Bayesian and L1 Approaches to Sparse Unsupervised Learning

The use of L1 regularisation for sparse learning has generated immense research interest, with successful application in such diverse areas as signal acquisition, image coding, genomics and collaborative filtering. While existing work highlights the many advantages of L1 methods, in this paper we find that L1 regularisation often dramatically underperforms in terms of predictive performance when compared with other methods for inferring sparsity. We focus on unsupervised latent variable models, and develop L1 minimising factor models, Bayesian variants of "L1", and Bayesian models with a stronger L0-like sparsity induced through spike-and-slab distributions. These spike-and-slab Bayesian factor models encourage sparsity while accounting for uncertainty in a principled manner and avoiding unnecessary shrinkage of non-zero values. We demonstrate on a number of data sets that in practice spike-and-slab Bayesian methods outperform L1 minimisation, even on a computational budget. We thus highlight the need to re-assess the wide use of L1 methods in sparsity-reliant applications, particularly when we care about generalising to previously unseen data, and provide an alternative that, over many varying conditions, provides improved generalisation performance.

KLAIR: A virtual infant for spoken language acquisition research

Recent research into the acquisition of spoken language has stressed the importance of learning through embodied linguistic interaction with caregivers rather than through passive observation. However the necessity of interaction makes experimental work into the simulation of infant speech acquisition difficult because of the technical complexity of building real-time embodied systems. In this paper we present KLAIR: a software toolkit for building simulations of spoken language acquisition through interactions with a virtual infant. The main part of KLAIR is a sensori-motor server that supplies a client machine learning application with a virtual infant on screen that can see, hear and speak. By encapsulating the real-time complexities of audio and video processing within a server that will run on a modern PC, we hope that KLAIR will encourage and facilitate more experimental research into spoken language acquisition through interaction. Copyright © 2009 ISCA.

Cross-Lingual spoken language understanding from unaligned data using discriminative classification models and machine translation

This paper investigates several approaches to bootstrapping a new spoken language understanding (SLU) component in a target language given a large dataset of semantically-annotated utterances in some other source language. The aim is to reduce the cost associated with porting a spoken dialogue system from one language to another by minimising the amount of data required in the target language. Since word-level semantic annotations are costly, Semantic Tuple Classifiers (STCs) are used in conjunction with statistical machine translation models both of which are trained from unaligned data to further reduce development time. The paper presents experiments in which a French SLU component in the tourist information domain is bootstrapped from English data. Results show that training STCs on automatically translated data produced the best performance for predicting the utterance's dialogue act type, however individual slot/value pairs are best predicted by training STCs on the source language and using them to decode translated utterances. © 2010 ISCA.

Monoculture of three native air breathing fishes in homestead ditches and plastic barrels

Six on-farm trials were conducted from 1 August to 23 November 2004 in two different environments such as homestead ditches (10 to 17 square meters) and plastic barrels (240 liters) to develop techniques for nonoculture [sic] of climbing perch, Anabas testudineus, stinging catfish, Heteropneustes fossilis and walking catfish, Clarias batrachus for poor and landless people who have no access to pond. Stocking density for ditch was 10 fry/square meter while that for barrel was 20 fry/cubic meter. The fishes were fed with 3-test diets viz. low-cost formulated feed (rice bran 20%, wheat meal 10%, mustered [sic] oil cake 35%, poultry offal 35%), live foods (chopped snails and clams), and a commercial feed (Saudi-Bangla feed, starter 3: first month and grower-1: subsequent two months) and designated as T1, T2 and T3, respectively. Feeding rate was the same in all the treatments viz. 10% of body weight (first two months), 8% (third month) and 6% (fourth month). T1 and T2 had three replications while T3 had two replications. Water temperature was recorded weekly while fish growth was monitored monthly. After 4 months' rearing, H. fossilis and C. batrachus in ditches and barrels attained higher average weight in T2 followed by T3 and T1 while A. testudineus in barrels also attained higher average weight in T2. The variation in net weight gain by A. testudineus in ditch fed test diets T2 and T3 was not significantly different (P>0.05) though the net gain in both T2 and T3 was significantly (P<0.05) higher than that of T1. The yield of climbing perch as obtained from T1, T2 and T3 was 988, 1136 and 1185 kg/ha, respectively while that stinging catfish was 395, 242 and 444 kg/ha and walking catfish was 1605, 2,099 and 1,654 kg/ha respectively. All the three species showed significantly lower growth rate in barrels than in ditches.

Genetic diversity of Chinese native pigs inferred from protein electrophoresis

We examined protein polymorphism of 20 native pig breeds in China and 3 introduced pig breeds. Thirty loci have been investigated, among which six loci were found to be polymorphic. Especially, the polymorphism of malate dehydrogenase (MDH), adenylate kinase (AK), and two new alleles of adenosine deaminase (ADA) had not been reported in domestic pigs and wild pigs. The percentage of polymorphic loci (P), the mean heterozygosity (H), and the mean number of alleles (A) are 0.200, 0.065, and 1.300, respectively. The degree of genetic variability of Chinese pigs as a whole was higher than that of goats, lower than that of cattle and horses, and similar to that of sheep. Using the gene frequencies of the 30 loci, Nei's genetic distance among the 20 native breeds in China and 3 introduced pig breeds was calculated by the formula of Nei. The program NEIGHBOR in PHYLIP 3.5c was chosen to construct an UPGMA tree and a NJ tree. Our results show that, of the total genetic variation found in the native pig breeds in China, 31% (0.31) is ascribable to genetic differences among breeds. About 69% of the total genetic variation is found within breeds. Most breeds are in linkage disequilibrium. The patterns of genetic similarities between the Chinese native pig breeds were not in agreement with the proposed pig type classification.

The origin and genetic diversity of Chinese native chicken breeds

The first 539 bases of mitochondrial DNA D-loop region of six Chinese native chicken breeds (Gallus gallus domesticus) were sequenced and compared to those of the red junglefowl (Gallus gallus), the gray junglefowl (Gallus sonneratii), the green junglefow