Cluster based training for scaling non-linear support vector machines

Support Vector Machines(SVMs) are hyperplane classifiers defined in a kernel induced feature space. The data size dependent training time complexity of SVMs usually prohibits its use in applications involving more than a few thousands of data points. In this paper we propose a novel kernel based incremental data clustering approach and its use for scaling Non-linear Support Vector Machines to handle large data sets. The clustering method introduced can find cluster abstractions of the training data in a kernel induced feature space. These cluster abstractions are then used for selective sampling based training of Support Vector Machines to reduce the training time without compromising the generalization performance. Experiments done with real world datasets show that this approach gives good generalization performance at reasonable computational expense.

Four group decodable differential scaled unitary linear space-time codes

Differential Unitary Space-Time Block codes (STBCs) offer a means to communicate on the Multiple Input Multiple Output (MIMO) channel without the need for channel knowledge at both the transmitter and the receiver. Recently Yuen-Guan-Tjhung have proposed Single-Symbol-Decodable Differential Space-Time Modulation based on Quasi-Orthogonal Designs (QODs) by replacing the original unitary criterion by a scaled unitary criterion. These codes were also shown to perform better than differential unitary STBCs from Orthogonal Designs (ODs). However the rate (as measured in complex symbols per channel use) of the codes of Yuen-Guan-Tjhung decay as the number of transmit antennas increase. In this paper, a new class of differential scaled unitary STBCs for all even number of transmit antennas is proposed. These codes have a rate of 1 complex symbols per channel use, achieve full diversity and moreover they are four-group decodable, i.e., the set of real symbols can be partitioned into four groups and decoding can be done for the symbols in each group separately. Explicit construction of multidimensional signal sets that yield full diversity for this new class of codes is also given.

Bond-order wave phase, spin solitons, and thermodynamics of a frustrated linear spin-1/2 Heisenberg antiferromagnet

The linear spin-1/2 Heisenberg antiferromagnet with exchanges J(1) and J(2) between first and second neighbors has a bond-order wave (BOW) phase that starts at the fluid-dimer transition at J(2)/J(1)=0.2411 and is particularly simple at J(2)/J(1)=1/2. The BOW phase has a doubly degenerate singlet ground state, broken inversion symmetry, and a finite-energy gap E-m to the lowest-triplet state. The interval 0.4 < J(2)/J(1) < 1.0 has large E-m and small finite-size corrections. Exact solutions are presented up to N = 28 spins with either periodic or open boundary conditions and for thermodynamics up to N = 18. The elementary excitations of the BOW phase with large E-m are topological spin-1/2 solitons that separate BOWs with opposite phase in a regular array of spins. The molar spin susceptibility chi(M)(T) is exponentially small for T << E-m and increases nearly linearly with T to a broad maximum. J(1) and J(2) spin chains approximate the magnetic properties of the BOW phase of Hubbard-type models and provide a starting point for modeling alkali-tetracyanoquinodimethane salts.

Novel window functions for digital filters

Window technique is one of the simplest methods to design Finite Impulse Response (FIR) filters. It uses special functions to truncate an infinite sequence to a finite one. In this paper, we propose window techniques based on integer sequences. The striking feature of the proposed work is that it overcomes all the problems posed by floating point numbers and inaccuracy, as the sequences are made of only integers. Some of these integer window sequences, yield sharp transition, while some of them result in zero ripple in passband and stopband.

On the PAPR of cosets of linear codes

Employing an error control code is one of the techniques to reduce the Peak-to-Average Power Ratio (PAPR) in a Orthogonal Frequency Division Multiplexing system, a well known class of such codes being the cosets of Reed-Muller codes. In this paper, we consider the class of such coset-codes of arbitrary linear codes and present a method of doubling the size of such a code without increasing the PAPR, by combining two such binary coset-codes. We identify the conditions under which we can employ this doubling more than once with no marginal increase in the PAPR value. Given a PAPR and length, our method has enabled to get the best coset-code (in terms of the size). Also, we show that the PAPR information of the coset-codes of the extended codes is obtainable from the PAPR of the corresponding coset-codes of the parent code. We have also shown a special type of lengthening is useful in PAPR studies.

Optimal resource allocation in conflicts with the Lanchester linear law (2,1) model

This paper presents a detailed analysis of a model for military conflicts where the defending forces have to determine an optimal partitioning of available resources to counter attacks from an adversary in two different fronts in an area fire situation. Lanchester linear law attrition model is used to develop the dynamical equations governing the variation in force strength. Here we address a static resource allocation problem namely, Time-Zero-Allocation (TZA) where the resource allocation is done only at the initial time. Numerical examples are given to support the analytical results.

Multi-Tone Testing of Linear and Nonlinear Analog Circuits using Polynomial Coefficients

A method of testing for parametric faults of analog circuits based on a polynomial representation of fault-free function of the circuit is presented. The response of the circuit under test (CUT) is estimated as a polynomial in the applied input voltage at relevant frequencies in addition to DC. Classification or Cur is based on a comparison of the estimated polynomial coefficients with those of the fault free circuit. This testing method requires no design for test hardware as might be added to the circuit fly some other methods. The proposed method is illustrated for a benchmark elliptic filter. It is shown to uncover several parametric faults causing deviations as small as 5% from the nominal values.

Structural Studies on the Second Mycobacterium smegmatis Dps: Invariant and Variable Features of Structure, Assembly and Function

A second DNA binding protein from stationary-phase cells of Mycobacterium smegmatis (MsDps2) has been identified from the bacterial genome. It was cloned, expressed and characterised and its crystal structure was determined. The core dodecameric structure of MsDps2 is the same as that of the Dps from the organism described earlier (MsDps1). However, MsDps2 possesses a long N-terminal tail instead of the C-terminal tail in MsDps1. This tail appears to be involved in DNA binding. It is also intimately involved in stabilizing the dodecamer. Partly on account of this factor, MsDps2 assembles straightway into the dodecamer, while MsDps1 does so on incubation after going through an intermediate trimeric stage. The ferroxidation centre is similar in the two proteins, while the pores leading to it exhibit some difference. The mode of sequestration of DNA in the crystalline array of molecules, as evidenced by the crystal structures, appears to be different in MsDps1 and MsDps2, highlighting the variability in the mode of Dps–DNA complexation. A sequence search led to the identification of 300 Dps molecules in bacteria with known genome sequences. Fifty bacteria contain two or more types of Dps molecules each, while 195 contain only one type. Some bacteria, notably some pathogenic ones, do not contain Dps. A sequence signature for Dps could also be derived from the analysis.

Software Pipelined Execution of Stream Programs on GPUs

The StreamIt programming model has been proposed to exploit parallelism in streaming applications on general purpose multi-core architectures. This model allows programmers to specify the structure of a program as a set of filters that act upon data, and a set of communication channels between them. The StreamIt graphs describe task, data and pipeline parallelism which can be exploited on modern Graphics Processing Units (GPUs), as they support abundant parallelism in hardware. In this paper, we describe the challenges in mapping StreamIt to GPUs and propose an efficient technique to software pipeline the execution of stream programs on GPUs. We formulate this problem - both scheduling and assignment of filters to processors - as an efficient Integer Linear Program (ILP), which is then solved using ILP solvers. We also describe a novel buffer layout technique for GPUs which facilitates exploiting the high memory bandwidth available in GPUs. The proposed scheduling utilizes both the scalar units in GPU, to exploit data parallelism, and multiprocessors, to exploit task and pipelin parallelism. Further it takes into consideration the synchronization and bandwidth limitations of GPUs, and yields speedups between 1.87X and 36.83X over a single threaded CPU.

Enantioselective syntheses of diquinane and (cis, anti, cis)-linear triquinanes

The enantioselective syntheses of diquinane and cis, anti, cis-linear triquinanes, starting from the readily available (S)-campholenaldehyde, employing an intramolecular rhodium carbenoid CH insertion reaction, are described. (C) 2010 Elsevier Ltd. All rights reserved.

Fluorescent resonant excitation energy transfer in linear polyenes

We have studied the dynamics of excitation transfer between two conjugated polyene molecules whose intermolecular separation is comparable to the molecular dimensions. We have employed a correlated electron model that includes both the charge-charge, charge-bond, and bond-bond intermolecular electron repulsion integrals. We have shown that the excitation transfer rate varies as inverse square of donor-acceptor separation R-2 rather than as R-6, suggested by the Foumlrster type of dipolar approximation. Our time-evolution study alsom shows that the orientational dependence on excitation transfer at a fixed short donor-acceptor separation cannot be explained by Foumlrster type of dipolar approximation beyond a certain orientational angle of rotation of an acceptor polyene with respect to the donor polyene. The actual excitation transfer rate beyond a certain orientational angle is faster than the Foumlrster type of dipolar approximation rate. We have also studied the excitation transfer process in a pair of push-pull polyenes for different push-pull strengths. We have seen that, depending on the push-pull strength, excitation transfer could occur to other dipole coupled states. Our study also allows for the excitation energy transfer to optically dark states which are excluded by Foumlrster theory since the one-photon transition intensity to these states (from the ground state) is zero.

Linear transconductor with flipped voltage follower in 130 nm CMOS

This paper presents a modified design method for linear transconductor circuit in 130 nm CMOS technology to improve linearity, robustness against process induced threshold voltage variability and reduce harmonic distortion. Source follower in the adaptively biased differential pair (ABDP) linear transconductor circuit is replaced with flipped voltage follower to improve the efficiency of the tail current source, which is connected to a conventional differential pair. The simulation results show the performance of the modified circuit also has better speed, noise performance and common mode rejection ratio compared to the ABDP circuit.

Generalised regression estimation for domain class frequencies

This study examines the properties of Generalised Regression (GREG) estimators for domain class frequencies and proportions. The family of GREG estimators forms the class of design-based model-assisted estimators. All GREG estimators utilise auxiliary information via modelling. The classic GREG estimator with a linear fixed effects assisting model (GREG-lin) is one example. But when estimating class frequencies, the study variable is binary or polytomous. Therefore logistic-type assisting models (e.g. logistic or probit model) should be preferred over the linear one. However, other GREG estimators than GREG-lin are rarely used, and knowledge about their properties is limited. This study examines the properties of L-GREG estimators, which are GREG estimators with fixed-effects logistic-type models. Three research questions are addressed. First, I study whether and when L-GREG estimators are more accurate than GREG-lin. Theoretical results and Monte Carlo experiments which cover both equal and unequal probability sampling designs and a wide variety of model formulations show that in standard situations, the difference between L-GREG and GREG-lin is small. But in the case of a strong assisting model, two interesting situations arise: if the domain sample size is reasonably large, L-GREG is more accurate than GREG-lin, and if the domain sample size is very small, estimation of assisting model parameters may be inaccurate, resulting in bias for L-GREG. Second, I study variance estimation for the L-GREG estimators. The standard variance estimator (S) for all GREG estimators resembles the Sen-Yates-Grundy variance estimator, but it is a double sum of prediction errors, not of the observed values of the study variable. Monte Carlo experiments show that S underestimates the variance of L-GREG especially if the domain sample size is minor, or if the assisting model is strong. Third, since the standard variance estimator S often fails for the L-GREG estimators, I propose a new augmented variance estimator (A). The difference between S and the new estimator A is that the latter takes into account the difference between the sample fit model and the census fit model. In Monte Carlo experiments, the new estimator A outperformed the standard estimator S in terms of bias, root mean square error and coverage rate. Thus the new estimator provides a good alternative to the standard estimator.