Principles and typical computational limitations of sparse speaker separation based on deterministic speech features

The separation of mixed auditory signals into their sources is an eminent neuroscience and engineering challenge. We reveal the principles underlying a deterministic, neural network-like solution to this problem. This approach is orthogonal to ICA/PCA that views the signal constituents as independent realizations of random processes. We demonstrate exemplarily that in the absence of salient frequency modulations, the decomposition of speech signals into local cosine packets allows for a sparse, noise-robust speaker separation. As the main result, we present analytical limitations inherent in the approach, where we propose strategies of how to deal with this situation. Our results offer new perspectives toward efficient noise cleaning and auditory signal separation and provide a new perspective of how the brain might achieve these tasks.     

Colors of the past: color image segmentation in historical topographic maps based on homogeneity

A novel approach to color image segmentation (CIS) in scanned archival topographic maps of the 19th century is presented. Archival maps provide unique information for GIS-based change detection and are the only spatially contiguous data sources prior to the establishment of remote sensing. Processing such documents is challenging due to their very low graphical quality caused by ageing, manual production and scanning. Typical artifacts are high degrees of mixed and false coloring, as well as blurring in the images. Existing approaches for segmentation in cartographic documents are normally presented using well-conditioned maps. The CIS approach presented here uses information from the local image plane, the frequency domain and color space. As a first step, iterative clustering is based on local homogeneity, frequency of homogeneity-tested pixels and similarity. By defining a peak-finding rule, “hidden” color layer prototypes can be identified without prior knowledge. Based on these prototypes a constrained seeded region growing (SRG) process is carried out to find connected regions of color layers using color similarity and spatial connectivity. The method was tested on map pages with different graphical properties with robust results as derived from an accuracy assessment.     

DGAT1 polymorphism in Bos indicus and Bos taurus cattle breeds

As a result of multiple QTL-mapping projects in recent years, a quantitative trait locus for milk fat percentage and milk yield has been described on BTA14. Recent reports name the acyl-CoA : diacylglycerol acyltransferase (DGAT1) gene on BTA14 as a potential candidate gene, with a nonconservative substitution of lysine by alanine (K232A) producing a major effect on milk composition and yield. DGAT1K appears to be the ancestral allele and the K232A substitution probably occurred after the divergence of the Bos indicus and Bos taurus lineages. These findings prompted us to genotype 1748 DNA samples of 38 different Bos taurus and Bos indicus cattle breeds from 13 countries on five continents (Europe, Africa, Asia, North America and South America), to examine the occurrence of the DGAT1 polymorphism and characterize the K232A substitution in cattle breeds of different origins and selected for different purposes (e.g., beef, dairy and dual purpose). Calculating pairwise FST values for pooled subpopulations showed least divergence for Bos indicus breeds with high milk fat percentage. Fixation of DGAT1A was found in some Bos taurus breeds and fixation of DGAT1K in one Bos indicus breed. Breeds of no known organized breeding background from the Near East domestication centre of Bos taurus and taurine African N'Dama cattle were found to possess intermediate frequencies of DGAT1K. While beef breeds tended to harbour higher DGAT1A levels, dairy cattle showed everything from very low levels of DGAT1K to unexpectedly high frequencies of this allele.     

A statistical model for identifying proteins by tandem mass spectrometry

A statistical model is presented for computing probabilities that proteins are present in a sample on the basis of peptides assigned to tandem mass (MS/MS) spectra acquired from a proteolytic digest of the sample. Peptides that correspond to more than a single protein in the sequence database are apportioned among all corresponding proteins, and a minimal protein list sufficient to account for the observed peptide assignments is derived using the expectation-maximization algorithm. Using peptide assignments to spectra generated from a sample of 18 purified proteins, as well as complex H. influenzae and Halobacterium samples, the model is shown to produce probabilities that are accurate and have high power to discriminate correct from incorrect protein identifications. This method allows filtering of large-scale proteomics data sets with predictable sensitivity and false positive identification error rates. Fast, consistent, and transparent, it provides a standard for publishing large-scale protein identification data sets in the literature and for comparing the results obtained from different experiments.     

Indications for the applicability of element signature analysis for the determination of the geographic origin of dried beef and poultry meat

In order to determine the geographic origin of poultry and dried beef, concentrations of a total of 72 different elements (occasionally represented with several isotopes) were analyzed by inductively coupled plasma high resolution mass spectrometry (ICP-HRMS). Additionally, gross chemical composition (GCC) was analyzed. The 25 poultry breast filets samples originated from Switzerland, France, Germany, Hungary, Brazil, and Thailand, and the 23 dried beef samples, made from M. biceps femoris and M. semitendinosus, were produced in Switzerland, Austria, Australia, United States, and Canada out of raw meat originating either from these or from other countries. A total of 66 and 46 of the elements and isotopes followed were detected in beef and poultry, respectively. For statistical analyses, only the most abundant isotopes per element were used. For both poultry meat and dried beef, a differentiation of the origins was possible using those elements, which were significantly different across countries (As, Na, Rb, and Tl in poultry; B, Ca, Cd, Cu, Dy, Eu, Ga, Li, Ni, Pd, Rb, Sr, Te, Tl, Tm, V, Yb, and Zn in beef). No sufficient differentiation between origins was possible with GCC. Further studies have to confirm the suitability of this approach for meat authentication with more samples.     

Kinds of patients

The basic goal of health outcomes research is to identify the kinds of patients who do (or do not) benefit substantially from specified medical or surgical treatments and procedures. Similarly, clinicians must determine whether particular patients are the kinds of patients who do (or do not) benefit from specified interventions. Such a kinds-based approach to clinical practice is often resisted, however, when physicians are asked to standardize their practices based on the results of health outcome data. In such settings, clinicians often assert that "every patient is unique". The present paper explores the coherence of this claim. In particular, I examine the applicability of the philosophical notion of natural kinds to a kinds-based approach to clinical research and practice. I conclude that the claim of patient uniqueness is misguided. Two key difficulties with a kinds-based approach are examined: the problems associated with (1) assigning single-case probabilities and (2) stereotyping and discrimination.