Search

Figure S1. Multi-Dimensional-Scaling of the distance matrix underlying the phylogenetic tree, based on chromosomes 1, 8 and 13. (DOCX 95 kb)

In the analysis and prediction of real-world systems, two of the key problems are nonstationarity (often in the form ofswitching between regimes) and overfitting (particularly serious for noisy processes). This article addresses these problems using gated experts, consisting of a (nonlinear) gating network and several (also nonlinear) competing experts. Each expert learns to predict the conditional mean and each expert adapts its width to match the noise level in its regime. The gating network learns to predict the probability of each expert, given the input. This article focuses on the case where the gating network bases its decision on information from the inputs. This can be contrasted to hidden Markov models where the decision is based on the previous state(s) (i.e., on the output of the gating network at the previous time step), as well as to averaging over several predictors. In contrast, gated experts soft-partition the input space. This article discusses the underlying statistical assumptions, derives the weight update rules, and compares the performance of gated experts to standard methods on three time series: (1) computer-generated series, obtained by randomly switching between two nonlinear processes, (2) a time series from the Santa Fe Time Series Competition (the light intensity of laser in chaotic state), and (3) the daily electricity demand of France, a real-world multivariate problem with structure on several timescales. The main results are (1) the gating network correctly discovers the different regimes of the process, (2) the widths associated with each expert are important for the segmentation task (and they can be used to characterize the subprocesses), and (3) there is less overfitting compared to single networks (homogeneous multilayer perceptrons), since the experts learn to match their variances to the (local) noise levels. This can be viewed as matching the local complexity of the model to the local complexity of the data.

Three genetically different clones of Toxoplasma gondii, also different in mouse virulence, were studied by experimental infection in chickens. For the experiments, four chicken lines were used, which differed in phylogenetic origin and performance level: two white egg layer lines, one with high laying performance (WLA), one with low (R11) and two brown layer lines, also displaying high (BLA) and low (L68) egg number. Chickens were intraperitoneally infected with three different T. gondii isolates representing type IIxIII recombinant clones, i.e. showing both, type II- and type III-specific alleles. These clones (K119/2 2C10, B136/1 B6H6, K119/2 A7) had exhibited virulence differences in a mouse model. In chickens, a significantly higher mortality was observed in white layer lines, but not in brown layer lines, suggesting that differences in the phylogenetic background may influence the susceptibility of chickens for toxoplasmosis. In addition, antibody (IgY) levels varied in surviving chickens at 31 days post infection. While low to intermediate antibody levels were observed in white layers, intermediate to high levels were measured in brown layers. Infection with a T. gondii clone showing low chicken virulence resulted in higher antibody levels in all chicken lines compared to infection with T. gondii clones of intermediate or high chicken virulence. This was in agreement with the parasite load as determined by real-time PCR. Overall, results show that progeny resulting from natural sexual recombination of T. gondii clonal lineages, may differ in their virulence for mice and chickens., Facultad de Ciencias Veterinarias

Ruskeissa kananmunissa esiintyvä kalanhaju on aiheuttanut mittavia ongelmia niin kananjalostukselle, tuottajille kuin kaupallekin. Ongelma koskettaa erityisesti eurooppalaisia kuluttajia, sillä EU-alueella yli 90% kulutetuista munista on ruskeita. Suomessa ongelma on toistaiseksi ollut marginaalinen, sillä kulutustottumukset ovat suosineet valkoisia munia, ruskeiden myyntiosuuden jäädessä alle kymmeneen prosenttiin. Kuluttajan näkökulmasta ongelma kärjistyy ruoanlaitossa: hajun on toisinaan havaittu voimistuvan käsiteltäessä munia kuten esimerkiksi keitettäessä tai paistettaessa. Kaupan laadunvalvonnan satunnaistesteistä löydetyt kalanhajuiset munat voivat johtaa koko sisäänostoerän hylkäämiseen. Pahimmassa tapauksessa tuottaja joutuu vaihtamaan koko kanalansa tuotantoeläimet muutaman haisevan munan takia. Hajun aiheuttava geeni tunnistettiin geenikartoituksen avulla. Taustalla on geenivirhe, joka johtaa hajuhaittaan, kuitenkin vain yhdessä tietynlaisen, yleensä rypsi- tai rapsipohjaisen ruokinnan kanssa. Valkoisilta kanoilta tämä geenivirhe puuttuu. Ruskeidenkaan kanojen munat eivät haise, elleivät geenivirheen perineet kanat ole saaneet rehua, joka sisältää hajun aiheuttajaa. Yleensä noin noin 5-10 prosenttia ruskeista kanoista voi tuottaa haisevia munia. Perinteisesti DNA-tutkimuksiin on vaadittu verinäytettä, jonka ottaminen voi pahimmillaan olla kanalle stressaava ja kivulias kokemus. Sen sijaan on mahdollista käyttää testauksessa höyheniä: jo yhdessä höyhentupessa on riittävä määrä DNA:ta, josta testi voidaan tehdä. Geenivirheen testaamiseksi onkin kehitetty höyhenestä tehtävä nopea ja yksinkertainen laboratoriomenetelmä. Menetelmän ansiosta voidaan ruskeita kanalinjoja valita risteytettäviksi niin, ettei jälkeläisissä ilmene hajuhaittaa aiheuttavaa geenivirhettä. Testausmenetelmälle on haettu kansainvälistä patenttia. Tällä menetelmällä jalostettujen kanalinjojen markkinointi aloitetaan vuoden lopussa ja ensimmäiset hajuttomat untuvikot ovat myynnissä elokuussa 2006. Nämä kanat aloittavat munintansa vuoden 2007 alussa. Testin ansiosta luomutuotannossa on mahdollista käyttää rehun valkuaislähteinä kotimaista rypsiä ja rapsia, mahdollisesti siirryttäessä 100 %:n luomuruokintaan.

The goal of the workshop on Implicit Measures was to explore how various implicit measures of user interest can be used in information retrieval and filtering applications. In most information retrieval or filtering applications, it is difficult to get explicit feedback from users about the relevance of the results, the appropriateness of the presentation, and more generally about the quality of their experience. Yet explicit judgments are assumed by researchers for many activities like the tuning and selection of ranking algorithms, information combination, user modeling, information presentation, etc. This workshop explored how implicit measures of user interest (such as dwell time, click through, and user activities like annotation, printing, and purchasing) can be used to develop predictive models for a variety of purposes. In the context of information retrieval such models could be used to: improve ranking and relevance assessment (e.g., the extent to which implicit measures can be used to evaluate the quality of systems, ranking algorithms and recommendations, or as input to relevance feedback algorithms); personalize search, filtering or presentation; personalize considering both individual and aggregate data; adaptive link generation or web site design; etc. An examination of theoretical issues such as modeling approaches (Bayesian techniques and other predictive models), gold standards for user behavior (e.g., relevance judgments, purchases), combining implicit and explicit preferences, and biases introduced by reliance on implicit measures were also encouraged. Since this was the first workshop on the topic at SIGIR, we encouraged participation from people with different backgrounds and perspectives including theoretical modeling, experimental analysis, and applications development. Applications from information retrieval, collaborative filtering, e-commerce, user modeling, and human-computer interaction were encouraged. Participation from both academia and industry was encouraged. As an outcome of the workshop we hoped to identify key theoretical modeling issues, systematize engineering principles and best practices, and spark new research directions. The workshop was held on August 1, 2003 in Toronto, Canada, immediately following the SIGIR 2003 conference. The workshop drew about 35 participants, representing a nice mix of academia and industry, and information retrieval and learning. To encourage interaction among participants, we limited the number and length of the talks to 15 minutes. This was followed by a 5 minute discussion by one of the organizers and additional open discussion among participants. This seemed to work well as a way to encourage interaction. We also had poster presentations available during the lunch break, which worked well, although only two posters were presented.

Summary Two Chinese domestic yak populations representing the Plateau type and the Huanhu Alpine type were analysed with 12 bovine microsatellite primers. All primer pairs functioned in the yak genome and polymorphism was found at all loci. The allele size ranges and frequencies of the two yak populations were similar and there was considerable overlap with the allele size ranges observed in cattle. Data for European cattle breeds was obtained from the Cattle Diversity Database (CaDBase) to interpret the heterozygosity and genetic distance estimates in yak populations. Heterozygosity estimated for the two yak populations was comparable with that of European cattle while Nei's Genetic Distance DA between the two yak populations was less than distances between the most closely related German cattle breeds. Bovine microsatellite primers proved to be a valuable tool for characterization of yak populations. Zusammenfassung Untersuchungen von zwei chinesischen Yak (Bos grunniens) Herden mittels Primersequenzen von Rindermikrosatelliten Primersequenzen von Rindermikrosatelliten wurden verwendet, um zwei chinesische Yak-Herden zu analysieren, eine vom Plateau-Typ und eine andere vom Huanhu-Bergtyp. Alle 12 Primerpaare fuhrten zu polymorphen Amplifikationsprodukten mit den DNA-Proben der Yaks. Der Allelgrosenbereich und die Allelfrequenzen aller 12 Marker waren in den beiden Yak-Herden sehr ahnlich. Zum Vergleich wurden Typisierungsergebnisse der 12 Mikrosatellitengenorte von 15 europaischen Rinderassen aus der Rinderdiversitatsdatenbank CaDBase entnommen. Ein groser Teil der Yak-Allele hatte die gleiche Grose wie beim Rind und auch der geschatzte Heterozygotiegrad der beiden Yak-Populationen war vergleichbar mit dem durchschnittlichen Heterozygotiegrad der 15 Rinderrassen. Die genetische Distanz (Nei's DA) zwischen beiden Yak-Herden wurde jedoch geringer geschatzt als zwischen zwei Paaren eng verwandter Rinderrassen. Die Ergebnisse zeigen, dass Sequenzinformationen von Rindermikrosatellitenprimern fur die Charakterisierung von Yak-Populationen auf der molekularen Ebene nutzlich sein konnen.

Contents: Part I:Neuroscience. T. Rebotier, J. Droulez, Sigma-Pi Properties of Spiking Neurons. H.S. Wan, D.S. Touretzky, A.D. Redish, Towards a Computational Theory of Rat Navigation. H.T. Blair, Evaluating Connectionist Models in Psychology and Neuroscience. Part II:Vision. J. Sirosh, R. Miikkulainen, Self-Organizing Feature Maps with Lateral Connections: Modeling Ocular Dominance. A.K. Bhattacharjya, B. Roysam, Joint Solution of Low, Intermediate, and High Level Vision Tasks by Global Optimization: Application to Computer Vision at Low SNR. T.B. Ghiselli-Crippa, P.W. Munro, Learning Global Spatial Structures from Local Associations. Part III:Cognitive Modeling. D. Ascher, A Connectionist Model of Auditory Morse Code Perception. V. Dragoi, J.E.R. Staddon, A Competitive Neural Network Model for the Process of Recurrent Choice. A.M. Lindemann, A Neural Network Simulation of Numerical Verbal-to-Arabic Transcoding. T. Lund, Combining Models of Single-Digit Arithmetic and Magnitude Comparison. I.E. Dror, Neural Network Models as Tools for Understanding High-Level Cognition: Developing Paradigms for Cognitive Interpretation of Neural Network Models. Part IV:Language. F.J. Eisenhart, Modeling Language as Sensorimotor Coordination. A. Govindjee, G. Dell, Structure and Content in Word Production: Why It's Hard to Say Dlorm. P. Gupta, Investigating Phonological Representations: A Modeling Agenda. H. Schutze, Y. Singer, Part-of-Speech Tagging Using a Variable Context Markov Model. M. Spivey-Knowlton, Quantitative Predictions from a Constraint-Based Theory of Syntactic Ambiguity Resolution. B.B. Tesar, Optimality Semantics. Part V:Symbolic Computation and Rules. K.G. Daugherty, M. Hare, What's in a Rule? The Past Tense by Some Other Name Might Be Called a Connectionist Net. A. Almor, M. Rindner, On the Proper Treatment of Symbolism -- A Lesson from Linguistics. L.F. Niklasson, Structure Sensitivity in Connectionist Models. M. Crucianu, Looking for Structured Representations in Recurrent Networks. I. Tchoumatchenko, Back Propagation with Understandable Results. M.W. Craven, J.W. Shavlik, Understanding Neural Networks via Rule Extraction and Pruning. A-H. Tan, Rule Learning and Extraction with Self-Organizing Neural Networks. Part VI:Recurrent Networks and Temporal Pattern Processing. J.F. Kolen, Recurrent Networks: State Machines or Iterated Function Systems? F. Cummins, R.F. Port, On the Treatment of Time in Recurrent Neural Networks. J.D. McAuley, Finding Metrical Structure in Time. C. Stevens, J. Wiles, Representations of Tonal Music: A Case Study in the Development of Temporal Relationships. M.A.S. Potts, D.S. Broomhead, J.P. Huke, Applications of Radial Basis Function Fitting to the Analysis of Dynamical Systems. M.E. Young, T.M. Bailey, Event Prediction: Faster Learning in a Layered Hebbian Network with Memory. Part VII:Control. S. Thrun, A. Schwartz, Issues in Using Function Approximation for Reinforcement Learning. P. Sabes, Approximating Q-Values with Basis Function Representations. K.L. Markey, Efficient Learning of Multiple Degree-of-Freedom Control Problems with Quasi-Independent Q-Agents. A.L. Tascillo, V.A. Skormin, Neural Adaptive Control of Systems with Drifting Parameters. Part VIII:Learning Algorithms and Architectures. R.C. O'Reilly, Temporally Local Unsupervised Learning: The MaxIn Algorithm for Maximizing Input Information. V.R. de Sa, Minimizing Disagreement for Self-Supervised Classification. S.N. Lindstaedt, Comparison of Two Unsupervised Neural Network Models for Redundancy Reduction. Z. Ghahramani, Solving Inverse Problems Using an EM Approach to Density Estimation. M. Finke, K-R. Muller, Estimating A-Posteriori Probabilities Using Stochastic Network Models. Part IX:Learning Theory. A.S. Weigend, On Overfitting and the Effective Number of Hidden Units. R. Dodier, Increase of Apparent Complexity Is Due to Decrease of Training Set Error. G.B. Orr, T.K. Leen, Momentum and Optimal Stochastic Search. R. Garces, Scheme to Improve the Generalization Error. M.P. Perrone, General Averaging Results for Convex Optimization. R.A. Caruana, Multitask Connectionist Learning. Z. Cataltepe, Y.S. Abu-Mostafa, Estimating Learning Performance Using Hints. Part X:Simulation Tools. A. Jagota, A Simulator for Asynchronous Hopfield Models. A. Linden, An Object-Oriented Dataflow Approach for Better Designs of Neural Net Architectures.

Chicken genetic resources comprise a wide range of breeds and populations including red jungle fowl (the assumed progenitor of all domestic breeds), native and fancy breeds, middle level food producers, industrial stocks and specialised lines. Based on the suggestion that the more distant a breed or population is the more likely it is to carry unique genetic features, the assessment of genetic distances by means of molecular marker information may provide useful information for initial evaluation of chicken genetic resources. During the last two decades several molecular marker classes have become available. Variable numbers of tandem repeat loci, in particular microsatellites, have been successfully used in chicken diversity studies. Genetic diversity measures using the highly polymorphic variable number of tandem repeat lociyield reliable and accurate information for the study of genetic relationships between chicken populations. First results of the European project on chicken biodiversity (AVIANDIV) obtained from microsatellite typing in DNA pools of 51 diverse chicken breeds showed that jungle fowl populations, traditional unselected breeds and broiler lines appear to be widely heterogeneous populations that may include a large portion of the genetic diversity of the tested breeds. In contrast, highly selected strains (layers and experimental lines) are characterised by a lower polymorphism. They behave as outliers from the set of breeds sampled. Single nucleotide polymorphism is a new and very promising molecular marker system which offers opportunities to assess the genetic diversity in farm animal species differently by investigating the mode and extent of changes in certain positions in the genome.

This paper presents ‘hidden Markov experts’, a framework for predicting conditional probability distributions of future values of a time series. On daily S&P500 data, the out-of- sample performance is compared to several baselines including GARCH and ‘gated experts’. The evaluation of the full density shows improvement over all competitors. Since the performance for point-predictions is comparable to the other methods, the main advantage of hidden Markov experts is their use for conditional density forecasting. Copyright © 2000 John Wiley & Sons, Ltd.

With the recent dramatic increase in electronic access to documents, text categorization—the task of assigning topics to a given document—has moved to the center of the information sciences and knowledge management. This article uses the structure that is present in the semantic space of topics in order to improve performance in text categorization: according to their meaning, topics can be grouped together into “meta-topics”, e.g., gold, silver, and copper are all metals. The proposed architecture matches the hierarchical structure of the topic space, as opposed to a flat model that ignores the structure. It accommodates both single and multiple topic assignments for each document. Its probabilistic interpretation allows its predictions to be combined in a principled way with information from other sources. The first level of the architecture predicts the probabilities of the meta-topic groups. This allows the individual models for each topic on the second level to focus on finer discriminations within the group. Evaluating the performance of a two-level implementation on the Reuters-22173 testbed of newswire articles shows the most significant improvement for rare classes.

This article exposes problems of the commonly used technique of splitting the available data into training, validation, and test sets that are held fixed, warns about drawing too strong conclusions from such static splits, and shows potential pitfalls of ignoring variability across splits. Using a bootstrap or resampling method, we compare the uncertainty in the solution stemming from the data splitting with neural network specific uncertainties (parameter initialization, choice of number of hidden units, etc.). We present two results on data from the New York Stock Exchange. First, the variation due to different resamplings is significantly larger than the variation due to different network conditions. This result implies that it is important to not over-interpret a model (or an ensemble of models) estimated on one specific split of the data. Second, on each split, the neural network solution with early stopping is very close to a linear model; no significant nonlinearities are extracted.

This study aimed to analyse the genetic diversity and population structure of five Ethiopian chicken ecotypes (N = 155), which were compared with six commercial purebreds (N = 180). For the analysis of genetic diversity, 26 AVIANDIV microsatellite markers were used. The number of alleles in Ethiopian ecotypes ranged from 2 to 19 per locus, with a mean of 6.1. The average observed heterozygosity within ecotype varied between 0.53 and 0.57. The overall heterozygote deficiency (F(IT)) in Ethiopian ecotypes was 0.124 ± 0.037. Over 68% of F(IT) was because of within-ecotype deficiency (F(IS)). In the phylogenetic tree, Ethiopian ecotypes clustered into two groups. The analysis of the relationship between populations using the structure program provided further evidence for the occurrence of at least two subgroups in the Ethiopian ecotypes. Findings of this study may provide the background for future studies to identify the origin of the two gene pools representing the Ethiopian chicken ecotypes and to characterize the gene variants influencing economically important traits.

Many studies in human genetics compare informativeness of single-nucleotide polymorphisms (SNPs) and microsatellites (single sequence repeats; SSR) in genome scans, but it is difficult to transfer the results directly to livestock because of different population structures. The aim of this study was to determine the number of SNPs needed to obtain the same differentiation power as with a given standard set of microsatellites. Eight chicken breeds were genotyped for 29 SSRs and 9216 SNPs. After filtering, only 2931 SNPs remained. The differentiation power was evaluated using two methods: partitioning of the Euclidean distance matrix based on a principal component analysis (PCA) and a Bayesian model-based clustering approach. Generally, with PCA-based partitioning, 70 SNPs provide a comparable resolution to 29 SSRs. In model-based clustering, the similarity coefficient showed significantly higher values between repeated runs for SNPs compared to SSRs. For the membership coefficients, reflecting the proportion to which a fraction segment of the genome belongs to the ith cluster, the highest values were obtained for 29 SSRs and 100 SNPs respectively. With a low number of loci (29 SSRs or ≤100 SNPs), neither marker types could detect the admixture in the Gödöllö Nhx population. Using more than 250 SNPs allowed a more detailed insight into the genetic architecture. Thus, the admixed population could be detected. It is concluded that breed differentiation studies will substantially gain power even with moderate numbers of SNPs.

This paper describes the fundamental change in paradigm in time series prediction that has occurred in the last decade. The examples are drawn from the Santa Fe Time Series Prediction and Analysis Competition . In 1979, the results of an earlier competition were published by the Royal Society. In that competition, and also in the follow up five years later, relatively large numbers of time series were provided (111 and 1001, respectively), taken from business (forecasting sales), economics (predicting recovery from the recession), finance, and the social sciences. However, all of the series used were very short, generally less than 100 values long. Most of the algorithms entered were fully automated, and most of the discussion centred around linear models. In contrast, the Santa Fe Competition focused on only six data-sets, ranging from 1000 to 100 000 points. All of the successful entries were fundamentally nonlinear and, even though significantly more computer power was used to analyse the larger data-sets with more complex models, the application of the techniques required more careful manual control than in the past. There was a general failure of simplistic ‘black-box’ approaches. In all successful entries, exploratory data analysis preceded the application of the algorithm. The Santa Fe Competition showed examples of nonlinear results going far beyond what is possible within the canon of linear systems analysis, but also showed that there are unprecedented opportunities for the analysis to go astray.