Conditional Deep Gaussian Processes: Empirical Bayes Hyperdata Learning

It is desirable to combine the expressive power of deep learning with Gaussian Process (GP) in one expressive Bayesian learning model. Deep kernel learning showed success as a deep network used for feature extraction. Then, a GP was used as the function model. Recently, it was suggested that, albeit training with marginal likelihood, the deterministic nature of a feature extractor might lead to overfitting, and replacement with a Bayesian network seemed to cure it. Here, we propose the conditional deep Gaussian process (DGP) in which the intermediate GPs in hierarchical composition are supported by the hyperdata and the exposed GP remains zero mean. Motivated by the inducing points in sparse GP, the hyperdata also play the role of function supports, but are hyperparameters rather than random variables. It follows our previous moment matching approach to approximate the marginal prior for conditional DGP with a GP carrying an effective kernel. Thus, as in empirical Bayes, the hyperdata are learned by optimizing the approximate marginal likelihood which implicitly depends on the hyperdata via the kernel. We show the equivalence with the deep kernel learning in the limit of dense hyperdata in latent space. However, the conditional DGP and the corresponding approximate inference enjoy the benefit of being more Bayesian than deep kernel learning. Preliminary extrapolation results demonstrate expressive power from the depth of hierarchy by exploiting the exact covariance and hyperdata learning, in comparison with GP kernel composition, DGP variational inference and deep kernel learning. We also address the non-Gaussian aspect of our model as well as way of upgrading to a full Bayes inference.

Pistachio Kernel Composition of ‘Kalehghouchi’, ‘Pete 1’, and ‘Lost Hills’ in California

Moisture and fat content, fatty acid profile, and volatile terpenes were measured for the first time for ‘Kalehghouchi’, ‘Pete 1’, and ‘Lost Hills’ pistachios grown at two California microclimates: Lost Hills and Parlier. ‘Kalehghouchi’ had the highest moisture content, followed by ‘Pete 1’ and ‘Lost Hills’, respectively. While the moisture content of ‘Kalehghouchi’ was not significantly affected by location, it was lower for ‘Pete 1’ grown at Parlier (40.8 vs. 40.8 g/100 g) and higher for ‘Lost Hill’ grown there (48.2 vs. 45.2 g/100 g). ‘Pete 1’ grown at the Parlier site had a higher fat content compared with ‘Lost Hills’ (47.7 vs. 43.0 g/100 g). ‘Kalehghouchi’ had a lower fat content at Parlier compared with Lost Hills (42.0 vs. 44.9 g/100 g), and ‘Lost Hills’ was unaffected by location. The main fatty acid measured in the pistachio samples was oleic acid (52% to 58%), followed by linoleic (26% to 33%) and palmitic acids (11% to 13%). While oleic acid content of ‘Lost Hills’ and ‘Kalehghouchi’ was higher for pistachios grown in Parlier, no impact of location was observed for ‘Pete 1’. The fatty acid profiles of all three cultivars appeared to be more dependent on genotype and less affected by microclimate. α-pinene (95–1682 ng/kg), limonene (37–741 ng/kg), and α-terpinolene (1–368 ng/kg) were the most abundant volatiles among all the cultivars and locations. Microclimate was the primary factor in determining volatile terpenes concentration in pistachio kernels.

Wheat Grain Composition, Dough Rheology and Bread Quality as Affected by Nitrogen and Sulfur Fertilization and Seeding Density

Flour from old varieties are usually considered very weak flours, and thus difficult to use in breadmaking especially when processed as Italian “Tipo 2” flour. Hence, the aim of our study was to understand if agronomic treatments can be used to improve flour processability and the quality of three old wheat varieties. An experimental strip-plot scheme was used: three old wheat varieties (Andriolo, Sieve, Verna), two seeding densities, three levels of nitrogen fertilization (N35, N80, and N135), and two levels of foliar sulfur fertilization. Analyzed parameters related to kernel composition, dough rheology and bread quality. Sulfur and nitrogen treatments significantly affected protein composition and dough alveograph strength, which increased by about 34% with nitrogen fertilization, and by about 14% with the sulfur treatment. However, only nitrogen fertilization affected bread characteristics. Crumb density significantly decreased from N35 to N135, while springiness and cohesiveness increased. On the other hand, sulfur did not improve breads. This highlight the importance of performing breadmaking tests in addition to the rheological determinations. The poor technological performance of old wheat flours can be improved with agronomical treatments designed to obtain higher-quality bread.

Morphological Traits, Kernel Composition and Sensory Evaluation of Hazelnut (Corylus avellana L.) Cultivars Grown in Poland

In the present study, the nut and kernel traits of six hazelnut cultivars (‘Barceloński’, ‘Kataloński’, ‘Webba Cenny’, ‘Olbrzym z Halle’, ‘Cosford’, and ‘Nottinghamski’) grown in Poland were investigated. Results showed that significant differences estimated among all six cultivars with all selected morphological traits and compositional properties. During statistical analysis the year of the study was not found to affect most of the investigated traits. The investigated cultivars showed a lower protein content (11.27–13.44%), higher carbohydrate content (16.40–21.79%) and similar fat content (58.91–63.83%) to nuts grown in a warmer climate like Turkey, Italy or Spain. The studied hazelnut varieties were large-sized with diameters greater than 20 mm. The nuts of the ‘Barceloński’, ‘Kataloński’ and ‘Olbrzym z Halle’ cultivars were characterized by the smallest diameters of nut and kernel, were the most spherical (0.85–0.95) and exhibited the largest average nut volume (4.26–4.46). Significant differences were found between the cultivars for oil content and the ratios of major fatty acids. The QDA results, estimated that other than shape, the evaluated nuts significantly differed only in the intensity of nutty flavor (4.88–5.92). The highest intensity of this attribute was found in three cultivars, ‘Nottinghamski’, ‘Cosford’, ‘Olbrzym z Halle’, whereas the lowest estimate was found in the ‘Barceloński’ cultivar, which was also slightly less sweet and more bitter. As a result, the present study showed that the investigated hazelnut kernels can be divided into two groups: ‘Nottinghamski’, ‘Cosford’, and ‘Webba Cenny’ are suitable for table consumption, while ‘Barceloński’, ‘Kataloński’, and ‘Olbrzym z Halle’ are suitable for the processing industry. The current experimental results may help to growers and breeders when choosing the cultivars for new plantations and possibilities of destinations of the produced nuts, for confectionery use or table consumption.

Assimilation of Standard Regularizer Contextual Model and Composite Kernel with Fuzzy-based Noise Classifier

The paper assay the effect of assimilating smoothness prior contextual model and composite kernel function with fuzzy based noise classifier using remote sensing data. The concept of the composite kernel has been taken by fusing two kernels together to improve the classification accuracy. Gaussian and Sigmoid kernel functions have opted for kernel composition. As a contextual model, Markov Random Field (MRF) Standard regularization model (smoothness prior) has been studied with the composite kernel-based Noise Classifier. Comparative analysis of new classifier with the conventional construes increase in overall accuracy.

Tempering-Drying Simulation and Experimental Analysis of Corn Kernel

In order to study the effect of corn kernel composition and physical structure on moisture distribution and transfer process and obtain the optimal tempering-drying parameters of corn kernel, a physical model was constructed with four different components as follows: seed coat, horny endosperm, farinaceous endosperm and embryo. The drying model was established based on the assumption of different diffusion coefficients and same thermal conductivity for the four components. The software of COMSOL Multiphysics was used to simulate the heat and mass transfer process inside the corn kernel during the thin-layer drying. The results showed that the least total drying time and the best drying quality were achieved under the multistage circulating drying of 10 min hot air drying and 60 min tempering, and the tempering degree was up to 0.9207.