que han sido identificadas, se lleva a cabo la adaptación dinámica del modelo de lenguaje. En la segunda etapa de la arquitectura de reconocimiento, usamos este modelo adaptado para realizar de nuevo el reconocimiento del segmento de audio. Para determinar los beneficios del marco de trabajo propuesto, llevamos a cabo la evaluación de cada uno de los sistemas principales previamente mencionados. Esta evaluación es realizada sobre discursos en el dominio de la política usando la base de datos EPPS (European Parliamentary Plenary Sessions - Sesiones Plenarias del Parlamento Europeo) del proyecto europeo TC-STAR. Analizamos distintas métricas acerca del rendimiento de los sistemas y evaluamos las mejoras propuestas con respecto a los sistemas de referencia. ABSTRACT The last decade has witnessed major advances in speech recognition technology. Today’s commercial systems are able to recognize continuous speech from numerous speakers, with acceptable levels of error and without the need for an explicit adaptation procedure. Despite this progress, speech recognition is far from being a solved problem. Most of these systems are adjusted to a particular domain and their efficacy depends significantly, among many other aspects, on the similarity between the language model used and the task that is being addressed. This dependence is even more important in scenarios where the statistical properties of the language fluctuates throughout the time, for example, in application domains involving spontaneous and multitopic speech. Over the last years there has been an increasing effort in enhancing the speech recognition systems for such domains. This has been done, among other approaches, by means of techniques of automatic adaptation. These techniques are applied to the existing systems, specially since exporting the system to a new task or domain may be both time-consuming and expensive. Adaptation techniques require additional sources of information, and the spoken language could provide some of them. It must be considered that speech not only conveys a message, it also provides information on the context in which the spoken communication takes place (e.g. on the subject on which it is being talked about). Therefore, when we communicate through speech, it could be feasible to identify the elements of the language that characterize the context, and at the same time, to track the changes that occur in those elements over time. This information can be extracted and exploited through techniques of information retrieval and machine learning. This allows us, within the development of more robust speech recognition systems, to enhance the adaptation of language models to the conditions of the context, thus strengthening the recognition system for domains under changing conditions (such as potential variations in vocabulary, style and topic). In this sense, the main contribution of this Thesis is the proposal and evaluation of a framework of topic-motivated contextualization based on the dynamic and non-supervised adaptation of language models for the enhancement of an automatic speech recognition system. This adaptation is based on an combined approach (from the perspective of both information retrieval and machine learning fields) whereby we identify the topics that are being discussed in an audio recording. The topic identification, therefore, enables the system to perform an adaptation of the language model according to the contextual conditions. The proposed framework can be divided in two major systems: a topic identification system and a dynamic language model adaptation system. This Thesis can be outlined from the perspective of the particular contributions made in each of the fields that composes the proposed framework: _ Regarding the topic identification system, we have focused on the enhancement of the document preprocessing techniques in addition to contributing in the definition of more robust criteria for the selection of index-terms. – Within both information retrieval and machine learning based approaches, the efficiency of topic identification systems, depends, to a large extent, on the mechanisms of preprocessing applied to the documents. Among the many operations that encloses the preprocessing procedures, an adequate selection of index-terms is critical to establish conceptual and semantic relationships between terms and documents. This process might also be weakened by a poor choice of stopwords or lack of precision in defining stemming rules. In this regard we compare and evaluate different criteria for preprocessing the documents, as well as for improving the selection of the index-terms. This allows us to not only reduce the size of the indexing structure but also to strengthen the topic identification process. – One of the most crucial aspects, in relation to the performance of topic identification systems, is to assign different weights to different terms depending on their contribution to the content of the document. In this sense we evaluate and propose alternative approaches to traditional weighting schemes (such as tf-idf ) that allow us to improve the specificity of terms, and to better identify the topics that are related to documents. _ Regarding the dynamic language model adaptation, we divide the contextualization process into different steps. – We propose supervised and unsupervised approaches for the generation of topic-based language models. The first of them is intended to generate topic-based language models by grouping the documents, in the training set, according to the original topic labels of the corpus. Nevertheless, a goal of this Thesis is to evaluate whether or not the use of these labels to generate language models is optimal in terms of recognition accuracy. For this reason, we propose a second approach, an unsupervised one, in which the objective is to group the data in the training set into automatic topic clusters based on the semantic similarity between the documents. By means of clustering approaches we expect to obtain a more cohesive association of the documents that are related by similar concepts, thus improving the coverage of the topic-based language models and enhancing the performance of the recognition system. – We develop various strategies in order to create a context-dependent language model. Our aim is that this model reflects the semantic context of the current utterance, i.e. the most relevant topics that are being discussed. This model is generated by means of a linear interpolation between the topic-based language models related to the most relevant topics. The estimation of the interpolation weights is based mainly on the outcome of the topic identification process. – Finally, we propose a methodology for the dynamic adaptation of a background language model. The adaptation process takes into account the context-dependent model as well as the information provided by the topic identification process. The scheme used for the adaptation is a linear interpolation between the background model and the context-dependent one. We also study different approaches to determine the interpolation weights used in this adaptation scheme. Once we defined the basis of our topic-motivated contextualization framework, we propose its application into an automatic speech recognition system. We focus on two aspects: the contextualization of the language models used by the system, and the incorporation of semantic-related information into a topic-based adaptation process. To achieve this, we propose an experimental framework based in ‘a two stages’ recognition architecture. In the first stage of the architecture, Information Retrieval and Machine Learning techniques are used to identify the topics in a transcription of an audio segment. This transcription is generated by the recognition system using a background language model. According to the confidence on the topics that have been identified, the dynamic language model adaptation is carried out. In the second stage of the recognition architecture, an adapted language model is used to re-decode the utterance. To test the benefits of the proposed framework, we carry out the evaluation of each of the major systems aforementioned. The evaluation is conducted on speeches of political domain using the EPPS (European Parliamentary Plenary Sessions) database from the European TC-STAR project. We analyse several performance metrics that allow us to compare the improvements of the proposed systems against the baseline ones.

Bio-QA-GNN: Reasoning with Language Models and Knowledge Graphs for Interpretable Biomedical Question Answering

Nowadays the idea of injecting world or domain-specific structured knowledge into pre-trained language models (PLMs) is becoming an increasingly popular approach for solving problems such as biases, hallucinations, huge architectural sizes, and explainability lack—critical for real-world natural language processing applications in sensitive fields like bioinformatics. One recent work that has garnered much attention in Neuro-symbolic AI is QA-GNN, an end-to-end model for multiple-choice open-domain question answering (MCOQA) tasks via interpretable text-graph reasoning. Unlike previous publications, QA-GNN mutually informs PLMs and graph neural networks (GNNs) on top of relevant facts retrieved from knowledge graphs (KGs). However, taking a more holistic view, existing PLM+KG contributions mainly consider commonsense benchmarks and ignore or shallowly analyze performances on biomedical datasets. This thesis start from a propose of a deep investigation of QA-GNN for biomedicine, comparing existing or brand-new PLMs, KGs, edge-aware GNNs, preprocessing techniques, and initialization strategies. By combining the insights emerged in DISI's research, we introduce Bio-QA-GNN that include a KG. Working with this part has led to an improvement in state-of-the-art of MCOQA model on biomedical/clinical text, largely outperforming the original one (+3.63\% accuracy on MedQA). Our findings also contribute to a better understanding of the explanation degree allowed by joint text-graph reasoning architectures and their effectiveness on different medical subjects and reasoning types. Codes, models, datasets, and demos to reproduce the results are freely available at: \url{https://github.com/disi-unibo-nlp/bio-qagnn}.

Preprocessing and time-frequency analysis of newborn EEG seizures

Neurological disease or dysfunction in newborn infants is often first manifested by seizures. Prolonged seizures can result in impaired neurodevelopment or even death. In adults, the clinical signs of seizures are well defined and easily recognized. In newborns, however, the clinical signs are subtle and may be absent or easily missed without constant close observation. This article describes the use of adaptive signal processing techniques for removing artifacts from newborn electroencephalogram (EEG) signals. Three adaptive algorithms have been designed in the context of EEG signals. This preprocessing is necessary before attempting a fine time-frequency analysis of EEG rhythmical activities, such as electrical seizures, corrupted by high amplitude signals. After an overview of newborn EEG signals, the authors describe the data acquisition set-up. They then introduce the basic physiological concepts related to normal and abnormal newborn EEGs and discuss the three adaptive algorithms for artifact removal. They also present time-frequency representations (TFRs) of seizure signals and discuss the estimation and modeling of the instantaneous frequency related to the main ridge of the TFR.