SEQUENTIAL DECISIONS AND PREDICTIONS IN NATURAL LANGUAGE PROCESSING

Natural language processing has achieved great success in a wide range of ap- plications, producing both commercial language services and open-source language tools. However, most methods take a static or batch approach, assuming that the model has all information it needs and makes a one-time prediction. In this disser- tation, we study dynamic problems where the input comes in a sequence instead of all at once, and the output must be produced while the input is arriving. In these problems, predictions are often made based only on partial information. We see this dynamic setting in many real-time, interactive applications. These problems usually involve a trade-off between the amount of input received (cost) and the quality of the output prediction (accuracy). Therefore, the evaluation considers both objectives (e.g., plotting a Pareto curve). Our goal is to develop a formal understanding of sequential prediction and decision-making problems in natural language processing and to propose efficient solutions. Toward this end, we present meta-algorithms that take an existent batch model and produce a dynamic model to handle sequential inputs and outputs. Webuild our framework upon theories of Markov Decision Process (MDP), which allows learning to trade off competing objectives in a principled way. The main machine learning techniques we use are from imitation learning and reinforcement learning, and we advance current techniques to tackle problems arising in our settings. We evaluate our algorithm on a variety of applications, including dependency parsing, machine translation, and question answering. We show that our approach achieves a better cost-accuracy trade-off than the batch approach and heuristic-based decision- making approaches. We first propose a general framework for cost-sensitive prediction, where dif- ferent parts of the input come at different costs. We formulate a decision-making process that selects pieces of the input sequentially, and the selection is adaptive to each instance. Our approach is evaluated on both standard classification tasks and a structured prediction task (dependency parsing). We show that it achieves similar prediction quality to methods that use all input, while inducing a much smaller cost. Next, we extend the framework to problems where the input is revealed incremen- tally in a fixed order. We study two applications: simultaneous machine translation and quiz bowl (incremental text classification). We discuss challenges in this set- ting and show that adding domain knowledge eases the decision-making problem. A central theme throughout the chapters is an MDP formulation of a challenging problem with sequential input/output and trade-off decisions, accompanied by a learning algorithm that solves the MDP.

Bcl-xL (S49) and (S62) sequential phosphorylation/dephosphorylation during mitosis prevents chromosome instability and aneuploidy

Une caractéristique intéressante de la protéine Bcl-xL est la présence d'un domaine en boucle non-structurée entre les hélices α1 and α2 de la protéine. Ce domaine protéique n'est pas essentiel pour sa fonction anti-apoptotique et absent chez CED-9, la protéine orthologue chez Caenorhabditis elegans. A l'intérieur de ce domaine, Bcl-xL subit une phosphorylation et déphosphorylation dynamique sur les résidus Ser49 et Ser62 en phase G2 du cycle cellulaire et lors de la mitose. Lorsque ces résidus sont mutés et les protéines exprimées dans des cellules cancéreuses, les cellules démontrent plusieurs défauts mitotiques liés à l'instabilité chromosomique. Pour analyser les effets de Bcl-xL Ser49 et Ser62 dans les cellules normales, les présentes études ont été réalisées dans des cellules diploïdes humaines normales, et in vivo chez Caenorhabditis elegans. Dans une première étude, nous avons utilisé la lignée cellulaire de cellules fibroblastiques diploïdes humaines normales BJ, exprimant Bcl-xL (type sauvage), (S49A), (S49D), (S62A), (S62D) et les double (S49/62A) et (S49/62D) mutants. Les cellules exprimant les mutants de phosphorylation ont montré des cinétiques de doublement de la population cellulaire réduites. Ces effets sur la cinétique de doublement de la population cellulaire corrèle avec l'apparition de la sénescence cellulaire, sans impact sur les taux de mort cellulaire. Ces cellules sénescentes affichent des phénotypes typiques de sénescence associés notamment à haut niveau de l'activité β-galactosidase associée à la sénescence, la sécrétion d' interleukine-6, l'activation de p53 et de p21WAF1/ Cip1, un inhibiteur des complexes kinase cycline-dépendant, ainsi que la formation de foyers de chromatine nucléaire associés à γH2A.X. Les analyses de fluorescence par hybridation in situ et des caryotypes par coloration au Giemsa ont révélé que l'expression des mutants de phosphorylation de Bcl-xL provoquent de l'instabilité chromosomique et l'aneuploïdie. Ces résultats suggèrent que les cycles de phosphorylation et déphosphorylation dynamiques de Bcl-xL Ser49 et Ser62 sont importants dans le maintien de l'intégrité des chromosomes lors de la mitose dans les cellules normales. Dans une deuxième étude, nous avons entrepris des expériences chez Caenorhabditis elegans pour comprendre l'importance des résidus Ser49 et Ser62 de Bcl-xL in vivo. Les vers transgéniques portant les mutations de Bcl-xL (S49A, S62A, S49D, S62D et S49/62A) ont été générés et leurs effets ont été analysés sur les cellules germinales des jeunes vers adultes. Les vers portant les mutations de Bcl-xL ont montré une diminution de ponte et d'éclosion des oeufs, des variations de la longueur de leurs régions mitotiques et des zones de transition, des anomalies chromosomiques à leur stade de diplotène, et une augmentation de l'apoptose des cellules germinales. Certaines de ces souches transgéniques, en particulier les variants Ser/Ala, ont également montré des variations de durée de vie par rapport aux vers témoins. Ces observations in vivo ont confirmé l'importance de Ser49 et Ser62 à l'intérieur du domaine à boucle de Bcl-xL pour le maintien de la stabilité chromosomique. Ces études auront une incidence sur les futures stratégies visant à développer et à identifier des composés qui pourraient cibler non seulement le domaine anti-apoptotique de la protéine Bcl-xL, mais aussi son domaine mitotique pour la thérapie du cancer.

Identifying stationary series in panels: A Monte Carlo evaluation of sequential panel selection methods

Sequential panel selection methods (spsms — procedures that sequentially use conventional panel unit root tests to identify I(0)I(0) time series in panels) are increasingly used in the empirical literature. We check the reliability of spsms by using Monte Carlo simulations based on generating directly the individual asymptotic pp values to be combined into the panel unit root tests, in this way isolating the classification abilities of the procedures from the small sample properties of the underlying univariate unit root tests. The simulations consider both independent and cross-dependent individual test statistics. Results suggest that spsms may offer advantages over time series tests only under special conditions.

Fitting phase type distribution to service process with sequential phases

The work of this thesis is concerned with fitting Hypo-exponential and Erlang phase type distributions for modeling real life processes with non-exponential service time. There exist situations where exponential distributions cannot explain the distribution of service time properly. This thesis presents the application of two traditional statistical estimation techniques to approximate the service distributions of processes with coefficient of variation less than one. It also presents an algorithm to fit Hypo-exponential distribution for complex situations which can’t be handled properly with traditional estimation techniques. The result shows the effect of variation of sample size and other parameters on the efficiency of the estimation techniques by comparing their respective outputs. Furthermore it checks how accurately the proposed algorithm approximates a given distribution.

Multilocus genetic analyses provide insight into speciation and hybridization in aquatic grasses, genus Ruppia

Aquatic plants of the genus Ruppia inhabit some of the most threatened habitats in the world, such as coastal lagoons and inland saline to brackish waters where their meadows play several key roles. The evolutionary history of this genus has been affected by the processes of hybridization, polyploidization, and vicariance, which have resulted in uncertainty regarding the number of species. In the present study, we apply microsatellite markers for the identification, genetic characterization, and detection of hybridization events among populations of putative Ruppia species found in the southern Iberian Peninsula, with the exception of a clearly distinct species, the diploid Ruppia maritima. Microsatellite markers group the populations into genetically distinct entities that are not coincident with geographical location and contain unique diagnostic alleles. These results support the interpretation of these entities as distinct species: designated here as (1) Ruppia drepanensis, (2) Ruppia cf. maritima, and (3) Ruppia cirrhosa. A fourth distinct genetic entity was identified as a putative hybrid between R. cf. maritima and R. cirrhosa because it contained a mixture of microsatellite alleles that are otherwise unique to these putative species. Hence, our analyses were able to discriminate among different genetic entities of Ruppia and, by adding multilocus nuclear markers, we confirm hybridization as an important process of speciation within the genus. In addition, careful taxonomic curation of the samples enabled us to determine the genotypic and genetic diversity and differentiation among populations of each putative Ruppia species. This will be important for identifying diversity hotspots and evaluating patterns of population genetic connectivity. © 2015 The Linnean Society of London, Biological Journal of the Linnean Society, 2015, 00, 000–000.

Using multivariate sequential patterns to improve survival prediction in intensive care burn unit

Resuscitation and stabilization are key issues in Intensive Care Burn Units and early survival predictions help to decide the best clinical action during these phases. Current survival scores of burns focus on clinical variables such as age or the body surface area. However, the evolution of other parameters (e.g. diuresis or fluid balance) during the first days is also valuable knowledge. In this work we suggest a methodology and we propose a Temporal Data Mining algorithm to estimate the survival condition from the patient’s evolution. Experiments conducted on 480 patients show the improvement of survival prediction.

European lampreys: new insights on postglacial colonization, gene flow and speciation.

Ice ages are known to be the most dominant palaeoclimatic feature occurring on Earth, producing severe climatic oscillations and consequently shaping the distribution and the population structure of several species. Lampreys constitute excellent models to study the colonization of freshwater systems, as they commonly appear in pairs of closely related species of anadromous versus freshwater resident adults, thus having the ability to colonize new habitats, through the anadromous species, and establish freshwater resident derivates. We used 10 microsatellite loci to investigate the spatial structure, patterns of gene flow and migration routes of Lampetra populations in Europe. We sampled 11 populations including the migratory L. fluviatilis and four resident species, L. planeri, L. alavariensis, L. auremensis and L. lusitanica, the last three endemic to the Iberian Peninsula. In this southern glacial refugium almost all sampled populations represent a distinct genetic cluster, showing high levels of allopatric differentiation, reflecting long periods of isolation. As result of their more recent common ancestor, populations from northern Europe are less divergent among them, they are represented by fewer genetic clusters, and there is evidence of strong recent gene flow among populations. These previously glaciated areas from northern Europe may have been colonized from lampreys expanding out of the Iberian refugia. The pair L. fluviatilis/L. planeri is apparently at different stages of speciation in different locations, showing evidences of high reproductive isolation in the southern refugium, and low differentiation in the north.

Sequential alcohol oxidation/putative homo Claisen-Tishchenko-type reaction to give esters: a key process in accessing novel biologically active lactone macrocycles

We report an efficient methodology for the direct oxidative esterification of primary alcohols to diether-esters using pyridinium chlorochromate (PCC). Numerous studies were carried out to probe the reaction mechanism and at the same time optimize the reaction conditions. The reaction could be conducted with 1 equivalent of PCC and 1 equivalent of BF3 center dot OEt2. Indications based on literature precedent were that the reaction may proceed via a sequential alcohol oxidation to the aldehyde followed by a putative Cr or boron catalyzed Claisen-Tishchenko-type reaction. Using this efficient methodology, we synthesized a family of novel diether-esters in very good yields; some of these molecules were subsequently tested against both acetylcholinesterase (AChE) and butyrylcholinesterase (BuChE). In addition, we also disclose a new synthetic strategy for the synthesis of lactam macrocycles with potential biological activity. This methodology included the regioselective borylation of the ester substrate and a subsequent Suzuki-Miyaura coupling to obtain the desired lactam macrocycle.

Development and application of a selenium speciation method in cattle feed and beef samples using HPLC-ICP-MS: evaluating the selenium metabolic process in cattle.

2016

Sequential factorial designs for method development of the determination of Cd and Pb in fish and shrimp by GF AAS after sample freezedrying and tetramethylammonium hydroxide solubilization.

2016

Sequential exhaustive extraction of a Mollisol soil, and characterizations of humic components, including humin, by solid and solution state NMR.

A comprehensive sequential extraction procedure was applied to isolate soil organic components using aqueous solvents at different pH values, base plus urea (base-urea), and finally dimethylsulfoxide (DMSO) plus concentrated H2SO4 (DMSO-acid) for the humin-enriched clay separates. The extracts from base-urea and DMSO-acid would be regarded as 'humin' in the classical definitions. The fractions isolated from aqueous base, base-urea and DMSO-acid were characterized by solid and solution state NMR spectroscopy. The base-urea solvent system isolated ca. 10% (by mass) additional humic substances. The combined base-urea and DMSO-acid solvents isolated ca. 93% of total organic carbon from the humin-enriched fine clay fraction (<2 ?m). Characterization of the humic fractions by solid-state NMR spectroscopy showed that oxidized char materials were concentrated in humic acids isolated at pH 7, and in the base-urea extract. Lignin-derived materials were in considerable abundance in the humic acids isolated at pH 12.6. Only very small amounts of char-derived structures were contained in the fulvic acids and fulvic acids-like material isolated from the base-urea solvent. After extraction with base-urea, the 0.5 m NaOH extract from the humin-enriched clay was predominantly composed of aliphatic hydrocarbon groups, and with lesser amounts of aromatic carbon (probably including some char material), and carbohydrates and peptides. From the combination of solid and solution-state NMR spectroscopy, it is clear that the major components of humin materials, from the DMSO-acid solvent, after the exhaustive extraction sequence, were composed of microbial and plant derived components, mainly long-chain aliphatic species (including fatty acids/ester, waxes, lipids and cuticular material), carbohydrate, peptides/proteins, lignin derivatives, lipoprotein and peptidoglycan (major structural components in bacteria cell walls). Black carbon or char materials were enriched in humic acids isolated at pH 7 and humic acids-like material isolated in the base-urea medium, indicating that urea can liberate char-derived material hydrogen bonded or trapped within the humin matrix.

Computational creativity: an interdisciplinary approach to sequential learning and creative generations

Creativity seems mysterious; when we experience a creative spark, it is difficult to explain how we got that idea, and we often recall notions like ``inspiration" and ``intuition" when we try to explain the phenomenon. The fact that we are clueless about how a creative idea manifests itself does not necessarily imply that a scientific explanation cannot exist. We are unaware of how we perform certain tasks, such as biking or language understanding, but we have more and more computational techniques that can replicate and hopefully explain such activities. We should understand that every creative act is a fruit of experience, society, and culture. Nothing comes from nothing. Novel ideas are never utterly new; they stem from representations that are already in mind. Creativity involves establishing new relations between pieces of information we had already: then, the greater the knowledge, the greater the possibility of finding uncommon connections, and the more the potential to be creative. In this vein, a beneficial approach to a better understanding of creativity must include computational or mechanistic accounts of such inner procedures and the formation of the knowledge that enables such connections. That is the aim of Computational Creativity: to develop computational systems for emulating and studying creativity. Hence, this dissertation focuses on these two related research areas: discussing computational mechanisms to generate creative artifacts and describing some implicit cognitive processes that can form the basis for creative thoughts.

Cross-sectional Study Comparing Different Therapeutic Modalities For Cystic Lymphangiomas In Children.

Here, we describe our experience with different therapeutic modalities used to treat cystic lymphangiomas in children in our hospital, including single therapy with OK-432, bleomycin and surgery, and a combination of the three modalities. We performed a retrospective, cross-sectional study including patients treated from 1998 to 2011. The effects on macrocystic lymphangiomas and adverse reactions were evaluated. Twenty-nine children with cystic lymphangiomas without any previous treatment were included. Under general anesthesia, patients given sclerosing agents underwent puncture of the lesion (guided by ultrasound when necessary) and complete aspiration of the intralesional liquid. The patients were evaluated with ultrasound and clinical examinations for a maximum follow-up time of 4 years. The proportions of patients considered cured after the first therapeutic approach were 44% in the surgery group, 29% in the bleomycin group and 31% in the OK-432 group. These proportions were not significantly different. Sequential treatment increased the rates of curative results to 71%, 74% and 44%, respectively, after the final treatment, which in our case was approximately 1.5 applications per patient. The results of this study indicate that most patients with cystic lymphangiomas do not show complete resolution after the initial therapy, regardless of whether the therapy is surgical or involves the use of sclerosing agents. To achieve complete resolution of the lesions, either multiple operations or a combination of surgery and sclerotherapy must be used and should be tailored to the characteristics of each patient.