Nested dictionary learning for hierarchical organization of imagery and text

A tree-based dictionary learning model is developed for joint analysis of imagery and associated text. The dictionary learning may be applied directly to the imagery from patches, or to general feature vectors extracted from patches or superpixels (using any existing method for image feature extraction). Each image is associated with a path through the tree (from root to a leaf), and each of the multiple patches in a given image is associated with one node in that path. Nodes near the tree root are shared between multiple paths, representing image characteristics that are common among different types of images. Moving toward the leaves, nodes become specialized, representing details in image classes. If available, words (text) are also jointly modeled, with a path-dependent probability over words. The tree structure is inferred via a nested Dirichlet process, and a retrospective stick-breaking sampler is used to infer the tree depth and width.

Lognormal and gamma mixed negative binomial regression

In regression analysis of counts, a lack of simple and efficient algorithms for posterior computation has made Bayesian approaches appear unattractive and thus underdeveloped. We propose a lognormal and gamma mixed negative binomial (NB) regression model for counts, and present efficient closed-form Bayesian inference; unlike conventional Poisson models, the proposed approach has two free parameters to include two different kinds of random effects, and allows the incorporation of prior information, such as sparsity in the regression coefficients. By placing a gamma distribution prior on the NB dispersion parameter r, and connecting a log-normal distribution prior with the logit of the NB probability parameter p, efficient Gibbs sampling and variational Bayes inference are both developed. The closed-form updates are obtained by exploiting conditional conjugacy via both a compound Poisson representation and a Polya-Gamma distribution based data augmentation approach. The proposed Bayesian inference can be implemented routinely, while being easily generalizable to more complex settings involving multivariate dependence structures. The algorithms are illustrated using real examples. Copyright 2012 by the author(s)/owner(s).

Cross-Domain Multitask Learning with Latent Probit Models

Learning multiple tasks across heterogeneous domains is a challenging problem since the feature space may not be the same for different tasks. We assume the data in multiple tasks are generated from a latent common domain via sparse domain transforms and propose a latent probit model (LPM) to jointly learn the domain transforms, and the shared probit classifier in the common domain. To learn meaningful task relatedness and avoid over-fitting in classification, we introduce sparsity in the domain transforms matrices, as well as in the common classifier. We derive theoretical bounds for the estimation error of the classifier in terms of the sparsity of domain transforms. An expectation-maximization algorithm is derived for learning the LPM. The effectiveness of the approach is demonstrated on several real datasets.

Communications inspired linear discriminant analysis

We study the problem of supervised linear dimensionality reduction, taking an information-theoretic viewpoint. The linear projection matrix is designed by maximizing the mutual information between the projected signal and the class label. By harnessing a recent theoretical result on the gradient of mutual information, the above optimization problem can be solved directly using gradient descent, without requiring simplification of the objective function. Theoretical analysis and empirical comparison are made between the proposed method and two closely related methods, and comparisons are also made with a method in which Rényi entropy is used to define the mutual information (in this case the gradient may be computed simply, under a special parameter setting). Relative to these alternative approaches, the proposed method achieves promising results on real datasets. Copyright 2012 by the author(s)/owner(s).

Scoup-SMT: Scalable Coupled Sparse Matrix-Tensor Factorization

How can we correlate neural activity in the human brain as it responds to words, with behavioral data expressed as answers to questions about these same words? In short, we want to find latent variables, that explain both the brain activity, as well as the behavioral responses. We show that this is an instance of the Coupled Matrix-Tensor Factorization (CMTF) problem. We propose Scoup-SMT, a novel, fast, and parallel algorithm that solves the CMTF problem and produces a sparse latent low-rank subspace of the data. In our experiments, we find that Scoup-SMT is 50-100 times faster than a state-of-the-art algorithm for CMTF, along with a 5 fold increase in sparsity. Moreover, we extend Scoup-SMT to handle missing data without degradation of performance. We apply Scoup-SMT to BrainQ, a dataset consisting of a (nouns, brain voxels, human subjects) tensor and a (nouns, properties) matrix, with coupling along the nouns dimension. Scoup-SMT is able to find meaningful latent variables, as well as to predict brain activity with competitive accuracy. Finally, we demonstrate the generality of Scoup-SMT, by applying it on a Facebook dataset (users, friends, wall-postings); there, Scoup-SMT spots spammer-like anomalies.

An innovative design and studio-based CS degree

The University of Queensland has recently established a new design-focused, studio-based computer science degree. The Bachelor of Information Environments degree augments the core courses from the University's standard CS degree with a stream of design courses and integrative studio-based projects undertaken every semester. The studio projects integrate and reinforce learning by requiring students to apply the knowledge and skills gained in other courses to open-ended real-world design projects. The studio model is based on the architectural studio and involves teamwork, collaborative learning, interactive problem solving, presentations and peer review. This paper describes the degree program, its curriculum and rationale, and reports on experiences in the first year of delivery.

The molecular structure of the borate mineral rhodizite (K, Cs)Al4Be4(B, Be)12O28 – A vibrational spectroscopic study

We have studied the borate mineral rhodizite (K, Cs)Al4Be4(B, Be)12O28 using a combination of DEM with EDX and vibrational spectroscopic techniques. The mineral occurs as colorless, gray, yellow to white crystals in the triclinic crystal system. The studied sample is from the Antandrokomby Mine, Sahatany valley, Madagascar. The mineral is prized as a semi-precious jewel. Semi-quantitative chemical composition shows a Al, Ca, borate with minor amounts of K, Mg and Cs. The mineral has a characteristic borate Raman spectrum and bands are assigned to the stretching and bending modes of B, Be and Al. No Raman bands in the OH stretching region were observed.

Gel growth and dielectric studies of K. Rb and Cs perchlorates

Single crystals of K, Rb and Cs perchlorates have been grown by the counter diffusion of the respective ions and CIO4 through the gel medium. Studies on nucleation, growth kinetics, morphological aspects and purity are discussed in this paper. The dielectric constant, ~b, as well as loss measured along the longest axis, exhibits an anomaly at the transition temperature, Tt, in all the three crystals. It is found that the peak values of Tt are approximately 800, 100 and 53 in K, Rb and Cs perchlorates, respectively. The dielectric anomaly and the large value of c b in the cubic phase are discussed in terms of the degree of disorder of the CIO~ group and the possible contribution from defects.

Synthesis of Layered Perovskite Oxides, ACa2-xLaxNb3-xTix010 (A = K, Rb, Cs), and Characterization of New Solid Acids, HCa2-xLaxNb3-xTixOlo (0 < x d 2), Exhibiting Variable Bronsted Acidity?

Layered perovskite oxides of the formula ACa~,La,Nb3-,Ti,010 (A = K, Rb, Cs and 0 < x d 2) have been prepared. The members adopt the structures of the parent ACazNb3010. Interlayer alkali cations in the niobium-titanium oxide series can be ion-exchanged with Li+, Na+, NH4+, or H+ to give new derivatives. Intercalation of the protonated derivatives with organic bases reveals that the Bronsted acidity of the solid solution series, HC~ ~ , L ~ ,N~ ~ , T ~ ,dOep~eOnd, s on the titanium content. While the x = 1 member (HCaLaNbzTiOlo) is nearly as acidic as the parent HCazNb3010, the x = 2 member (HLazNbTizOlo) is a weak acid hardly intercalating organic bases with pKa - 11.3. The variation of acidity is probably due to an ordering of Nb/Ti atoms in the triple octahedral perovskite slabs, [Ca~,La,Nb~,Ti,0~0], such that protons are attached to NbO6 octahedra in the x = 1 member and to Ti06 octahedra in the x = 2 member.

ChemInform Abstract: Synthesis of Anion-Deficient Layered Perovskites, ACa2Nb3-xMxO10-x (A: Rb, Cs; M: Al, Fe), Exhibiting Ion-Exchange and Intercalation. Evidence for the Formation of Layered Brownmillerites, ACa2Nb2AlO9 (A: Cs, H)

Anion-deficient layered perovskite oxides of the formula, ACa2Nb3-xMxO10-x (A = Rb, Cs; M = Al, Fe) for 0 < x less-than-or-equal-to 1.0, possessing tetragonal structures similar to the parent ACa2Nb3O10, have been synthesized. The interlayer A cations in these materials are readily exchanged with protons in aqueous HNO3 to give the protonated derivatives, HCa2Nb3-xMxO10-x; the latter are solid Bronsted acids intercalating a number of organic amines including aniline (pK(a) = 4.63). The distribution of acid sites in the interlayer region of HCa2Nb2MO9 inferred from n-alkylamine intercalation suggests that oxygen vacancies and Nb/M atoms are disordered in the ACa2Nb2MO9 samples prepared at 1100-1200-degrees-C. Annealing a disordered sample of CsCa2Nb2AlO9 for a long time at lower temperatures tends to order the Nb/Al atoms and oxygen vacancies to produce octahedral (NbO6/2)-tetrahedral (AlO4/2)-octahedral (NbO6/2) layer sequence reminiscent of the brownmillerite structure.

ChemInform Abstract: Slicing the Perovskite Structure into Layers: Synthesis of Novel Three- Dimensional and Layered Perovskite Oxides, ALaSrNb2M(II)O9 (A: Na, Cs)

ChemInform is a weekly Abstracting Service, delivering concise information at a glance that was extracted from about 100 leading journals. To access a ChemInform Abstract of an article which was published elsewhere, please select a “Full Text” option. The original article is trackable via the “References” option.

Preparation and characterization of Cd(0.75)PS(3)A(0.5)(H2O)(y) [A = Na, K and Cs]

Cd(0.75)PS(3)A(0.5)(H2O)(y) [A = Na, K and Cs], synthesized by the ion-exchange intercalation reaction of the insulating layered CdPS3, exhibits interesting electrical properties. The electrical properties are strongly dependent on the extent of hydration of the alkali ion which resides in the interlamellar space. In the potassium and caesium ion-exchanged compounds, y = I, the lattice expansion is similar to 3 Angstrom and the electric response characteristic of a dielectric. In the as prepared A = Na compound, y = 2, the lattice expansion is 5.6 Angstrom, the compound is conducting with a DC conductance of 3 x 10(-5) S cm(-1) at 300 K. Cd0.75PS3Na0.5(H2O)(y), y = 2, on evacuation or on heating looses water, reversibly, to form a y = 1 phase with electrical properties similar to that of the K and Cs ion exchange intercalation compounds.

A[Bi3Ti4O13] and A[Bi3PbTi5O16] (A = K, Cs): New n=4 and n=5 members of the layered perovskite series, A[A ' n-1B nO3n+1], and their hydrates

We describe the synthesis and structural characterization of new layered bismuth titanates, A[Bi3Ti4O13] and A[Bi3PbTi5O16]for A = K, Cs, corresponding to n = 4 and 5 members of the Dion-Jacobson series of layered perovskites of the general formula, A[A'n-1BnO3n+1]. These materials have been prepared by solid state reaction of the constituents containing excess alkali, which is required to suppress the formation of competitive Aurivillius phases. Unlike the isostructural niobates and niobium titanates of the same series, the new phases reported here are spontaneously hydrated-a feature which could make them potentially useful as photocatalysts for water splitting reaction. On hydration of the potassium compounds, the c axis expands by ca. 2 Angstrom and loses its doubling [for example, the tetragonal lattice parameters of K[Bi3Ti4O13] and its dihydrate are respectively a = 3900(1)Angstrom c 37.57(2) Angstrom; a 3.885(1) Angstrom, c = 20.82(4) Angstrom]; surprisingly, the cesium analogues do not show a similar change on hydration.