Meta-learning for informed model searches in Automated Machine Learning

Day by day, machine learning is changing our lives in ways we could not have imagined just 5 years ago. ML expertise is more and more requested and needed, though just a limited number of ML engineers are available on the job market, and their knowledge is always limited by an inherent characteristic of theirs: they are humans. This thesis explores the possibilities offered by meta-learning, a new field in ML that takes learning a level higher: models are trained on other models' training data, starting from features of the dataset they were trained on, inference times, obtained performances, to try to understand the relationship between a good model and the way it was obtained. The so-called metamodel was trained on data collected by OpenML, the largest ML metadata platform that's publicly available today. Datasets were analyzed to obtain meta-features that describe them, which were then tied to model performances in a regression task. The obtained metamodel predicts the expected performances of a given model type (e.g., a random forest) on a given ML task (e.g., classification on the UCI census dataset). This research was then integrated into a custom-made AutoML framework, to show how meta-learning is not an end in itself, but it can be used to further progress our ML research. Encoding ML engineering expertise in a model allows better, faster, and more impactful ML applications across the whole world, while reducing the cost that is inevitably tied to human engineers.

Machine Learning in Automatic Tabulation for Constraint Model Reformulation

Combinatorial decision and optimization problems belong to numerous applications, such as logistics and scheduling, and can be solved with various approaches. Boolean Satisfiability and Constraint Programming solvers are some of the most used ones and their performance is significantly influenced by the model chosen to represent a given problem. This has led to the study of model reformulation methods, one of which is tabulation, that consists in rewriting the expression of a constraint in terms of a table constraint. To apply it, one should identify which constraints can help and which can hinder the solving process. So far this has been performed by hand, for example in MiniZinc, or automatically with manually designed heuristics, in Savile Row. Though, it has been shown that the performances of these heuristics differ across problems and solvers, in some cases helping and in others hindering the solving procedure. However, recent works in the field of combinatorial optimization have shown that Machine Learning (ML) can be increasingly useful in the model reformulation steps. This thesis aims to design a ML approach to identify the instances for which Savile Row’s heuristics should be activated. Additionally, it is possible that the heuristics miss some good tabulation opportunities, so we perform an exploratory analysis for the creation of a ML classifier able to predict whether or not a constraint should be tabulated. The results reached towards the first goal show that a random forest classifier leads to an increase in the performances of 4 different solvers. The experimental results in the second task show that a ML approach could improve the performance of a solver for some problem classes.

[impact Variables On The Decline In Infant Mortality In The State Of São Paulo, Brazil: 1998-2008].

This is an ecological, analytical and retrospective study comprising the 645 municipalities in the State of São Paulo, the scope of which was to determine the relationship between socioeconomic, demographic variables and the model of care in relation to infant mortality rates in the period from 1998 to 2008. The ratio of average annual change for each indicator per stratum coverage was calculated. Infant mortality was analyzed according to the model for repeated measures over time, adjusted for the following correction variables: the city's population, proportion of Family Health Programs (PSFs) deployed, proportion of Growth Acceleration Programs (PACs) deployed, per capita GDP and SPSRI (São Paulo social responsibility index). The analysis was performed by generalized linear models, considering the gamma distribution. Multiple comparisons were performed with the likelihood ratio with chi-square approximate distribution, considering a significance level of 5%. There was a decrease in infant mortality over the years (p < 0.05), with no significant difference from 2004 to 2008 (p > 0.05). The proportion of PSFs deployed (p < 0.0001) and per capita GDP (p < 0.0001) were significant in the model. The decline of infant mortality in this period was influenced by the growth of per capita GDP and PSFs.

Oxidative Stress And Susceptibility To Mitochondrial Permeability Transition Precedes The Onset Of Diabetes In Autoimmune Non-obese Diabetic Mice.

Beta cell destruction in type 1 diabetes (TID) is associated with cellular oxidative stress and mitochondrial pathway of cell death. The aim of this study was to determine whether oxidative stress and mitochondrial dysfunction are present in T1D model (non-obese diabetic mouse, NOD) and if they are related to the stages of disease development. NOD mice were studied at three stages: non-diabetic, pre-diabetic, and diabetic and compared with age-matched Balb/c mice. Mitochondria respiration rates measured at phosphorylating and resting states in liver and soleus biopsies and in isolated liver mitochondria were similar in NOD and Balb/c mice at the three disease stages. However, NOD liver mitochondria were more susceptible to calcium-induced mitochondrial permeability transition as determined by cyclosporine-A-sensitive swelling and by decreased calcium retention capacity in all three stages of diabetes development. Mitochondria H2O2 production rate was higher in non-diabetic, but unaltered in pre-diabetic and diabetic NOD mice. The global cell reactive oxygen species (ROS), but not specific mitochondria ROS production, was significantly increased in NOD lymphomononuclear and stem cells in all disease stages. In addition, marked elevated rates of 2',7'-dichlorodihydrofluorescein (H2DCF) oxidation were observed in pancreatic islets from non-diabetic NOD mice. Using matrix-assisted laser desorption/ionization (MALDI) mass spectrometry (MS) and lipidomic approach, we identified oxidized lipid markers in NOD liver mitochondria for each disease stage, most of them being derivatives of diacylglycerols and phospholipids. These results suggest that the cellular oxidative stress precedes the establishment of diabetes and may be the cause of mitochondrial dysfunction that is involved in beta cell death.

Successful transplant of mesenchymal stem cells in induced osteonecrosis of the ovine femoral head: preliminary results

PURPOSE: Evaluate the bone tissue recovery following transplantation of ovine mesenchymal stem cells (MSC) from bone marrow and human immature dental-pulp stem cells (hIDPSC) in ovine model of induced osteonecrosis of femoral head (ONFH). METHODS: Eight sheep were divided in three experimental groups. First group was composed by four animals with ONFH induced by ethanol through central decompression (CD), for control group without any treatment. The second and third group were compose by two animals, six weeks after ONFH induction received transplantation of heterologous ovine MSC (CD + oMSC), and hIDPSC (CD + hIDPSC), respectively. In both experiments the cells were transplanted without application of any type of immunosupression protocol. RESULTS: Our data indicate that both cell types used in experiments were able to proliferate within injured site providing bone tissue recovery. The histological results obtained from CD+hIDPSC suggested that the bone regeneration in such animals was better than that observed in CD animals. CONCLUSION: Mesenchymal stem cell transplant in induced ovine osteonecrosis of femoral head by central decompression technique is safe, and apparently favors bone regeneration of damaged tissues.

Experimental nodel of C6 brain tumors in athymic rats

Malignant brain tumor experimental models tend to employ cells that are immunologically compatible with the receptor animal. In this study, we have proposed an experimental model of encephalic tumor development by injecting C6 cells into athymic Rowett rats, aiming at reaching a model which more closely resembles to the human glioma tumor. In our model, we observed micro-infiltration of tumor cell clusters in the vicinity of the main tumor mass, and of more distal isolated tumor cells immersed in normal encephalic parenchyma. This degree of infiltration is superior to that usually observed in other C6 models.

Quality of water sources used as drinking water in a Brazilian peri-urban area

The objective of this paper was to assess bacteriological quality of drinking water in a peri-urban area located in the Metropolitan Region of São Paulo, Brazil. A total of 89 water samples were collected from community plastic tanks and 177 water samples from wells were collected bimonthly, from September 2007 to November 2008, for evaluating bacteriological parameters including: Escherichia coli, Enterococcus and heterotrophic plate count (HPC). Clostridium perfringens was investigated in a subsample (40 samples from community plastic tank and 40 from wells). E. coli was present in 5 (5.6%) samples from community plastic tanks (2.0 - 5.1x10(4) MPN/100mL) and in 70 (39.5%) well samples (2.0 - 8.6x10(4) MPN/100mL). Thus, these samples were not in accordance with the Brazilian Regulation. Enterococcus was detected in 20 (22.5%) samples of the community plastic tanks (1 to 79 NC/100mL) and in 142 (80.2%) well samples (1 to >200 NC/100mL). C. perfringens was detected in 5 (12.5%) community plastic tanks samples and in 35 (87.5%) wells samples (2.2 to >16 MPN/100mL). HPC were above 500 CFU/mL in 5 (5.6%) waters from community plastic tanks. In wells samples, the HPC ranged from <1 to 1.6x10(4) CFU/mL. The residual chlorine did not attend the standard established in the drinking water legislation (0.2 mg/L), except in 20 (22.5%) samples. These results confirm the vulnerability of the water supply systems in this peri-urban area what is clearly a public health concern.

Remyelination in experimentally demyelinated connexin 32 KnockOut mice

The aim of this study was to evaluate the role of connexin 32 (Cx 32) during remyelination of the peripheral nervous system, through a local injection of either 0,1% ethidium bromide solution or saline in the sciatic nerve of Cx 32 knockout mice. Euthanasia was performed ranging from 1, 2, 3, 7, 15, 21 to 30 days after injection. Histochemical, immunohistochemical, immunofluorescence and transmission electron microscopical techniques were used to analyze the development of the lesions. Within the sciatic nerves, Schwann cells initially showed signs of intoxication and rejected their sheaths; after seven days, some thin newly formed myelin sheaths with uneven compactness and redundant loops (tomacula) were conspicuous. We concluded that the regeneration of lost myelin sheaths within the PNS followed the pattern already reported for this model in other laboratory species. Therefore, these results suggest that absence of Cx 32 did not interfere with the normal pattern of remyelination in this model in young mice.

Evaluating gene by sex and age interactions on cardiovascular risk factors in Brazilian families

Background: In family studies, it is important to evaluate the impact of genes and environmental factors on traits of interest. In particular, the relative influences of both genes and the environment may vary in different strata of the population of interest, such as young and old individuals, or males and females. Methods: In this paper, extensions of the variance components model are used to evaluate heterogeneity in the genetic and environmental variance components due to the effects of sex and age (the cutoff between young and old was 43 yrs). The data analyzed were from 81 Brazilian families (1,675 individuals) of the Baependi Family Heart Study. Results: The models allowing for heterogeneity of variance components by sex suggest that genetic and environmental variances are not different in males and females for diastolic blood pressure, LDL-cholesterol, and HDL-cholesterol, independent of the covariates included in the models. However, for systolic blood pressure, fasting glucose and triglycerides, the evidence for heterogeneity was dependent on the covariates in the model. For instance, in the presence of sex and age covariates, heterogeneity in the genetic variance component was suggested for fasting glucose. But, for systolic blood pressure, there was no evidence of heterogeneity in any of the two variance components. Except for the LDL-cholesterol, models allowing for heterogeneity by age provide evidence of heterogeneity in genetic variance for triglycerides and systolic and diastolic blood pressure. There was evidence of heterogeneity in environmental variance in fasting glucose and HDL-cholesterol. Conclusions: Our results suggest that heterogeneity in trait variances should not be ignored in the design and analyses of gene-finding studies involving these traits, as it may generate additional information about gene effects, and allow the investigation of more sophisticated models such as the model including sex-specific oligogenic variance components.

Protein aggregation containing beta-amyloid, alpha-synuclein and hyperphosphorylated tau in cultured cells of hippocampus, substantia nigra and locus coeruleus after rotenone exposure

Background: Protein aggregates containing alpha-synuclein, beta-amyloid and hyperphosphorylated tau are commonly found during neurodegenerative processes which is often accompanied by the impairment of mitochondrial complex I respiratory chain and dysfunction of cellular systems of protein degradation. In view of this, we aimed to develop an in vitro model to study protein aggregation associated to neurodegenerative diseases using cultured cells from hippocampus, locus coeruleus and substantia nigra of newborn Lewis rats exposed to 0.5, 1, 10 and 25 nM of rotenone, which is an agricultural pesticide, for 48 hours. Results: We demonstrated that the proportion of cells in culture is approximately the same as found in the brain nuclei they were extracted from. Rotenone at 0.5 nM was able to induce alpha-synuclein and beta amyloid aggregation, as well as increased hyperphosphorylation of tau, although high concentrations of this pesticide (over 1 nM) lead cells to death before protein aggregation. We also demonstrated that the 14kDa isoform of alpha-synuclein is not present in newborn Lewis rats. Conclusion: Rotenone exposure may lead to constitutive protein aggregation in vitro, which may be of relevance to study the mechanisms involved in idiopathic neurodegeneration.

Nonsingular bounce in modified gravity

We investigate bouncing solutions in the framework of the nonsingular gravity model of Brandenberger, Mukhanov and Sornborger. We show that a spatially flat universe filled with ordinary matter undergoing a phase of contraction reaches a stage of minimal expansion factor before bouncing in a regular way to reach the expanding phase. The expansion can be connected to the usual radiation-and matter-dominated epochs before reaching a final expanding de Sitter phase. In general relativity (GR), a bounce can only take place provided that the spatial sections are positively curved, a fact that has been shown to translate into a constraint on the characteristic duration of the bounce. In our model, on the other hand, a bounce can occur also in the absence of spatial curvature, which means that the time scale for the bounce can be made arbitrarily short or long. The implication is that constraints on the bounce characteristic time obtained in GR rely heavily on the assumed theory of gravity. Although the model we investigate is fourth order in the derivatives of the metric (and therefore unstable vis-a-vis the perturbations), this generic bounce dynamics should extend to string-motivated nonsingular models which can accommodate a spatially flat bounce.

A Markovian model market-Akerlof`s lemons and the asymmetry of information

In this work we study an agent based model to investigate the role of asymmetric information degrees for market evolution. This model is quite simple and may be treated analytically since the consumers evaluate the quality of a certain good taking into account only the quality of the last good purchased plus her perceptive capacity beta. As a consequence, the system evolves according to a stationary Markov chain. The value of a good offered by the firms increases along with quality according to an exponent alpha, which is a measure of the technology. It incorporates all the technological capacity of the production systems such as education, scientific development and techniques that change the productivity rates. The technological level plays an important role to explain how the asymmetry of information may affect the market evolution in this model. We observe that, for high technological levels, the market can detect adverse selection. The model allows us to compute the maximum asymmetric information degree before the market collapses. Below this critical point the market evolves during a limited period of time and then dies out completely. When beta is closer to 1 (symmetric information), the market becomes more profitable for high quality goods, although high and low quality markets coexist. The maximum asymmetric information level is a consequence of an ergodicity breakdown in the process of quality evaluation. (C) 2011 Elsevier B.V. All rights reserved.

Estimation of hexenuronic acids and kappa number in kraft pulps of Eucalyptus globulus by Fourier transform near infrared spectroscopy and multivariate analysis

Fourier transform near infrared (FT-NIR) spectroscopy was evaluated as an analytical too[ for monitoring residual Lignin, kappa number and hexenuronic acids (HexA) content in kraft pulps of Eucalyptus globulus. Sets of pulp samples were prepared under different cooking conditions to obtain a wide range of compound concentrations that were characterised by conventional wet chemistry analytical methods. The sample group was also analysed using FT-NIR spectroscopy in order to establish prediction models for the pulp characteristics. Several models were applied to correlate chemical composition in samples with the NIR spectral data by means of PCR or PLS algorithms. Calibration curves were built by using all the spectral data or selected regions. Best calibration models for the quantification of lignin, kappa and HexA were proposed presenting R-2 values of 0.99. Calibration models were used to predict pulp titers of 20 external samples in a validation set. The lignin concentration and kappa number in the range of 1.4-18% and 8-62, respectively, were predicted fairly accurately (standard error of prediction, SEP 1.1% for lignin and 2.9 for kappa). The HexA concentration (range of 5-71 mmol kg(-1) pulp) was more difficult to predict and the SEP was 7.0 mmol kg(-1) pulp in a model of HexA quantified by an ultraviolet (UV) technique and 6.1 mmol kg(-1) pulp in a model of HexA quantified by anion-exchange chromatography (AEC). Even in wet chemical procedures used for HexA determination, there is no good agreement between methods as demonstrated by the UV and AEC methods described in the present work. NIR spectroscopy did provide a rapid estimate of HexA content in kraft pulps prepared in routine cooking experiments.

A numerical model for the description of the nonlinear behaviour of multi-leaf masonry walls

A nonlinear finite element model was developed to simulate the nonlinear response of three-leaf masonry specimens, which were subjected to laboratory tests with the aim of investigating the mechanical behaviour of multiple-leaf stone masonry walls up to failure. The specimens consisted of two external leaves made of stone bricks and mortar joints, and an internal leaf in mortar and stone aggregate. Different loading conditions, typologies of the collar joints, and stone types were taken into account. The constitutive law implemented in the model is characterized by a damage tensor, which allows the damage-induced anisotropy accompanying the cracking process to be described. To follow the post-peak behaviour of the specimens with sufficient accuracy it was necessary to make the damage model non-local, to avoid mesh-dependency effects related to the strain-softening behaviour of the material. Comparisons between the predicted and measured failure loads are quite satisfactory in most of the studied cases. (c) 2007 Elsevier Ltd. All rights reserved.

Constitutive model for long term municipal solid waste mechanical behavior

This paper presents some improvements in the model proposed by Machado et al. [Machado SL, Carvalho MF, Vilar OM. Constitutive model for municipal solid waste. J Geotech Geoenviron Eng ASCE 2002; 128(11):940-51] now considering the influence of biodegradation of organic matter in the mechanical behavior of municipal solid waste. The original framework considers waste as composed of two component groups; fibers and organic paste. The particular laws of behavior are assessed for each component group and then coupled to represent waste behavior. The improvements introduced in this paper take into account the changes in the properties of fibers and mass loss due to organic matter depletion over time. Mass loss is indirectly calculated considering the MSW gas generation potential through a first order decay model. It is shown that as the biodegradation process occurs the proportion of fibers increases, however, they also undergo a degradation process which tends to reduce their ultimate tensile stress and Young modulus. The way these changes influence the behavior of MSW is incorporated in the final framework which captures the main features of the MSW stress-strain behavior under different loading conditions. (C) 2007 Elsevier Ltd. All rights reserved.