Maximum Margin Classifiers with Specified False Positive and False Negative Error Rates

This paper addresses the problem of maximum margin classification given the moments of class conditional densities and the false positive and false negative error rates. Using Chebyshev inequalities, the problem can be posed as a second order cone programming problem. The dual of the formulation leads to a geometric optimization problem, that of computing the distance between two ellipsoids, which is solved by an iterative algorithm. The formulation is extended to non-linear classifiers using kernel methods. The resultant classifiers are applied to the case of classification of unbalanced datasets with asymmetric costs for misclassification. Experimental results on benchmark datasets show the efficacy of the proposed method.

The generalized Onsager model for the secondary flow in a high-speed rotating cylinder

The generalizations of the Onsager model for the radial boundary layer and the Carrier-Maslen model for the end-cap axial boundary layer in a high-speed rotating cylinder are formulated for studying the secondary gas flow due to wall heating and due to insertion of mass, momentum and energy into the cylinder. The generalizations have wider applicability than the original Onsager and Carrier-Maslen models, because they are not restricted to the limit A >> 1, though they are restricted to the limit R e >> 1 and a high-aspect-ratio cylinder whose length/diameter ratio is large. Here, the stratification parameter A = root m Omega(2)R(2)/2k(B)T). This parameter A is the ratio of the peripheral speed, Omega R, to the most probable molecular speed, root 2k(B)T/m, the Reynolds number Re = rho w Omega R(2)/mu, where m is the molecular mass, Omega and R are the rotational speed and radius of the cylinder, k(B) is the Boltzmann constant, T is the gas temperature, rho(w) is the gas density at wall, and mu is the gas viscosity. In the case of wall forcing, analytical solutions are obtained for the sixth-order generalized Onsager equations for the master potential, and for the fourth-order generalized Carrier-Maslen equation for the velocity potential. For the case of mass/momentum/energy insertion into the flow, the separation-of-variables procedure is used, and the appropriate homogeneous boundary conditions are specified so that the linear operators in the axial and radial directions are self-adjoint. The discrete eigenvalues and eigenfunctions of the linear operators (sixth-order and second-order in the radial and axial directions for the Onsager equation, and fourth-order and second-order in the axial and radial directions for the Carrier-Maslen equation) are determined. These solutions are compared with direct simulation Monte Carlo (DSMC) simulations. The comparison reveals that the boundary conditions in the simulations and analysis have to be matched with care. The commonly used `diffuse reflection' boundary conditions at solid walls in DSMC simulations result in a non-zero slip velocity as well as a `temperature slip' (gas temperature at the wall is different from wall temperature). These have to be incorporated in the analysis in order to make quantitative predictions. In the case of mass/momentum/energy sources within the flow, it is necessary to ensure that the homogeneous boundary conditions are accurately satisfied in the simulations. When these precautions are taken, there is excellent agreement between analysis and simulations, to within 10 %, even when the stratification parameter is as low as 0.707, the Reynolds number is as low as 100 and the aspect ratio (length/diameter) of the cylinder is as low as 2, and the secondary flow velocity is as high as 0.2 times the maximum base flow velocity. The predictions of the generalized models are also significantly better than those of the original Onsager and Carrier-Maslen models, which are restricted to thin boundary layers in the limit of high stratification parameter.

Maximum rate of 3- and 4-real-symbol ML decodable unitary weight STBCs

It has been shown recently that the maximum rate of a 2-real-symbol (single-complex-symbol) maximum likelihood (ML) decodable, square space-time block codes (STBCs) with unitary weight matrices is 2a/2a complex symbols per channel use (cspcu) for 2a number of transmit antennas [1]. These STBCs are obtained from Unitary Weight Designs (UWDs). In this paper, we show that the maximum rates for 3- and 4-real-symbol (2-complex-symbol) ML decodable square STBCs from UWDs, for 2a transmit antennas, are 3(a-1)/2a and 4(a-1)/2a cspcu, respectively. STBCs achieving this maximum rate are constructed. A set of sufficient conditions on the signal set, required for these codes to achieve full-diversity are derived along with expressions for their coding gain.

Maximum Rate of Unitary-Weight, Single-Symbol Decodable STBCs

It is well known that the space-time block codes (STBCs) from complex orthogonal designs (CODs) are single-symbol decodable/symbol-by-symbol decodable (SSD). The weight matrices of the square CODs are all unitary and obtainable from the unitary matrix representations of Clifford Algebras when the number of transmit antennas n is a power of 2. The rate of the square CODs for n = 2(a) has been shown to be a+1/2(a) complex symbols per channel use. However, SSD codes having unitary-weight matrices need not be CODs, an example being the minimum-decoding-complexity STBCs from quasi-orthogonal designs. In this paper, an achievable upper bound on the rate of any unitary-weight SSD code is derived to be a/2(a)-1 complex symbols per channel use for 2(a) antennas, and this upper bound is larger than that of the CODs. By way of code construction, the interrelationship between the weight matrices of unitary-weight SSD codes is studied. Also, the coding gain of all unitary-weight SSD codes is proved to be the same for QAM constellations and conditions that are necessary for unitary-weight SSD codes to achieve full transmit diversity and optimum coding gain are presented.

On maximum entropy and minimum KL-divergence optimization by Grobner basis methods

In this paper we study constrained maximum entropy and minimum divergence optimization problems, in the cases where integer valued sufficient statistics exists, using tools from computational commutative algebra. We show that the estimation of parametric statistical models in this case can be transformed to solving a system of polynomial equations. We give an implicit description of maximum entropy models by embedding them in algebraic varieties for which we give a Grobner basis method to compute it. In the cases of minimum KL-divergence models we show that implicitization preserves specialization of prior distribution. This result leads us to a Grobner basis method to embed minimum KL-divergence models in algebraic varieties. (C) 2012 Elsevier Inc. All rights reserved.

Maximum weight independent sets in hole- and dart-free graphs

The Maximum Weight Independent Set (MWIS) problem on graphs with vertex weights asks for a set of pairwise nonadjacent vertices of maximum total weight. The complexity of the MWIS problem for hole-free graphs is unknown. In this paper, we first prove that the MWIS problem for (hole, dart, gem)-free graphs can be solved in O(n(3))-time. By using this result, we prove that the MWIS problem for (hole, dart)-free graphs can be solved in O(n(4))-time. Though the MWIS problem for (hole, dart, gem)-free graphs is used as a subroutine, we also give the best known time bound for the solvability of the MWIS problem in (hole, dart, gem)-free graphs. (C) 2012 Elsevier B.V. All rights reserved.

Selective excitation and detection of maximum quantum coherence of a group of scalar coupled protons in chiral molecules: an NMR experiment for enantiodiscrimination

The H-1 NMR spectroscopic discrimination of enantiomers in the solution state and the measurement of enantiomeric composition is most often hindered due to either very small chemical shift differences between the discriminated peaks or severe overlap of transitions from other chemically non-equivalent protons. In addition the use of chiral auxiliaries such as, crown ether and chiral lanthanide shift reagent may often cause enormous line broadening or give little degree of discrimination beyond the crown ether substrate ratio, hampering the discrimination. In circumventing such problems we are proposing the utilization of the difference in the additive values of all the chemical shifts of a scalar coupled spin system. The excitation and detection of appropriate highest quantum coherence yields the measurable difference in the frequencies between two transitions, one pertaining to each enantiomer in the maximum quantum dimension permitting their discrimination and the F-2 cross section at each of these frequencies yields an enantiopure spectrum. The advantage of the utility of the proposed method is demonstrated on several chiral compounds where the conventional one dimensional H-1 NMR spectra fail to differentiate the enantiomers.

Differential cellular recognition pattern to M. tuberculosis targets defined by IFN-gamma and IL-17 production in blood from TB plus patients from Honduras as compared to health care workers: TB and immune responses in patients from Honduras

Background: A better understanding of the quality of cellular immune responses directed against molecularly defined targets will guide the development of TB diagnostics and identification of molecularly defined, clinically relevant M.tb vaccine candidates. Methods: Recombinant proteins (n = 8) and peptide pools (n = 14) from M. tuberculosis (M.tb) targets were used to compare cellular immune responses defined by IFN-gamma and IL-17 production using a Whole Blood Assay (WBA) in a cohort of 148 individuals, i.e. patients with TB + (n = 38), TB- individuals with other pulmonary diseases (n = 81) and individuals exposed to TB without evidence of clinical TB (health care workers, n = 29). Results: M.tb antigens Rv2958c (glycosyltransferase), Rv2962c (mycolyltransferase), Rv1886c (Ag85B), Rv3804c (Ag85A), and the PPE family member Rv3347c were frequently recognized, defined by IFN-gamma production, in blood from healthy individuals exposed to M.tb (health care workers). A different recognition pattern was found for IL-17 production in blood from M.tb exposed individuals responding to TB10.4 (Rv0288), Ag85B (Rv1886c) and the PPE family members Rv0978c and Rv1917c. Conclusions: The pattern of immune target recognition is different in regard to IFN-gamma and IL-17 production to defined molecular M.tb targets in PBMCs from individuals frequently exposed to M.tb. The data represent the first mapping of cellular immune responses against M.tb targets in TB patients from Honduras.

Maximum likelihood algorithms for joint estimation of synchronisation impairments and channel in multiple input multiple output-orthogonal frequency division multiplexing system

Maximum likelihood (ML) algorithms, for the joint estimation of synchronisation impairments and channel in multiple input multiple output-orthogonal frequency division multiplexing (MIMO-OFDM) system, are investigated in this work. A system model that takes into account the effects of carrier frequency offset, sampling frequency offset, symbol timing error and channel impulse response is formulated. Cramer-Rao lower bounds for the estimation of continuous parameters are derived, which show the coupling effect among different impairments and the significance of the joint estimation. The authors propose an ML algorithm for the estimation of synchronisation impairments and channel together, using the grid search method. To reduce the complexity of the joint grid search in the ML algorithm, a modified ML (MML) algorithm with multiple one-dimensional searches is also proposed. Further, a stage-wise ML (SML) algorithm using existing algorithms, which estimate less number of parameters, is also proposed. Performance of the estimation algorithms is studied through numerical simulations and it is found that the proposed ML and MML algorithms exhibit better performance than SML algorithm.

Learning from positive and unlabelled examples using maximum margin clustering

Learning from Positive and Unlabelled examples (LPU) has emerged as an important problem in data mining and information retrieval applications. Existing techniques are not ideally suited for real world scenarios where the datasets are linearly inseparable, as they either build linear classifiers or the non-linear classifiers fail to achieve the desired performance. In this work, we propose to extend maximum margin clustering ideas and present an iterative procedure to design a non-linear classifier for LPU. In particular, we build a least squares support vector classifier, suitable for handling this problem due to symmetry of its loss function. Further, we present techniques for appropriately initializing the labels of unlabelled examples and for enforcing the ratio of positive to negative examples while obtaining these labels. Experiments on real-world datasets demonstrate that the non-linear classifier designed using the proposed approach gives significantly better generalization performance than the existing relevant approaches for LPU.

A Phase Transition for the Uniform Distribution in the Pattern Maximum Likelihood Problem

In this paper, we consider the setting of the pattern maximum likelihood (PML) problem studied by Orlitsky et al. We present a well-motivated heuristic algorithm for deciding the question of when the PML distribution of a given pattern is uniform. The algorithm is based on the concept of a ``uniform threshold''. This is a threshold at which the uniform distribution exhibits an interesting phase transition in the PML problem, going from being a local maximum to being a local minimum.

Generative Maximum Entropy Learning for Multiclass Classification

Maximum entropy approach to classification is very well studied in applied statistics and machine learning and almost all the methods that exists in literature are discriminative in nature. In this paper, we introduce a maximum entropy classification method with feature selection for large dimensional data such as text datasets that is generative in nature. To tackle the curse of dimensionality of large data sets, we employ conditional independence assumption (Naive Bayes) and we perform feature selection simultaneously, by enforcing a `maximum discrimination' between estimated class conditional densities. For two class problems, in the proposed method, we use Jeffreys (J) divergence to discriminate the class conditional densities. To extend our method to the multi-class case, we propose a completely new approach by considering a multi-distribution divergence: we replace Jeffreys divergence by Jensen-Shannon (JS) divergence to discriminate conditional densities of multiple classes. In order to reduce computational complexity, we employ a modified Jensen-Shannon divergence (JS(GM)), based on AM-GM inequality. We show that the resulting divergence is a natural generalization of Jeffreys divergence to a multiple distributions case. As far as the theoretical justifications are concerned we show that when one intends to select the best features in a generative maximum entropy approach, maximum discrimination using J-divergence emerges naturally in binary classification. Performance and comparative study of the proposed algorithms have been demonstrated on large dimensional text and gene expression datasets that show our methods scale up very well with large dimensional datasets.