Prototype learning methods for online handwriting recognition

In this paper, we study different methods for prototype selection for recognizing handwritten characters of Tamil script. In the first method, cumulative pairwise- distances of the training samples of a given class are used to select prototypes. In the second method, cumulative distance to allographs of different orientation is used as a criterion to decide if the sample is representative of the group. The latter method is presumed to offset the possible orientation effect. This method still uses fixed number of prototypes for each of the classes. Finally, a prototype set growing algorithm is proposed, with a view to better model the differences in complexity of different character classes. The proposed algorithms are tested and compared for both writer independent and writer adaptation scenarios.

Active learning in keyword search-based data integration

The problem of scaling up data integration, such that new sources can be quickly utilized as they are discovered, remains elusive: Global schemas for integrated data are difficult to develop and expand, and schema and record matching techniques are limited by the fact that data and metadata are often under-specified and must be disambiguated by data experts. One promising approach is to avoid using a global schema, and instead to develop keyword search-based data integration-where the system lazily discovers associations enabling it to join together matches to keywords, and return ranked results. The user is expected to understand the data domain and provide feedback about answers' quality. The system generalizes such feedback to learn how to correctly integrate data. A major open challenge is that under this model, the user only sees and offers feedback on a few ``top-'' results: This result set must be carefully selected to include answers of high relevance and answers that are highly informative when feedback is given on them. Existing systems merely focus on predicting relevance, by composing the scores of various schema and record matching algorithms. In this paper, we show how to predict the uncertainty associated with a query result's score, as well as how informative feedback is on a given result. We build upon these foundations to develop an active learning approach to keyword search-based data integration, and we validate the effectiveness of our solution over real data from several very different domains.

Natural actor-critic algorithms

We present four new reinforcement learning algorithms based on actor-critic, natural-gradient and functi approximation ideas,and we provide their convergence proofs. Actor-critic reinforcement learning methods are online approximations to policy iteration in which the value-function parameters are estimated using temporal difference learning and the policy parameters are updated by stochastic gradient descent. Methods based on policy gradients in this way are of special interest because of their compatibility with function-approximation methods, which are needed to handle large or infinite state spaces. The use of temporal difference learning in this way is of special interest because in many applications it dramatically reduces the variance of the gradient estimates. The use of the natural gradient is of interest because it can produce better conditioned parameterizations and has been shown to further reduce variance in some cases. Our results extend prior two-timescale convergence results for actor-critic methods by Konda and Tsitsiklis by using temporal difference learning in the actor and by incorporating natural gradients. Our results extend prior empirical studies of natural actor-critic methods by Peters, Vijayakumar and Schaal by providing the first convergence proofs and the first fully incremental algorithms.

Incremental natural-gradient actor-critic algorithms

We present four new reinforcement learning algorithms based on actor-critic and natural-gradient ideas, and provide their convergence proofs. Actor-critic rein- forcement learning methods are online approximations to policy iteration in which the value-function parameters are estimated using temporal difference learning and the policy parameters are updated by stochastic gradient descent. Methods based on policy gradients in this way are of special interest because of their com- patibility with function approximation methods, which are needed to handle large or infinite state spaces. The use of temporal difference learning in this way is of interest because in many applications it dramatically reduces the variance of the gradient estimates. The use of the natural gradient is of interest because it can produce better conditioned parameterizations and has been shown to further re- duce variance in some cases. Our results extend prior two-timescale convergence results for actor-critic methods by Konda and Tsitsiklis by using temporal differ- ence learning in the actor and by incorporating natural gradients, and they extend prior empirical studies of natural actor-critic methods by Peters, Vijayakumar and Schaal by providing the first convergence proofs and the first fully incremental algorithms.

Anticancer property of Bryophyllum pinnata (Lam.) Oken. leaf on human cervical cancer cells

Background: Bryophyllum pinnata (B. pinnata) is a common medicinal plant used in traditional medicine of India and of other countries for curing various infections, bowel diseases, healing wounds and other ailments. However, its anticancer properties are poorly defined. In view of broad spectrum therapeutic potential of B. pinnata we designed a study to examine anti-cancer and anti-Human Papillomavirus (HPV) activities in its leaf extracts and tried to isolate its active principle. Methods: A chloroform extract derived from a bulk of botanically well-characterized pulverized B. pinnata leaves was separated using column chromatography with step-gradient of petroleum ether and ethyl acetate. Fractions were characterized for phyto-chemical compounds by TLC, HPTLC and NMR and Biological activity of the fractions were examined by MTT-based cell viability assay, Electrophoretic Mobility Shift Assay, Northern blotting and assay of apoptosis related proteins by immunoblotting in human cervical cancer cells. Results: Results showed presence of growth inhibitory activity in the crude leaf extracts with IC50 at 552 mu g/ml which resolved to fraction F4 (Petroleum Ether: Ethyl Acetate:: 50: 50) and showed IC50 at 91 mu g/ml. Investigations of anti-viral activity of the extract and its fraction revealed a specific anti-HPV activity on cervical cancer cells as evidenced by downregulation of constitutively active AP1 specific DNA binding activity and suppression of oncogenic c-Fos and c-Jun expression which was accompanied by inhibition of HPV18 transcription. In addition to inhibiting growth, fraction F4 strongly induced apoptosis as evidenced by an increased expression of the pro-apoptotic protein Bax, suppression of the anti-apoptotic molecules Bcl-2, and activation of caspase-3 and cleavage of PARP-1. Phytochemical analysis of fraction F4 by HPTLC and NMR indicated presence of activity that resembled Bryophyllin A. Conclusions: Our study therefore demonstrates presence of anticancer and anti-HPV an activity in B. pinnata leaves that can be further exploited as a potential anticancer, anti-HPV therapeutic for treatment of HPV infection and cervical cancer.

Optical microscopy methods for understanding learning and memory

Structural dynamics of dendritic spines is one of the key correlative measures of synaptic plasticity for encoding short-term and long-term memory. Optical studies of structural changes in brain tissue using confocal microscopy face difficulties of scattering. This results in low signal-to-noise ratio and thus limiting the imaging depth to few tens of microns. Multiphoton microscopy (MpM) overcomes this limitation by using low-energy photons to cause localized excitation and achieve high resolution in all three dimensions. Multiple low-energy photons with longer wavelengths minimize scattering and allow access to deeper brain regions at several hundred microns. In this article, we provide a basic understanding of the physical phenomena that give MpM an edge over conventional microscopy. Further, we highlight a few of the key studies in the field of learning and memory which would not have been possible without the advent of MpM.

The role of lysine-41 in RNase A catalysis - A quantum chemical study on the active site ligand complex

Several mechanisms have been proposed to explain the action of enzymes at the atomic level. Among them, the recent proposals involving short hydrogen bonds as a step in catalysis by Gerlt and Gassman [1] and proton transfer through low barrier hydrogen bonds (LBHBs) [2, 3] have attracted attention. There are several limitations to experimentally testing such hypotheses, Recent developments in computational methods facilitate the study of active site-ligand complexes to high levels of accuracy, Our previous studies, which involved the docking of the dinucleotide substrate UpA to the active site of RNase A [4, 5], enabled us to obtain a realistic model of the ligand-bound active site of RNase A. From these studies, based on empirical potential functions, we were able to obtain the molecular dynamics averaged coordinates of RNase A, bound to the ligand UpA. A quantum mechanical study is required to investigate the catalytic process which involves the cleavage and formation of covalent bonds. In the present study, we have investigated the strengths of some of the hydrogen bonds between the active site residues of RNase A and UpA at the ab initio quantum chemical level using the molecular dynamics averaged coordinates as the starting point. The 49 atom system and other model systems were optimized at the 3-21G level and the energies of the optimized systems were obtained at the 6-31G* level. The results clearly indicate the strengthening of hydrogen bonds between neutral residues due to the presence of charged species at appropriate positions. Such a strengthening manifests itself in the form of short hydrogen bonds and a low barrier for proton transfer. In the present study, the proton transfer between the 2'-OH of ribose (from the substrate) and the imidazole group from the H12 of RNase A is influenced by K41, which plays a crucial role in strengthening the neutral hydrogen bond, reducing the barrier for proton transfer.

Molecular structure of Boc-Aib-Aib-Phe-Met-NH2·DMSO. A fragment of a biologically active enkephalin analogue

The tetrapeptide t-butyloxycarbonyl--aminoisobutyryl--aminoisobutyryl-L- phenylalanyl-L-methionyl amide crystallizes in the orthorhombic space group P212121 with a= 9.096, b= 18.067, c= 21.701 Å and Z= 4. The crystals contain one molecule of dimethyl sulphoxide (DMSO) associated with each peptide. The structure has been solved by direct methods and refined to an R value of 0.103 for 2 672 observed reflections. The peptide adopts a distorted 310 helical structure stabilized by two intramolecular 4 1 hydrogen bonds between the Boc CO and Aib(1) CO groups and the NH groups of Phe(3) and Met(4), respectively. A long hydrogen bond (N O = 3.35 Å) is also observed between Aib(2) CO and one of the terminal amide hydrogens. The DMSO molecule is strongly hydrogen bonded to the Aib(1) NH group. The solid-state conformation agrees well with proposals made on the basis of n.m.r. studies in solution.

Can Nano Silicon Be Made Optically Active?

There has been a lot of effort to make Silicon optically active. In this work we examine two methods of generating nanocrystals of Silicon from bulk fragments. This approach of ours allows us to play with the shape of the nanocrystals and therefore the degeneracy of the conduction band minimum. We go on to examine whether similar sized particles with different shapes have the same physical properties, and finally whether Silicon may be rendered optically active by this route. While we do find that similar sized particles with different shapes may have different band gaps, this route of modifying the degeneracy of the conduction band minimum makes nano Si slightly optically active.

A Team of Continuous-Action Learning Automata for Noise-Tolerant Learning of Half-Spaces

Learning automata are adaptive decision making devices that are found useful in a variety of machine learning and pattern recognition applications. Although most learning automata methods deal with the case of finitely many actions for the automaton, there are also models of continuous-action-set learning automata (CALA). A team of such CALA can be useful in stochastic optimization problems where one has access only to noise-corrupted values of the objective function. In this paper, we present a novel formulation for noise-tolerant learning of linear classifiers using a CALA team. We consider the general case of nonuniform noise, where the probability that the class label of an example is wrong may be a function of the feature vector of the example. The objective is to learn the underlying separating hyperplane given only such noisy examples. We present an algorithm employing a team of CALA and prove, under some conditions on the class conditional densities, that the algorithm achieves noise-tolerant learning as long as the probability of wrong label for any example is less than 0.5. We also present some empirical results to illustrate the effectiveness of the algorithm.

Amino acid sequence of the winged bean (Psophocarpus tetragonolobus) basic lectin. Adenine binding and identification of the active-site tryptophan residue

The complete amino acid sequence of winged bean basic agglutinin (WBA I) was obtained by a combination of manual and gas-phase sequencing methods. Peptide fragments for sequence analyses were obtained by enzymatic cleavages using trypsin and Staphylococcus aureus V8 endoproteinase and by chemical cleavages using iodosobenzoic acid, hydroxylamine, and formic acid. COOH-terminal sequence analysis of WBA I and other peptides was performed using carboxypeptidase Y. The primary structure of WBA I was homologous to those of other legume lectins and more so to Erythrina corallodendron. Interestingly, the sequence shows remarkable identities in the regions involved in the association of the two monomers of E. corallodendron lectin. Other conserved regions are the double metal-binding site and residues contributing to the formation of the hydrophobic cavity and the carbohydrate-binding site. Chemical modification studies both in the presence and absence of N-acetylgalactosamine together with sequence analyses of tryptophan-containing tryptic peptides demonstrate that tryptophan 133 is involved in the binding of carbohydrate ligands by the lectin. The location of tryptophan 133 at the active center of WBA I for the first time subserves to explain a role for one of the most conserved residues in legume lectins.

NDDO-based CI methods for the prediction of electronic spectra and sum-over-states molecular hyperpolarization

NDDO-based (AM1) configuration interaction (CI) calculations have been used to calculate the wavelength and oscillator strengths of electronic absorptions in organic molecules and the results used in a sum-over-states treatment to calculate second-order-hyperpolarizabilities. The results for both spectra and hyperpolarizabilities are of acceptable quality as long as a suitable CI-expansion is used. We have found that using an active space of eight electrons in eight orbitals and including all single and pair-double excitations in the CI leads to results that agree well with experiment and that do not change significantly with increasing active space for most organic molecules. Calculated second-order hyperpolarizabilities using this type of CI within a sum-over-states calculation appear to be of useful accuracy.

Structured Learning for Non-Smooth Ranking losses

Learning to rank from relevance judgment is an active research area. Itemwise score regression, pairwise preference satisfaction, and listwise structured learning are the major techniques in use. Listwise structured learning has been applied recently to optimize important non-decomposable ranking criteria like AUC (area under ROC curve) and MAP(mean average precision). We propose new, almost-lineartime algorithms to optimize for two other criteria widely used to evaluate search systems: MRR (mean reciprocal rank) and NDCG (normalized discounted cumulative gain)in the max-margin structured learning framework. We also demonstrate that, for different ranking criteria, one may need to use different feature maps. Search applications should not be optimized in favor of a single criterion, because they need to cater to a variety of queries. E.g., MRR is best for navigational queries, while NDCG is best for informational queries. A key contribution of this paper is to fold multiple ranking loss functions into a multi-criteria max-margin optimization.The result is a single, robust ranking model that is close to the best accuracy of learners trained on individual criteria. In fact, experiments over the popular LETOR and TREC data sets show that, contrary to conventional wisdom, a test criterion is often not best served by training with the same individual criterion.

A comparison of three methods for downscaling daily precipitation in the Punjab region

Many downscaling techniques have been developed in the past few years for projection of station-scale hydrological variables from large-scale atmospheric variables simulated by general circulation models (GCMs) to assess the hydrological impacts of climate change. This article compares the performances of three downscaling methods, viz. conditional random field (CRF), K-nearest neighbour (KNN) and support vector machine (SVM) methods in downscaling precipitation in the Punjab region of India, belonging to the monsoon regime. The CRF model is a recently developed method for downscaling hydrological variables in a probabilistic framework, while the SVM model is a popular machine learning tool useful in terms of its ability to generalize and capture nonlinear relationships between predictors and predictand. The KNN model is an analogue-type method that queries days similar to a given feature vector from the training data and classifies future days by random sampling from a weighted set of K closest training examples. The models are applied for downscaling monsoon (June to September) daily precipitation at six locations in Punjab. Model performances with respect to reproduction of various statistics such as dry and wet spell length distributions, daily rainfall distribution, and intersite correlations are examined. It is found that the CRF and KNN models perform slightly better than the SVM model in reproducing most daily rainfall statistics. These models are then used to project future precipitation at the six locations. Output from the Canadian global climate model (CGCM3) GCM for three scenarios, viz. A1B, A2, and B1 is used for projection of future precipitation. The projections show a change in probability density functions of daily rainfall amount and changes in the wet and dry spell distributions of daily precipitation. Copyright (C) 2011 John Wiley & Sons, Ltd.

Performance Modeling based on Multidimensional Surface Learning for Performance Predictions of Parallel Applications in Non-Dedicated Environments

Modeling the performance behavior of parallel applications to predict the execution times of the applications for larger problem sizes and number of processors has been an active area of research for several years. The existing curve fitting strategies for performance modeling utilize data from experiments that are conducted under uniform loading conditions. Hence the accuracy of these models degrade when the load conditions on the machines and network change. In this paper, we analyze a curve fitting model that attempts to predict execution times for any load conditions that may exist on the systems during application execution. Based on the experiments conducted with the model for a parallel eigenvalue problem, we propose a multi-dimensional curve-fitting model based on rational polynomials for performance predictions of parallel applications in non-dedicated environments. We used the rational polynomial based model to predict execution times for 2 other parallel applications on systems with large load dynamics. In all the cases, the model gave good predictions of execution times with average percentage prediction errors of less than 20%