Chance constrained uncertain classification via robust optimization

This paper studies the problem of constructing robust classifiers when the training is plagued with uncertainty. The problem is posed as a Chance-Constrained Program (CCP) which ensures that the uncertain data points are classified correctly with high probability. Unfortunately such a CCP turns out to be intractable. The key novelty is in employing Bernstein bounding schemes to relax the CCP as a convex second order cone program whose solution is guaranteed to satisfy the probabilistic constraint. Prior to this work, only the Chebyshev based relaxations were exploited in learning algorithms. Bernstein bounds employ richer partial information and hence can be far less conservative than Chebyshev bounds. Due to this efficient modeling of uncertainty, the resulting classifiers achieve higher classification margins and hence better generalization. Methodologies for classifying uncertain test data points and error measures for evaluating classifiers robust to uncertain data are discussed. Experimental results on synthetic and real-world datasets show that the proposed classifiers are better equipped to handle data uncertainty and outperform state-of-the-art in many cases.

Variable Sparsity Kernel Learning

This paper(1) presents novel algorithms and applications for a particular class of mixed-norm regularization based Multiple Kernel Learning (MKL) formulations. The formulations assume that the given kernels are grouped and employ l(1) norm regularization for promoting sparsity within RKHS norms of each group and l(s), s >= 2 norm regularization for promoting non-sparse combinations across groups. Various sparsity levels in combining the kernels can be achieved by varying the grouping of kernels-hence we name the formulations as Variable Sparsity Kernel Learning (VSKL) formulations. While previous attempts have a non-convex formulation, here we present a convex formulation which admits efficient Mirror-Descent (MD) based solving techniques. The proposed MD based algorithm optimizes over product of simplices and has a computational complexity of O (m(2)n(tot) log n(max)/epsilon(2)) where m is no. training data points, n(max), n(tot) are the maximum no. kernels in any group, total no. kernels respectively and epsilon is the error in approximating the objective. A detailed proof of convergence of the algorithm is also presented. Experimental results show that the VSKL formulations are well-suited for multi-modal learning tasks like object categorization. Results also show that the MD based algorithm outperforms state-of-the-art MKL solvers in terms of computational efficiency.

Robust Formulations for Handling Uncertainty in Kernel Matrices

We study the problem of uncertainty in the entries of the Kernel matrix, arising in SVM formulation. Using Chance Constraint Programming and a novel large deviation inequality we derive a formulation which is robust to such noise. The resulting formulation applies when the noise is Gaussian, or has finite support. The formulation in general is non-convex, but in several cases of interest it reduces to a convex program. The problem of uncertainty in kernel matrix is motivated from the real world problem of classifying proteins when the structures are provided with some uncertainty. The formulation derived here naturally incorporates such uncertainty in a principled manner leading to significant improvements over the state of the art. 1.

Efficient algorithms for learning kernels from multiple similarity matrices with general convex loss functions

In this paper we consider the problem of learning an n × n kernel matrix from m(1) similarity matrices under general convex loss. Past research have extensively studied the m = 1 case and have derived several algorithms which require sophisticated techniques like ACCP, SOCP, etc. The existing algorithms do not apply if one uses arbitrary losses and often can not handle m > 1 case. We present several provably convergent iterative algorithms, where each iteration requires either an SVM or a Multiple Kernel Learning (MKL) solver for m > 1 case. One of the major contributions of the paper is to extend the well knownMirror Descent(MD) framework to handle Cartesian product of psd matrices. This novel extension leads to an algorithm, called EMKL, which solves the problem in O(m2 log n 2) iterations; in each iteration one solves an MKL involving m kernels and m eigen-decomposition of n × n matrices. By suitably defining a restriction on the objective function, a faster version of EMKL is proposed, called REKL,which avoids the eigen-decomposition. An alternative to both EMKL and REKL is also suggested which requires only an SVMsolver. Experimental results on real world protein data set involving several similarity matrices illustrate the efficacy of the proposed algorithms.

Precoder Designs for MIMO Broadcast Channels with Imperfect CSI

[1] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.Cambridge University Press, 2006. [2] H. Bolcskei, D. Gesbert, C. B. Papadias, and A.-J. van der Veen, Spacetime Wireless Systems: From Array Processing to MIMO Communications.Cambridge University Press, 2006. [3] Q. H. Spencer, C. B. Peel, A. L. Swindlehurst, and M. Haardt, “An introduction to the multiuser MIMO downlink,” IEEE Commun. Mag.,vol. 42, pp. 60–67, Oct. 2004. [4] K. Kusume, M. Joham,W. Utschick, and G. Bauch, “Efficient tomlinsonharashima precoding for spatial multiplexing on flat MIMO channel,”in Proc. IEEE ICC’2005, May 2005, pp. 2021–2025. [5] R. Fischer, C. Windpassinger, A. Lampe, and J. Huber, “MIMO precoding for decentralized receivers,” in Proc. IEEE ISIT’2002, 2002, p.496. [6] M. Schubert and H. Boche, “Iterative multiuser uplink and downlink beamforming under SINR constraints,” IEEE Trans. Signal Process.,vol. 53, pp. 2324–2334, Jul. 2005. [7] ——, “Solution of multiuser downlink beamforming problem with individual SINR constraints,” IEEE Trans. Veh. Technol., vol. 53, pp.18–28, Jan. 2004. [8] A. Wiesel, Y. C. Eldar, and Shamai, “Linear precoder via conic optimization for fixed MIMO receivers,” IEEE Trans. Signal Process., vol. 52,pp. 161–176, Jan. 2006. [9] N. Jindal, “MIMO broadcast channels with finite rate feed-back,” in Proc. IEEE GLOBECOM’2005, Nov. 2005. [10] R. Hunger, F. Dietrich, M. Joham, and W. Utschick, “Robust transmit zero-forcing filters,” in Proc. ITG Workshop on Smart Antennas, Munich,Mar. 2004, pp. 130–137. [11] M. B. Shenouda and T. N. Davidson, “Linear matrix inequality formulations of robust QoS precoding for broadcast channels,” in Proc.CCECE’2007, Apr. 2007, pp. 324–328. [12] M. Payaro, A. Pascual-Iserte, and M. A. Lagunas, “Robust power allocation designs for multiuser and multiantenna downlink communication systems through convex optimization,” IEEE J. Sel. Areas Commun.,vol. 25, pp. 1392–1401, Sep. 2007. [13] M. Biguesh, S. Shahbazpanahi, and A. B. Gershman, “Robust downlink power control in wireless cellular systems,” EURASIP Jl. Wireless Commun. Networking, vol. 2, pp. 261–272, 2004. [14] B. Bandemer, M. Haardt, and S. Visuri, “Liner MMSE multi-user MIMO downlink precoding for users with multple antennas,” in Proc.PIMRC’06, Sep. 2006, pp. 1–5. [15] J. Zhang, Y. Wu, S. Zhou, and J. Wang, “Joint linear transmitter and receiver design for the downlink of multiuser MIMO systems,” IEEE Commun. Lett., vol. 9, pp. 991–993, Nov. 2005. [16] S. Shi, M. Schubert, and H. Boche, “Downlink MMSE transceiver optimization for multiuser MIMO systems: Duality and sum-mse minimization,”IEEE Trans. Signal Process., vol. 55, pp. 5436–5446, Nov.2007. [17] A. Mezghani, M. Joham, R. Hunger, and W. Utschick, “Transceiver design for multi-user MIMO systems,” in Proc. WSA 2006, Mar. 2006. [18] R. Doostnejad, T. J. Lim, and E. Sousa, “Joint precoding and beamforming design for the downlink in a multiuser MIMO system,” in Proc.WiMob’2005, Aug. 2005, pp. 153–159. [19] N. Vucic, H. Boche, and S. Shi, “Robust transceiver optimization in downlink multiuser MIMO systems with channel uncertainty,” in Proc.IEEE ICC’2008, Beijing, China, May 2008. [20] A. Ben-Tal and A. Nemirovsky, “Selected topics in robust optimization,”Math. Program., vol. 112, pp. 125–158, Feb. 2007. [21] D. Bertsimas and M. Sim, “Tractable approximations to robust conic optimization problems,” Math. Program., vol. 107, pp. 5–36, Jun. 2006. [22] P. Ubaidulla and A. Chockalingam, “Robust Transceiver Design for Multiuser MIMO Downlink,” in Proc. IEEE Globecom’2008, New Orleans, USA, Dec. 2008, to appear. [23] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge University Press, 2004. [24] G. H. Golub and C. F. V. Loan, Matrix Computations. The John Hopkins University Press, 1996.

Aromatic, microporous polymer networks with high surface area generated in Friedel-Crafts-type polycondensations

A series of novel, microporous polymer networks (MPNs) have been generated in a simple, acid catalysed Friedel-Crafts-type self-condensation of A(2)B(2)- and A(2)B(4)-type fluorenone monomers. Two A2B4-type monomers with 2,7-bis(N, N-diphenylamino) A or 2,7-bis [4-(N, N-diphenylamino) phenyl] D substitution of the fluorenone cores lead to MPNs with high S(BET) surface areas of up to 1400 m(2) g(-1). Two MPNs made of binary monomer mixtures showed the highest Brunauer-Emmett-Teller (BET) surface areas S(BET) of our series (SBET of up to 1800 m(2) g(-1)) after washing the powdery samples with supercritical carbon dioxide. Total pore volumes of up to 1.6 cm(3) g(-1) have been detected. It is observed that the substitution pattern of the monomers is strongly influencing the resulting physicochemical properties of the microporous polymer networks (MPNs).

Coupled phonons, magnetic excitations, and ferroelectricity in AlFeO3: Raman and first-principles studies

We determine the nature of coupled phonons and magnetic excitations in AlFeO3 using inelastic light scattering from 5 to 315 K covering a spectral range from 100 to 2200 cm(-1) and complementary first-principles density functional theory-based calculations. A strong spin-phonon coupling and magnetic ordering-induced phonon renormalization are evident in (1) anomalous temperature dependence of many modes with frequencies below 850 cm(-1), particularly near the magnetic transition temperature T-c approximate to 250 K, and (2) distinct changes in band positions of high-frequency Raman bands between 1100 and 1800 cm(-1); in particular, a broad mode near 1250 cm(-1) appears only below T-c, attributed to the two-magnon Raman scattering. We also observe weak anomalies in the mode frequencies similar to 100 K due to a magnetically driven ferroelectric phase transition. Understanding of these experimental observations has been possible on the basis of first-principles calculations of the phonons' spectrum and their coupling with spins.

Si isotope variability in Proterozoic cherts

We report Si-isotopic compositions of 75 sedimentologically and petrographically characterized chert samples with ages ranging from similar to 2600 to 750 Ma using multi-collector inductively coupled plasma mass spectrometry. delta Si-30 values of the cherts analyzed in this study show a similar to 7 parts per thousand range, from -4.29 to +2.85. This variability can be explained in part by (1) simple mixing of silica derived from continental (higher delta Si-30) and hydrothermal (lower delta Si-30) sources, (2) multiple mechanisms of silica precipitation and (3) Rayleigh-type fractionations within pore waters of individual basins. We observe similar to 3 parts per thousand variation in peritidal cherts from a single Neoproterozoic sedimentary basin (Spitsbergen). This variation can be explained by Rayleigh-type fractionation during precipitation from silica-saturated porewaters. In some samples, post-dissolution and reprecipitation of silica could have added to this effect. Our data also indicate that peritidal cherts are enriched in the heavier isotopes of Si whereas basinal cherts associated with banded iron formations (BIF) show lower delta Si-30. This difference could partly be due to Si being derived from hydrothermal sources in BIFs. We postulate that the difference in delta Si-30 between non-BIF and BIF cherts is consistent with the contrasting genesis of these deposits. Low delta Si-30 in BIF is consistent with laboratory experiments showing that silica adsorbed onto Fe-hydroxide particles preferentially incorporates lighter Si isotopes. Despite large intrabasinal variation and environmental differences, the data show a clear pattern of secular variation. Low delta Si-30 in Archean cherts is consistent with a dominantly hydrothermal source of silica to the oceans at that time. The monotonically increasing delta Si-30 from 3.8 to 1.5 Ga appears to reflect a general increase in continental versus hydrothermal sources of Si in seawater, as well as the preferential removal of lighter Si isotopes during silica precipitation in iron-associated cherts from silica-saturated seawater. The highest delta Si-30 values are observed in 1.5 Ga peritidal cherts; in part, these enriched values could reflect increasing sequestration of light silica during soil-forming processes, thus, delivering relatively heavy dissolved silica to the oceans from continental sources. The causes behind the reversal in trend towards lower delta Si-30 in cherts younger than 1.5 Ga old are less clear. Cherts deposited 1800-1900 Ma are especially low delta Si-30, a possible indication of transiently strong hydrothermal input at this time. (C) 2012 Elsevier Ltd. All rights reserved.

Power Controlled Reverse Channel Training Achieves an Infinite Diversity Order in a TDD-SIMO System with Perfect CSIR

In this letter, we analyze the Diversity Multiplexinggain Tradeoff (DMT) performance of a training-based reciprocal Single Input Multiple Output (SIMO) system. Assuming Channel State Information (CSI) is available at the Receiver (CSIR), we propose a channel-dependent power-controlled Reverse Channel Training (RCT) scheme that enables the transmitter to directly estimate the power control parameter to be used for the forwardlink data transmission. We show that, with an RCT power of (P) over bar (gamma), gamma > 0 and a forward data transmission power of (P) over bar, our proposed scheme achieves an infinite diversity order for 0 <= g(m) < L-c-L-B,L-tau/L-c min(gamma, 1) and r > 2, where g(m) is the multiplexing gain, L-c is the channel coherence time, L-B,L-tau is the RCT duration and r is the number of receive antennas. We also derive an upper bound on the outage probability and show that it goes to zero asymptotically as exp(-(P) over bar (E)), where E (sic) (gamma - g(m)L(c)/L-c-L-B,L-tau), at high (P) over bar. Thus, the proposed scheme achieves a significantly better DMT performance compared to the finite diversity order achieved by channel-agnostic, fixed-power RCT schemes.

Efficient Methods for Robust Classification Under Uncertainty in Kernel Matrices

In this paper we study the problem of designing SVM classifiers when the kernel matrix, K, is affected by uncertainty. Specifically K is modeled as a positive affine combination of given positive semi definite kernels, with the coefficients ranging in a norm-bounded uncertainty set. We treat the problem using the Robust Optimization methodology. This reduces the uncertain SVM problem into a deterministic conic quadratic problem which can be solved in principle by a polynomial time Interior Point (IP) algorithm. However, for large-scale classification problems, IP methods become intractable and one has to resort to first-order gradient type methods. The strategy we use here is to reformulate the robust counterpart of the uncertain SVM problem as a saddle point problem and employ a special gradient scheme which works directly on the convex-concave saddle function. The algorithm is a simplified version of a general scheme due to Juditski and Nemirovski (2011). It achieves an O(1/T-2) reduction of the initial error after T iterations. A comprehensive empirical study on both synthetic data and real-world protein structure data sets show that the proposed formulations achieve the desired robustness, and the saddle point based algorithm outperforms the IP method significantly.

Robust formulations for clustering-based large-scale classification

Chebyshev-inequality-based convex relaxations of Chance-Constrained Programs (CCPs) are shown to be useful for learning classifiers on massive datasets. In particular, an algorithm that integrates efficient clustering procedures and CCP approaches for computing classifiers on large datasets is proposed. The key idea is to identify high density regions or clusters from individual class conditional densities and then use a CCP formulation to learn a classifier on the clusters. The CCP formulation ensures that most of the data points in a cluster are correctly classified by employing a Chebyshev-inequality-based convex relaxation. This relaxation is heavily dependent on the second-order statistics. However, this formulation and in general such relaxations that depend on the second-order moments are susceptible to moment estimation errors. One of the contributions of the paper is to propose several formulations that are robust to such errors. In particular a generic way of making such formulations robust to moment estimation errors is illustrated using two novel confidence sets. An important contribution is to show that when either of the confidence sets is employed, for the special case of a spherical normal distribution of clusters, the robust variant of the formulation can be posed as a second-order cone program. Empirical results show that the robust formulations achieve accuracies comparable to that with true moments, even when moment estimates are erroneous. Results also illustrate the benefits of employing the proposed methodology for robust classification of large-scale datasets.

Non-polio enteroviruses and their association with acute diarrhea in children in India

A causative agent in approximately 40% of diarrhea] cases. still remains unidentified. Though many enteroviruses (EVs) are transmitted through fecal-oral route and replicate in the intestinal cells, their association with acute diarrhea has not so far been recognized due to lack of detailed epidemiological investigations. This long-term, detailed molecular epidemiological study aims to conclusively determine the association of non-polio enteroviruses (NPEVs) with acute diarrhea in comaparison with rotavirus (RV) in children. Diarrheal stool specimens from 2161 children aged 0-2 years and 169 children between 2 and 9 years, and 1800 normal stool samples from age-matched healthy children between 0 and 9 years were examined during 2008-2012 for enterovirus (oral polio vaccine strains (OPVs) and NPEVs). Enterovirus serotypes were identified by complete VP1 gene sequence analysis. Enterovirus and rotavirus were detected in 19.01% (380/2330) and 13.82% (322/2330) diarrheal stools. During the study period, annual prevalence of EV- and RV-associated diarrhea ranged between 8% and 22%, but with contrasting seasonal prevalence with RV predominating during winter months and NPEV prevailing in other seasons. NPEVs are associated with epidemics-like outbreaks during which they are detected in up to 50% of diarrheic children, and in non-epidemic seasons in 0-10% of the patients. After subtraction of OPV-positive diarrheal cases (1.81%), while NPEVs are associated with about 17% of acute diarrhea, about 6% of healthy children showed asymptomatic NPEV excretion. Of 37 NPEV serotypes detected in diarrheal children, seven echovirus types 1, 7, 11, 13, 14, 30 and 33 are frequently observed, with Ell being more prevalent followed by E30. In conclusion, NPEVs are significantly associated with acute diarrhea, and NPEVs and rotavirus exhibit contrasting seasonal predominance. This study signifies the need for a new direction of research on enteroviruses involving systematic analysis of their contribution to diarrheal burden. (C) 2013 Elsevier B.V. All rights reserved.

Solvatochromism of 9,10-phenanthrenequinone: An electronic and resonance Raman spectroscopic study

Solvent effects play a vital role in various chemical, physical, and biological processes. To gain a fundamental understanding of the solute-solvent interactions and their implications on the energy level re-ordering and structure, UV-VIS absorption, resonance Raman spectroscopic, and density functional theory calculation studies on 9,10-phenanthrenequinone (PQ) in different solvents of diverse solvent polarity has been carried out. The solvatochromic analysis of the absorption spectra of PQ in protic dipolar solvents suggests that the longest (1n-pi(1)*; S-1 state) and the shorter (1 pi-pi(1)*; S-2 state) wavelength band undergoes a hypsochromic and bathochromic shift due to intermolecular hydrogen bond weakening and strengthening, respectively. It also indicates that hydrogen bonding plays a major role in the differential solvation of the S-2 state relative to the ground state. Raman excitation profiles of PQ (400-1800 cm(-1)) in various solvents followed their corresponding absorption spectra therefore the enhancements on resonant excitation are from single-state rather than mixed states. The hyperchromism of the longer wavelength band is attributed to intensity borrowing from the nearby allowed electronic transition through vibronic coupling. Computational calculation with C-2 nu symmetry constraint on the S-2 state resulted in an imaginary frequency along the low-frequency out-of-plane torsional modes involving the C=O site and therefore, we hypothesize that this mode could be involved in the vibronic coupling. (C) 2015 AIP Publishing LLC.

Path Following Using Trajectory Shaping Guidance

Unmanned vehicle path following by pursuing a virtual target moving along the path is considered. Limitations for pure pursuit guidance are analyzed while following the virtual target on curved paths. Trajectory shaping guidance is proposed as an alternate guidance scheme for a general curvature path. It is proven that under certain tenable assumptions trajectory shaping guidance yields an identical path as that of the virtual target. By linear analysis it is shown that the convergence to the path for trajectory shaping guidance is twice as fast as pure pursuit. Simulations highlight significant improvement in position errors by using trajectory shaping guidance. Comparative simulation studies comply with analytic findings and present better performance as compared with pure pursuit and a nonlinear guidance methodology from the literature. Experimental validation supports the analytic and simulations studies as the guidance laws are implemented on a radio-controlled car in a laboratory environment.

BIOMASS GASIFICATION TO SUBSTITUTE FOSSIL FUEL IN FOOD INDUSTRY

Food industries like biscuit and confectionary use significant amount of fossil fuel for thermal energy. Biscuit manufacturing in India is carried out both by organized and unorganized sector. The ratio of organized to unorganized sector is 60 : 40 (1). The total biscuit manufacturing in the organized sector India in 2008 was about 1.7 million metric tons (1). Accounting for the unorganized sector in India, the total biscuit manufacturing would have been about 2.9 million metric tons/annum. A typical biscuit baking is carried in a long tunnel kiln with varying temperature in different zones. Generally diesel is used to provide the necessary heat energy for the baking purpose, with temperature ranging from 190 C in the drying zone to about 300 C in the baking area and has to maintain in the temperature range of +/- 5 C. Typical oil consumption is about 40 litres per ton of biscuit production. The paper discusses the experience in substituting about 120 lts per hour kiln for manufacturing about 70 tons of biscuit daily. The system configuration consists of a 500 kg/hr gasification system comprising of a reactor, multicyclone, water scrubbers, and two blowers for maintaining the constant gas pressure in the header before the burners. Cold producer gas is piped to the oven located about 200 meters away from the gasifier. Fuel used in the gasification system is coconut shells. All the control system existing on the diesel burner has been suitably adapted for producer gas operation to maintain the total flow, A/F control so as to maintain the temperature. A total of 7 burners are used in different zones. Over 17000 hour of operation has resulted in replacing over 1800 tons of diesel over the last 30 months. The system operates for over 6 days a week with average operational hours of 160. It has been found that on an average 3.5 kg of biomass has replaced one liter of diesel.