Advanced search


Knowledge area




523 results, page 1 of 10

Towards a predictive framework for biocrust mediation of plant performance: a meta-analysis

Caroline A. Havrilla Veer Bala Chaudhary Scott Ferrenberg Anita J. Antoninka Jayne Belnap Matthew Bowker David Eldridge Akasha M. Faist Elisabeth Huber Sannwald Alexander D. Leslie Emilio Rodriguez-Caballero Yuanming Zhang Nichole N. Barger (2019)

"Understanding the importance of biotic interactions in driving the distribution and abundance of species is a central goal of plant ecology. Early vascular plants likely colonized land occupied by biocrusts - photoautotrophic, surface-dwelling soil communities comprised of cyanobacteria, bryophytes, lichens and fungi - suggesting biotic interactions between biocrusts and plants have been at play for some 2,000 million years. Today, biocrusts coexist with plants in dryland ecosystems worldwide, and have been shown to both facilitate or inhibit plant species performance depending on ecological context. Yet, the factors that drive the direction and magnitude of these effects remain largely unknown. We conducted a meta-analysis of plant responses to biocrusts using a global dataset encompassing 1,004 studies from six continents. Meta-analysis revealed there is no simple positive or negative effect of biocrusts on plants. Rather, plant responses differ by biocrust composition and plant species traits and vary across plant ontogeny. Moss-dominated biocrusts facilitated, while lichen-dominated biocrusts inhibited overall plant performance. Plant responses also varied among plant functional groups: C-4 grasses received greater benefits from biocrusts compared to C-3 grasses, and plants without N-fixing symbionts responded more positively to biocrusts than plants with N-fixing symbionts. Biocrusts decreased germination but facilitated growth of non-native plant species. Synthesis. Results suggest that interspecific variation in plant responses to biocrusts, contingent on biocrust type, plant traits, and ontogeny can have strong impacts on plant species performance. These findings have important implications for understanding biocrust contributions to plant productivity and community assembly processes in ecosystems worldwide."

Article

Biological soil crust Biotic interactions Biotic resistance Biotic soil community Germination Facilitation Meta-analysis Plant functional traits Plant-soil (below-ground) interactions CIENCIAS AGROPECUARIAS Y BIOTECNOLOGÍA CIENCIAS AGRARIAS AGRONOMÍA ECOLOGÍA VEGETAL ECOLOGÍA VEGETAL

Meta-analyses of QTL for grain yield and anthesis silking interval in 18 maize populations evaluated under water-stressed and well-watered environments

Kassa Semagn Yoseph Beyene Marilyn Warburton Amsal Tarekegne Prasanna Boddupalli (2013)

Identification of QTL with large phenotypic effects conserved across genetic backgrounds and environments is one of the prerequisites for crop improvement using marker assisted selection (MAS). The objectives of this study were to identify meta-QTL (mQTL) for grain yield (GY) and anthesis silking interval (ASI) across 18 bi-parental maize populations evaluated in the same conditions across 2-4 managed water stressed and 3-4 well watered environments.Results: The meta-analyses identified 68 mQTL (9 QTL specific to ASI, 15 specific to GY, and 44 for both GY and ASI). Mean phenotypic variance explained by each mQTL varied from 1.2 to 13.1% and the overall average was 6.5%. Few QTL were detected under both environmental treatments and/or multiple (>4 populations) genetic backgrounds. The number and 95% genetic and physical confidence intervals of the mQTL were highly reduced compared to the QTL identified in the original studies. Each physical interval of the mQTL consisted of 5 to 926 candidate gene. Conclusions: Meta-analyses reduced the number of QTL by 68% and narrowed the confidence intervals up to 12-fold. At least the 4 mQTL (mQTL2.2, mQTL6.1, mQTL7.5 and mQTL9.2) associated with GY under both water-stressed and well-watered environments and detected up to 6 populations may be considered for fine mapping and validation to confirm effects in different genetic backgrounds and pyramid them into new drought resistant breeding lines. This is the first extensive report on meta-analysis of data from over 3100 individuals genotyped using the same SNP platform and evaluated in the same conditions across a wide range of managed water-stressed and well-watered environments

Article

Breeding Drought Heritability Maize Managed water stress Meta analysis SNP CIENCIAS AGROPECUARIAS Y BIOTECNOLOGÍA

A meta-analysis of effects of chemical composition of incubated diet and bioactive compounds on in vitro ruminal fermentation

ALBERTO MURO REYES (2012)

This study examined the role of supplementation of several bioactive compounds (BC) and the chemical composition of the diet used as substrate for in vitro incubations, on in vitro ruminal fermentation profile and nutrient degradation. A meta-analytical approach was used to weigh the sample size used in each experiment, and account for the random effect of each as well as unequal variance among studies. A total of 20 recently conducted experiments with 354 treatments, each including one control (i.e., no BC supplementation), fulfilled the criteria for inclusion. Doses of BC supplementation varied from 0.03 to 500 mg/g dry matter (DM) of incubated diet. Contents of crude protein (CP) and neutral detergent fibre (NDF) of the incubated diets (DM basis) ranged from 139 to 189 g/kg and 160 to 420 g/kg, respectively. Results indicate that supplementation of BC linearly decreased (137.4 versus 116.5 mmol/L; P<0.05) concentration of total volatile fatty acids (VFA) and proportion of acetate (P<0.05). Also, the concentration of ammonia in the in vitro rumen fluid was lower with BC supplementation (22.9 versus 15.6 mg/dL; P<0.05). Analysis by backward elimination correlation analysis revealed that inclusion of the chemical composition of the incubated diet into the model with BC supplementation improved the accuracy of estimation of responses of fermentation variables. Thus, higher NDF and CP contents of the substrate and higher BC dosage were associated with lower concentrations of total VFA (r2 = 0.54), whereas both lower CP contents of the substrate and BC supplementation lowered the concentration of ammonia (r2 = 0.32). This analysis showed negative associations between BC supplementation and in vitro disappearance of DM and NDF, and positive correlations with dietary NDF content. In contrast, higher BC inclusion and lowering NDF content in the diet was accompanied with decreased in vitro CH4 formation (r2 = 0.21). Results indicate that BC supplementation and chemical composition of the incubated diet are determining factors which impact responses of in vitro ruminal fermentation and degradation.

This study examined the role of supplementation of several bioactive compounds (BC) and the chemical composition of the diet used as substrate for in vitro incubations, on in vitro ruminal fermentation profile and nutrient degradation. A meta-analytical approach was used to weigh the sample size used in each experiment, and account for the random effect of each as well as unequal variance among studies. A total of 20 recently conducted experiments with 354 treatments, each including one control (i.e., no BC supplementation), fulfilled the criteria for inclusion. Doses of BC supplementation varied from 0.03 to 500 mg/g dry matter (DM) of incubated diet. Contents of crude protein (CP) and neutral detergent fibre (NDF) of the incubated diets (DM basis) ranged from 139 to 189 g/kg and 160 to 420 g/kg, respectively. Results indicate that supplementation of BC linearly decreased (137.4 versus 116.5 mmol/L; P<0.05) concentration of total volatile fatty acids (VFA) and proportion of acetate (P<0.05). Also, the concentration of ammonia in the in vitro rumen fluid was lower with BC supplementation (22.9 versus 15.6 mg/dL; P<0.05). Analysis by backward elimination correlation analysis revealed that inclusion of the chemical composition of the incubated diet into the model with BC supplementation improved the accuracy of estimation of responses of fermentation variables. Thus, higher NDF and CP contents of the substrate and higher BC dosage were associated with lower concentrations of total VFA (r2 = 0.54), whereas both lower CP contents of the substrate and BC supplementation lowered the concentration of ammonia (r2 = 0.32). This analysis showed negative associations between BC supplementation and in vitro disappearance of DM and NDF, and positive correlations with dietary NDF content. In contrast, higher BC inclusion and lowering NDF content in the diet was accompanied with decreased in vitro CH4 formation (r2 = 0.21). Results indicate that BC supplementation and chemical composition of the incubated diet are determining factors which impact responses of in vitro ruminal fermentation and degradation.

Producción Científica de la Universidad Autónoma de Zacatecas UAZ

Article

CIENCIAS AGROPECUARIAS Y BIOTECNOLOGÍA Meta-analysis Essential oils In vitro rumen fermentation Volatile fatty acids

Macrophage migration inhibitory factor -173 G/C polymorphism: A global meta-analysis across the disease spectrum

Oscar Illescas Juan Carlos Gomez Verjan LIZBETH ESMERALDA GARCIA VELAZQUEZ Tzipe Govezensky MIRIAM RODRIGUEZ SOSA (2018)

Human macrophage migration inhibitory factor (MIF) is a cytokine that plays a role in several metabolic and inflammatory processes. Single nucleotide polymorphism (SNP) -173 G/C (rs755622) on MIF gene has been associated with numerous diseases, such as arthritis and cancer. However, most of the reports concerning the association of MIF with these and other pathologies are inconsistent and remain quite controversial. Therefore, we performed a meta-analysis from 96 case-control studies on -173 G/C MIF SNP and stratified the data according to the subjects geographic localization or the disease pathophysiology, in order to determine a more meaningful significance to this SNP. The polymorphism was strongly associated with an increased risk in autoimmune-inflammatory, infectious and age-related diseases on the dominant (OR: 0.74 [0.58–0.93], P < 0.01; OR: 0.81 [0.74–0.89], P < 0.0001; and OR: 0.81 [0.76–0.87], P < 0.0001, respectively) and the recessive models (OR: 0.74 [0.57–0.095], P < 0.01; OR: 0.66 [0.48–0.92], P < 0.0154; and OR: 0.70 [0.60–0.82], P < 0.0001, respectively). Also, significant association was found in the geographic localization setting for Asia, Europe and Latin America subdivisions in the dominant (OR: 0.76 [0.69–0.84], P < 0.0001; OR: 0.77 [0.72–0.83], P < 0.0001; OR: 0.61 [0.44–0.83], P-value: 0.0017, respectively) and overdominant models (OR: 0.85 [0.77–0.94], P < 0.0001; OR: 0.80 [0.75–0.86], P < 0.0001; OR: 0.73 [0.63–0.85], P-value: 0.0017, respectively). Afterwards, we implemented a network meta-analysis to compare the association of the polymorphism for two different subdivisions. We found a stronger association for autoimmune than for age-related or autoimmune-inflammatory diseases, and stronger association for infectious than for autoimmune-inflammatory diseases. We report for the first time a meta-analysis of rs755622 polymorphism with a variety of stratified diseases and populations. The study reveals a strong association of the polymorphism with autoimmune and infectious diseases. These results may help direct future research on MIF-173 G/C in diseases in which the relation is clearer and thus assist the search for more plausible applications.

Article

MEDICINA Y CIENCIAS DE LA SALUD BIOLOGÍA Y QUÍMICA Metaanálisis Metanálisis Macrófagos Inflamación Envejecimiento Poliformismo genético Enfermedades autoinmunes Meta-Analysis Macrophages Inflammation Aging Polymorphism, Genetic Autoimmune Diseases

Macrophage migration inhibitory factor -173 G/C polymorphism: A global meta-analysis across the disease spectrum

Oscar Illescas Juan Carlos Gomez Verjan LIZBETH ESMERALDA GARCIA VELAZQUEZ Tzipe Govezensky MIRIAM RODRIGUEZ SOSA (2018)

Human macrophage migration inhibitory factor (MIF) is a cytokine that plays a role in several metabolic and inflammatory processes. Single nucleotide polymorphism (SNP) -173 G/C (rs755622) on MIF gene has been associated with numerous diseases, such as arthritis and cancer. However, most of the reports concerning the association of MIF with these and other pathologies are inconsistent and remain quite controversial. Therefore, we performed a meta-analysis from 96 case-control studies on -173 G/C MIF SNP and stratified the data according to the subjects geographic localization or the disease pathophysiology, in order to determine a more meaningful significance to this SNP. The polymorphism was strongly associated with an increased risk in autoimmune-inflammatory, infectious and age-related diseases on the dominant (OR: 0.74 [0.58–0.93], P < 0.01; OR: 0.81 [0.74–0.89], P < 0.0001; and OR: 0.81 [0.76–0.87], P < 0.0001, respectively) and the recessive models (OR: 0.74 [0.57–0.095], P < 0.01; OR: 0.66 [0.48–0.92], P < 0.0154; and OR: 0.70 [0.60–0.82], P < 0.0001, respectively). Also, significant association was found in the geographic localization setting for Asia, Europe and Latin America subdivisions in the dominant (OR: 0.76 [0.69–0.84], P < 0.0001; OR: 0.77 [0.72–0.83], P < 0.0001; OR: 0.61 [0.44–0.83], P-value: 0.0017, respectively) and overdominant models (OR: 0.85 [0.77–0.94], P < 0.0001; OR: 0.80 [0.75–0.86], P < 0.0001; OR: 0.73 [0.63–0.85], P-value: 0.0017, respectively). Afterwards, we implemented a network meta-analysis to compare the association of the polymorphism for two different subdivisions. We found a stronger association for autoimmune than for age-related or autoimmune-inflammatory diseases, and stronger association for infectious than for autoimmune-inflammatory diseases. We report for the first time a meta-analysis of rs755622 polymorphism with a variety of stratified diseases and populations. The study reveals a strong association of the polymorphism with autoimmune and infectious diseases. These results may help direct future research on MIF-173 G/C in diseases in which the relation is clearer and thus assist the search for more plausible applications.

Article

MEDICINA Y CIENCIAS DE LA SALUD BIOLOGÍA Y QUÍMICA Metaanálisis Metanálisis Macrófagos Inflamación Envejecimiento Poliformismo genético Enfermedades autoinmunes Meta-Analysis Macrophages Inflammation Aging Polymorphism, Genetic Autoimmune Diseases

Meta-análisis de dietas con baja proteína adicionadas con aminoácidos sintéticos para cerdos en engorda

MONICA GONZALEZ REYES (2013)

Tesis (Doctorado en Ciencias, especialista en Ganadería).- Colegio de Postgraduados, 2013.

Se han reportado resultados inconsistentes utilizando dietas bajas en proteína cruda formuladas con sorgo-pasta de soya adicionadas con AA sintéticos que mantengan los resultados productivos y las características de la canal de cerdos alimentados con dieta estándar. Esto ha motivado a realizar un meta-análisis para combinar los resultados originales de estudios independientes. El conjunto de datos utilizados se basó en una línea de investigación sobre la alimentación de cerdos con dietas bajas en proteína en las etapas de iniciación, crecimiento y finalización. Se realizó un análisis de regresión lineal usando un modelo de superficie de respuesta con el comando RSREG de SAS (2009) para obtener el óptimo nivel de proteína para las variables productivas, características de la canal y concentración de urea en plasma. Una vez que se analizaron los datos de los experimentos con el meta-análisis, se prosiguió a montar un experimento para corroborar los resultados de las regresiones con 20 cerdos machos castrados y 20 hembras (Yorkshire×Duroc×Pietrain) en iniciación (11.5 Kg), crecimiento (25.36 Kg) y finalización (54.01 Kg), en respuesta a diferentes niveles de PC (16.5, 17.3, 18.10 y 19.25% en iniciación; n=10; 18.99, 15.11, 13.66 y 12.55% en crecimiento; n=10; y 12.8, 11.3 y 9.5% en finalización; n=12). El diseño experimental fue completamente al azar; cada cerdo se consideró una unidad experimental. El meta-análisis indicó que en iniciación disminuyó ganancia diaria de peso y ganancia de carne magra al bajar el nivel de proteína de 20.9 a 14.5% (P≤0.05); en crecimiento, la ganancia diaria de peso, conversión alimenticia, peso final, porcentaje de carne magra, área del músculo longissimus y urea en plasma fueron menores y la grasa dorsal mayor al bajar la PC (P≤0.05); para finalización, la ganancia diaria de peso, consumo de alimento, conversión alimenticia, peso final, ganancia de carne magra, porcentaje de carne magra, área del músculo longissimus y concentración de urea en plasma disminuyeron al bajar la PC (P≤0.05. En el experimento realizado para corroborar los niveles óptimos de PC, para cerdos en iniciación no hubo diferencias en las variables analizadas, excepto que el porcentaje de carne magra fue mejor (P≤0.05) cuando se utilizaron 16.5 y 19.25% PC. En crecimiento, ganancia diaria de peso, peso final y ganancia de carne magra disminuyeron (P≤0.05) al bajar la PC de la dieta de 18.99 a 12.55% de PC. En finalización, ganancia de peso, consumo de alimento, peso final y ganancia de carne magra se redujeron (P≤0.05) al bajar la PC de 12.8 a 9.5%. La concentración de urea en plasma se redujo (P≤0.05) al disminuir la PC de la dieta en las tres etapas. _______________ META-ANALYSIS OF LOW–PROTEIN DIETS SUPPLEMENTED WITH CRYSTALLINE AMINO ACIDS FOR FATTENING PIGS. ABSTRACT: Inconsistent results have been reported on the effects of low-protein diets formulated with sorghum-soybean meal supplemented with synthetic AA to maintain the productive results of the standard diet for growth performance and carcass characteristics of fattening pigs. This has led to conduct a meta-analysis to combine the results of independent original studies. The data set used was based on a research line of feeding pigs with low-protein diets for nursery, growing and finishing phases of growth. A linear regression analysis was performed using a response surface model with the RSREG command of SAS (2009) to determine the optimum protein level for growth performance, carcass characteristics and plasma urea concentration. Once the meta-analysis was performed, an additional experiment was conducted to corroborate those values using 20 barrows and 20 gilts in nursery (11.5 kg), growing (25.36 kg) and finishing (54.01 kg) phases, in response to different levels of PC (19.25, 18, 10, 17.30 and 16.5% for nursery; n=10; 18.99, 15.11, 13.66 and 12.55% for growing; n=10; 12.8, 11.3 and 9.5% for finishing stage; n=12). The experimental design was a completely randomized; each pig was considered an experimental unit. The meta-analysis of nursery phase indicated that average daily gain and lean meat percentage decreased when lowering the protein level from 20.9 to 14.5% (P≤ 0.05); in growing pigs, average daily gain, feed:gain ratio, body weight, lean meat percentage, longissimus muscle area and plasma urea nitrogen concentration decreased (P≤0.05) while backfat thickness increased (P≤0.05) when CP was reduced; for finishing phase, average daily gain, average daily feed intake, feed:gain ratio, body weight, fat free lean gain, lean meat percentage, longissimus muscle area and plasma urea concentration decreased by lowering dietary CP (P≤0.05). In the experiment conducted to corroborate the optimal levels of PC, for nursery pigs there were no differences in the variables analyzed, except for lean meat percentage, which was higher (P≤0.05) when fed 16.5 and 19.25% CP. In growing pigs there was a reduction on average daily gain, body weight and fat free lean gain (P≤0.05) when fed the lowest protein diet. For finishing pigs, average daily gain, average daily feed intake, body weight and fat free lean gain diminished (P≤0.05) when lowering dietary CP from 12.8 to 9.5%. The plasma urea nitrogen concentration was reduced (P≤0.05) with decreasing dietary PC is the three stages of growth.

Doctoral thesis

Cerdos Dietas bajas en proteína Meta-análisis Pigs Low-protein diets Meta-analysis Ganadería Doctorado CIENCIAS AGROPECUARIAS Y BIOTECNOLOGÍA

Evaluación de las interacciones entre las nanopartículas de plata y microorganismos patógenos

Evaluation of the interactions of silver nanoparticles and pathogenic microorganisms

ROBERTO VAZQUEZ MUÑOZ (2017)

Los nanomateriales antimicrobianos –nanoantibióticos- tales como las nanopartículas de plata –AgNPs-, podrían ayudar a mitigar el reto de las enfermedades infecciosas, que son una de las principales causas de muerte en el mundo. Las AgNPs presentan actividad antiviral y antimicrobiana, aunque pueden afectar a todos los sistemas biológicos (SB). A pesar de su uso, se desconoce la toxicidad relativa de las AgNPs en los diferentes SB, así como qué parámetros fisicoquímicos–medio de cultivo, pH, entre otros- afectan su estabilidad o actividad. Por otro lado, aunque se ha demostrado que pueden mejorar la actividad de algunos antibióticos, hasta el momento no se conoce el mecanismo que favorece la sinergia. El objetivo del presente trabajo fue establecer la toxicidad relativa de las AgNPs en sistemas biológicos de distinto grado de complejidad, desde virus hasta células humanas. Además, se buscó esclarecer qué factores influyen en la estabilidad y actividad de las AgNPs, así como elucidar el mecanismo que favorece la sinergia en los tratamientos combinados AgNPs- antibióticos. Se encontró que las Concentraciones Inhibitorias (CI) de las AgNPs fueron: 10-12 μg ml de plata en bacterias (E. coli, S. Typhimurium, S. aureus y B. subtilis), 20

para Fusarium oxysporum y 45 para Candida albicans, independientemente de su perfil de susceptibilidad a antibióticos. La mayoría de las CI de las AgNPs en todos los sistemas biológicos in vitro –desde virus hasta líneas celulares de cáncer- estuvieron en el orden de 10-1 de plata. Por otro lado, dependiendo del medio de cultivo usado, las AgNPs mostraron diferencias en sus características –plasmón superficial, tamaño, estabilidad- y su actividad antimicrobiana –variaciones en las CI- en hongos y bacterias. Las AgNPs alteran la estructura celular de bacterias y hongos; asimismo, se bioacumulan y se pueden biosintetizar en el medio intracelular. Además, las AgNPs despolarizan la membrana celular de manera significativa. Algunas combinaciones de AgNPs con antibióticos muestran una actividad potenciada; particularmente con aquellos que se internalizan para ejercer su efecto antimicrobiano. Considerando lo anterior, proponemos que el efecto sinérgico se debe al mecanismo de acción de los antibióticos, facilitado por la actividad de las AgNPs sobre la membrana de la célula bacteriana. Hasta donde sabemos, éste es el primer estudio experimental –y sustentado por un meta-análisis- que evalúa el efecto tóxico del mismo nanomaterial

Antimicrobial nanomaterials –nanoantibiotics- such as silver nanoparticles –AgNPs- could help to mitigate the challenge of infectious diseases, which are one of the leading causes of death in the world. AgNPs have antiviral and antimicrobial activity, although they can affect all biological systems (BSs). In spite of their use, the relative toxicity of AgNPs in different BSs is unknown, as well as what physicochemical parameters–culture medium, pH, among others- affect their stability or activity. On the other hand, although it has been shown that they can improve the activity of some antibiotics, so far the mechanism that favors synergy is unknown. The objective of the present work was to establish the relative toxicity of AgNPs in biological systems of varying degrees of complexity, from viruses to human cells. In addition, it was sought to clarify which factors influence the stability and activity of AgNPs, as well as to elucidate the mechanism that favors synergy in the combined AgNPs-antibiotic treatments. It was found that the Inhibitory Concentrations (ICs) of the AgNPs were: 10-12 μg ml of silver in bacteria (E. coli, S. Typhimurium, S. aureus and B. subtilis), 20 for Fusarium oxysporum and 45 for Candida albicans, regardless of their susceptibility profile to antibiotics. The majority of ICs of AgNPs in all in vitro biological systems –from viruses to cancer cell lines- were on the order of 10-1of silver. On the other hand, depending on the culture medium used, the AgNPs showed differences in their characteristics –surface plasmon, size, stability - and their antimicrobial activity- variations in ICs in fungi and bacteria. AgNPs alter the cellular structure of bacteria and fungi; they also bioaccumulate and can be biosynthesized in the intracellular medium. In addition, AgNPs depolarize the cell membrane significantly. Some combinations of AgNPs with antibiotics show enhanced activity; particularly with those that are internalized to exert their antimicrobial effect. Considering the above, we propose that the synergistic effect is due to the mechanism of action of antibiotics, facilitated by the activity of AgNPs on the membrane of the bacterial cell. As far as we know, this is the first experimental study –and supported by a meta-analysis- that evaluates the toxic effect of the same nanomaterial on such a wide range of biological systems. The toxicity of AgNPs is independentof the physiological or structural complexity of the biological sys

Doctoral thesis

Nanoantibióticos, nanopartículas de plata, efecto sinérgico, meta-análisis Nanoantibiotics, silver nanoparticles, synergistic effect, meta-analysis BIOLOGÍA Y QUÍMICA CIENCIAS DE LA VIDA OTRAS ESPECIALIDADES DE LA BIOLOGÍA OTRAS

A full meta-learning approach to assist in the full model selection problem in high volume datasets

Angel Diaz Pacheco (2019)

Data by itself is not information, it becomes information only when it is analyzed and, a big part of such analysis is performed through machine learning techniques. Choosing the right technique for a dataset is not a trivial task because it requires to test the performance of the available alternatives and, taking into account that many of these techniques possess a set of configurable parameters, the process becomes harder. With the advent of social networks, the information in ambits outside the scientific research grew to unprecedented scales, this situation favored the storing of bigger quantities of data potentially rich in information and economical value, but the challenge of selecting the most adequate technique for those datasets got worsened. The full model selection analysis emerged as a way to address this issue finding the best combination of a learning algorithm, a subset of features and a combination of data-preparation techniques to a dataset. Full model selection is capable to obtain models with high predictive accuracy and key information about datasets where there is no prior knowledge but, is not the first alternative when datasets become bigger. This analysis supposes to perform a series of transformations in the dataset and the construction of a classifier when a single model is evaluated but, bearing in mind the combination of all factors involved in the problem, the number of posible models is huge or even infinite. Consequently, this problem cannot be addressed through simpler methods as a grid search. Furthermore, the time of this process grows proportionally to the size of the datasets under analysis, therefore, with bigger datasets, the problem becomes intractable. Several approaches as the use of: proxy models, meta-learning techniques and tools from the Big Data paradigm were explored to be capable to address the huge search space of the full model selection problem and to enable the analysis of highvolume datasets within an affordable computing time. The obtained results of this work showed an important reduction of the time in the search process in comparison with a robust algorithm for model selection and with models of higher predictive accuracy. The contributions of this work were: a framework to perform the full model selection analysis in datasets of any size, based in the MapReduce programming model.

Los datos en sí mismos no son información, se convierten en información solo cuando son analizados y gran parte de tal análisis es realizado a través de técnicas de aprendizaje automático. Elegir la técnica adecuada para un conjunto de datos no es una labor trivial ya que requiere de probar el desempeño de cada alternativa disponible y tomando en cuenta que muchas de estas poseen un conjunto de parámetros configurables, el proceso se hace más complicado. Con el advenimiento de las redes sociales, la información en ámbitos ajenos a la investigación científica creció a escalas sin precedentes, esta situación favoreció el almacenamiento de grandes cantidades de datos potencialmente ricos en información y valor económico, pero, el desafío de seleccionar la técnica más adecuada para un conjunto de datos se hizo más difícil. El análisis de selección de modelo completo emergió como una forma de afrontar el problema de encontrar la mejor combinación de un algoritmo de aprendizaje, un

subconjunto de características y una combinación de técnicas de preprocesamiento para un conjunto de datos. La selección de modelo completo es capaz de obtener modelos de gran poder predictivo e información de interés en conjuntos de datos que no han sido analizados, pero, no es la primera alternativa cuando los conjuntos de datos se hacen más grandes. Este análisis supone realizar una serie de transformaciones en el conjunto de datos y la construcción de un clasificador cuando solo un modelo es evaluado, pero, teniendo en mente la combinación de todos los factores involucrados en el problema, el número de modelos posibles es enorme e incluso infinito. En consecuencia, este problema no puede ser enfrentado a través de métodos más simples como la búsqueda en rejilla. Además, el tiempo de dicho proceso crece en proporción al tamaño del conjunto de datos bajo análisis, por lo tanto, con conjuntos de datos más grandes el problema se hace intratable. Varios enfoques como el uso de: modelos proxy, técnicas de meta aprendizaje y herramientas provenientes del paradigma de Big Data fueron exploradas para tener la capacidad de enfrentar el enorme espacio de búsqueda del problema de selección de modelo completo y habilitar el análisis de conjuntos con gran volumen de datos dentro de un tiempo de computo razonable.

Doctoral thesis

Model selection MapReduce Meta-learning Machine learning CIENCIAS FÍSICO MATEMÁTICAS Y CIENCIAS DE LA TIERRA MATEMÁTICAS CIENCIA DE LOS ORDENADORES LENGUAJES DE PROGRAMACIÓN LENGUAJES DE PROGRAMACIÓN

Aprendizaje supervisado por la web basado en información multi-modal de imágenes

Ricardo Benitez-Jimenez (2019)

Nowadays, in domestic assistance robotics it is of great interest to find objects as part of a sub-task of a particular activity, such as carrying an object from one place to another. However, sometimes the object to be recognized is not part of the existing models for object recognition. An alternative it is learning to recognize a new object category is by retrieving images of the object via the Internet, this technique is known as Webly-supervised learning (WSL). However, use all the retrieved images to train a classifier produces a low performance in this task due to the amount of irrelevant images retrieved from the Web. Recent approaches use irrelevant images in their initial classifiers in order to filter irrelevant images, furthermore, they do not show evidence that their approaches work to learn unknown object categories. In this thesis we present a new method inspired by meta-learning allowing to take advantage of the information (textual and visual) in the Web to image filtering of unknown object categories in a multi-modal approach. The experimental results show that the proposed method is an alternative that offers a similar performance to the manual selection of relevant images in object recognition task.

Actualmente en robótica de asistencia doméstica es de gran interés encontrar objetos como parte de una subtarea de una actividad en particular, tal es el caso de mover un objeto de un lugar a otro. Sin embargo, en ocasiones el objeto a reconocer no forma parte de los modelos existentes para el reconocimiento de objetos. Una manera de aprender a reconocer un nuevo objeto es recuperando imágenes del objeto por medio de Internet, a esta técnica se le conoce como aprendizaje supervisado por la Web (WSL por sus siglas en inglés). No obstante, utilizar todas las imágenes recuperadas para el entrenamiento de un clasificador produce un bajo rendimiento en esta tarea debido a la cantidad de imágenes irrelevantes recuperadas de la Web. Los enfoques existentes conservan imágenes irrelevantes en sus clasificadores iniciales para el filtrado de imágenes irrelevantes, además de no presentar evidencia de que sus enfoques funcionen adecuadamente en objetos desconocidos. En este trabajo se propone un nuevo método inspirado en el meta-aprendizaje que permite tomar ventaja de la información (textual y visual) presente en la Web para filtrar de manera multi-modal imágenes de categorías desconocidas. Los resultados experimentales muestran que el método propuesto es una alternativa que ofrece un rendimiento similar a la selección manual de imágenes relevantes en la tarea de reconocimientos de objetos.

Master thesis

Meta-learning Multi-modal Web supervised learning CIENCIAS FÍSICO MATEMÁTICAS Y CIENCIAS DE LA TIERRA MATEMÁTICAS CIENCIA DE LOS ORDENADORES DISPOSITIVOS DE ALMACENAMIENTO DISPOSITIVOS DE ALMACENAMIENTO

Meta-aprendizaje auto-organizado temporal para predicción a largo plazo de series de tiempo caóticas

RIGOBERTO SALOMON FONSECA DELGADO (2017)

Long-term prediction of time series from chaotic systems is a difficult task but required in various fields such as economics, medicine, government, etc. In recent years, several studies have turned their attention to reuse prior knowledge with the aim of combining models and improving prediction. The process of learning from previous results is called meta-learning. On the context of model combination, the meta-learning strategy lets to automatically generate an expert, able to recommend an appropriate combination of models to predict a time series in particular. However, the process of meta-learning in time series imposes nontrivial challenges such as: analyzing model behavior with time series data, look for models that can be combined with each series and even generate new methods of meta-learning those consider variations of model performance over time. This research focuses on the problem of meta-learning of model combination, using self-organization and an automatic analysis of the behavior of the models. The goal is to improve the accuracy on long-term prediction achieved by state of the art algorithms in chaotic time series. The main contribution of this work is a new method, based on meta-features and self-organized maps neural networks, capable of generating combinations of prediction models considering the accuracy of the models and their diversity of behavior over time. The developed method was compared with other methods of the state of the art, and validated using synthetic series and series obtained from real applications, that have a chaotic and non-chaotic behavior.

La predicción a largo plazo de series de tiempo de sistemas caóticos es una tarea difícil pero requerida en diversos dominios como economía, medicina, gobierno, etc. En los últimos años, varias investigaciones han puesto su atención en reutilizar conocimiento previo a fin de combinar modelos y mejorar la predicción. El proceso de aprender a partir de resultados previos es denominado meta-aprendizaje. En el contexto de combinación de modelos, la estrategia de meta-aprendizaje permite generar automáticamente un experto, capaz de recomendar una combinación de modelos apropiada para predecir una serie de tiempo en particular. Sin embargo, el proceso de meta-aprendizaje en series de tiempo impone retos no triviales tales como: analizar el comportamiento de los modelos con los datos, buscar modelos que puedan combinarse adecuadamente para cada serie, e incluso generar nuevos métodos de meta-aprendizaje, que consideren las variaciones de desempeño de los modelos en el tiempo. Esta investigación se centra en el problema de meta-aprendizaje de combinación de modelos, utilizando auto-organización y un análisis automático del comportamiento de los modelos. La meta es mejorar la exactitud en predicción a largo plazo alcanzada por los algoritmos del estado del arte en series de tiempo caóticas. La principal contribución de este trabajo es un nuevo método, basado en meta-características y en las redes neuronales de mapas auto-organizados, capaz de generar combinaciones de modelos de predicción considerando la exactitud de los modelos y su diversidad de comportamiento en el tiempo. El método desarrollado fue comparado con otros métodos del estado del arte, y validado utilizando series sintéticas y series obtenidas de aplicaciones reales, que presentan comportamientos tanto caóticos como no caóticos.

Doctoral thesis

Time series Meta-learning Forecasting Self-organijing maps Neural networks CIENCIAS FÍSICO MATEMÁTICAS Y CIENCIAS DE LA TIERRA MATEMÁTICAS CIENCIA DE LOS ORDENADORES