Bayesian Active Learning for Classification and Preference Learning

Information theoretic active learning has been widely studied for probabilistic models. For simple regression an optimal myopic policy is easily tractable. However, for other tasks and with more complex models, such as classification with nonparametric models, the optimal solution is harder to compute. Current approaches make approximations to achieve tractability. We propose an approach that expresses information gain in terms of predictive entropies, and apply this method to the Gaussian Process Classifier (GPC). Our approach makes minimal approximations to the full information theoretic objective. Our experimental performance compares favourably to many popular active learning algorithms, and has equal or lower computational complexity. We compare well to decision theoretic approaches also, which are privy to more information and require much more computational time. Secondly, by developing further a reformulation of binary preference learning to a classification problem, we extend our algorithm to Gaussian Process preference learning.

THE PHYLOGENETIC-RELATIONSHIPS AND CLASSIFICATION OF THE DOUCS AND SNUB-NOSED LANGURS OF CHINA AND VIETNAM

The taxonomy of the douc and snub-nosed langurs has changed several times during the 20th century. The controversy over the systematic position of these animals has been due in part to difficulties in studying them: both the doucs and the snub-nosed langurs are rare in the wild and are generally poorly represented in institutional collections. This review is based on a detailed examination of relatively large numbers of specimens of most of the species of langurs concerned. An attempt was made to draw upon as many types of information as were available in order to make an assessment of the phyletic relationships between the langur species under discussion. Toward this end, quantitative and qualitative features of the skeleton, specific features of visceral anatomy and characteristics of the pelage were utilized. The final data matrix comprised 178 characters. The matrix was analyzed using the program Hennig86. The results of the analysis support the following conclusions: (1) that the douc and snub-nosed langurs are generically distinct and should be referred to as species of Pygathrix and Rhinopithecus, respectively; (2) that the Tonkin snub-nosed langur be placed in its own subgenus as Rhinopithecus (Presbytiscus) avunculus and that the Chinese snub-nosed langur thus be placed in the subgenus Rhinopithecus (Rhinopithecus); (3) that four extant species of Rhinopithecus be recognized: R. (Rhinopithecus) roxellana Milne Edwards, 1870; R. (Rhinopithecus) bieti Milne Edwards, 1897; R. (Rhinopithecus) brelichi Thomas, 1903, and R. (Presbytiscus) avunculus Dollman, 1912; (4) that the Chinese snub-nosed langurs fall into northern and southern subgroups divided by the Yangtze river; (5) that R. lantianensis Hu and Qi, 1978, is a valid fossil species, and (6) the precise affinities and taxonomic status of the fossil species R. tingianus Matthew and Granger, 1923, are unclear because the type specimen is a subadult.

CLASSIFICATION AND EVOLUTION OF ASIAN COLOBINES

In order to study the differentiation of Asian colobines, 14 variables measured on 123 skulls, including Rhinopithecus, Presbytis, Presbytiscus (Rhinopithecus avunculus), Pygathrix and Nasalis were analyzed by one-way, cluster and discriminant function analyses. Information on paleoenvironmental changes in China and southeast Asia since the late Tertiary was used to examine the influences of migratory routes and range of distribution in Asian colobines. A cladogram for 6 genera of Asian colobines was constructed from the results of various analyses. Some new points or revisions were suggested: (1) Following one of two migratory routes, ancient species of Asian colobines perhaps passed through Xizang (Tibet) along the northern bank of the Tethys sea and through the Heng Duan Shan regions of Yunnan into Vietnam. An ancient landmass linking Yunnan and Xizang was already present on the east bank of the Tethys sea. Accordingly, Asian colobines would have two centers of evolutionary origin: Sundaland and the Heng Duan Shan regions of China. (2) Pygathrix shares more cranial features with Presbytiscus than with Rhinopithecus. This differs somewhat from the conclusion reached by Groves. (3) Nasalis (karyotype: 2n = 48) may be the most primitive genus among Asian colobines. Certain features shared with Rhinopithecus, e.g. large body size, terrestrial activity and limb proportions, can be interpreted as symple-siomorphic characters. (4) Rhinopithecus, with respect to craniofacial features, is a special case among Asian colobines. It combines a high degree of evolutionary specialization with retention of some primitive features thought to have been present in the ancestral Asian colobine.

An Online Expectation-Maximization Algorithm for Changepoint Models

Changepoint models are widely used to model the heterogeneity of sequential data. We present a novel sequential Monte Carlo (SMC) online Expectation-Maximization (EM) algorithm for estimating the static parameters of such models. The SMC online EM algorithm has a cost per time which is linear in the number of particles and could be particularly important when the data is representable as a long sequence of observations, since it drastically reduces the computational requirements for implementation. We present an asymptotic analysis for the stability of the SMC estimates used in the online EM algorithm and demonstrate the performance of this scheme using both simulated and real data originating from DNA analysis.

An Online Expectation-Maximization Algorithm for Changepoint Models

Changepoint models are widely used to model the heterogeneity of sequential data. We present a novel sequential Monte Carlo (SMC) online Expectation-Maximization (EM) algorithm for estimating the static parameters of such models. The SMC online EM algorithm has a cost per time which is linear in the number of particles and could be particularly important when the data is representable as a long sequence of observations, since it drastically reduces the computational requirements for implementation. We present an asymptotic analysis for the stability of the SMC estimates used in the online EM algorithm and demonstrate the performance of this scheme using both simulated and real data originating from DNA analysis.

Phrase linguistic classification and generalization for improving statistical machine translation

In this paper a method to incorporate linguistic information regarding single-word and compound verbs is proposed, as a first step towards an SMT model based on linguistically-classified phrases. By substituting these verb structures by the base form of the head verb, we achieve a better statistical word alignment performance, and are able to better estimate the translation model and generalize to unseen verb forms during translation. Preliminary experiments for the English - Spanish language pair are performed, and future research lines are detailed. © 2005 Association for Computational Linguistics.