Smoothed Functional Algorithms for Stochastic Optimization Using q-Gaussian Distributions

Smoothed functional (SF) schemes for gradient estimation are known to be efficient in stochastic optimization algorithms, especially when the objective is to improve the performance of a stochastic system However, the performance of these methods depends on several parameters, such as the choice of a suitable smoothing kernel. Different kernels have been studied in the literature, which include Gaussian, Cauchy, and uniform distributions, among others. This article studies a new class of kernels based on the q-Gaussian distribution, which has gained popularity in statistical physics over the last decade. Though the importance of this family of distributions is attributed to its ability to generalize the Gaussian distribution, we observe that this class encompasses almost all existing smoothing kernels. This motivates us to study SF schemes for gradient estimation using the q-Gaussian distribution. Using the derived gradient estimates, we propose two-timescale algorithms for optimization of a stochastic objective function in a constrained setting with a projected gradient search approach. We prove the convergence of our algorithms to the set of stationary points of an associated ODE. We also demonstrate their performance numerically through simulations on a queuing model.

Subexponential Algorithm for d-Cluster Edge Deletion: Exception or Rule?

The correlation clustering problem is a fundamental problem in both theory and practice, and it involves identifying clusters of objects in a data set based on their similarity. A traditional modeling of this question as a graph theoretic problem involves associating vertices with data points and indicating similarity by adjacency. Clusters then correspond to cliques in the graph. The resulting optimization problem, Cluster Editing (and several variants) are very well-studied algorithmically. In many situations, however, translating clusters to cliques can be somewhat restrictive. A more flexible notion would be that of a structure where the vertices are mutually ``not too far apart'', without necessarily being adjacent. One such generalization is realized by structures called s-clubs, which are graphs of diameter at most s. In this work, we study the question of finding a set of at most k edges whose removal leaves us with a graph whose components are s-clubs. Recently, it has been shown that unless Exponential Time Hypothesis fail (ETH) fails Cluster Editing (whose components are 1-clubs) does not admit sub-exponential time algorithm STACS, 2013]. That is, there is no algorithm solving the problem in time 2 degrees((k))n(O(1)). However, surprisingly they show that when the number of cliques in the output graph is restricted to d, then the problem can be solved in time O(2(O(root dk)) + m + n). We show that this sub-exponential time algorithm for the fixed number of cliques is rather an exception than a rule. Our first result shows that assuming the ETH, there is no algorithm solving the s-Club Cluster Edge Deletion problem in time 2 degrees((k))n(O(1)). We show, further, that even the problem of deleting edges to obtain a graph with d s-clubs cannot be solved in time 2 degrees((k))n(O)(1) for any fixed s, d >= 2. This is a radical contrast from the situation established for cliques, where sub-exponential algorithms are known.

Descending from infinity: Convergence of tailed distributions

We investigate the relaxation of long-tailed distributions under stochastic dynamics that do not support such tails. Linear relaxation is found to be a borderline case in which long tails are exponentially suppressed in time but not eliminated. Relaxation stronger than linear suppresses long tails immediately, but may lead to strong transient peaks in the probability distribution. We also find that a delta-function initial distribution under stronger than linear decay displays not one but two different regimes of diffusive spreading.

Bacterial growth rate and host factors as determinants of intracellular bacterial distributions in systemic Salmonella enterica infections.

Bacteria of the species Salmonella enterica cause a range of life-threatening diseases in humans and animals worldwide. The within-host quantitative, spatial, and temporal dynamics of S. enterica interactions are key to understanding how immunity acts on these infections and how bacteria evade immune surveillance. In this study, we test hypotheses generated from mathematical models of in vivo dynamics of Salmonella infections with experimental observation of bacteria at the single-cell level in infected mouse organs to improve our understanding of the dynamic interactions between host and bacterial mechanisms that determine net growth rates of S. enterica within the host. We show that both bacterial and host factors determine the numerical distributions of bacteria within host cells and thus the level of dispersiveness of the infection.

Kinetic Assessment of Measured Mass Flow Rates and Streamwise Pressure Distributions in Microchannel Gas Flows

Measured mass flow rates and streamwise pressure distributions of gas flowing through microchannels were reported by many researchers. Assessment of these data is crucial before they are used in the examination of slip models and numerical schemes, and in the design of microchannel elements in various MEMS devices. On the basis of kinetic solutions of the mass flow rates and pressure distributions in microchannel gas flows, the measured data available are properly normalized and then are compared with each other. The 69 normalized data of measured pressure distributions are in excellent agreement, and 67 of them are within 1 +/- 0.05. The normalized data of mass flow-rates ranging between 0.95 and 1 agree well with each other as the inlet Knudsen number Kn (i) < 0.02, but they scatter between 0.85 and 1.15 as Kn (i) > 0.02 with, to some extent, a very interesting bifurcation trend.

Thickness And Component Distributions Of Yttrium-Titanium Alloy Films In Electron-Beam Physical Vapor Deposition

Thickness and component distributions of large-area thin films are an issue of international concern in the field of material processing. The present work employs experiments and direct simulation Monte Carlo (DSMC) method to investigate three-dimensional low-density, non-equilibrium jets of yttrium and titanium vapor atoms in an electron-beams physical vapor deposition (EBPVD) system furnished with two or three electron-beams, and obtains their deposition thickness and component distributions onto 4-inch and 6-inch mono-crystal silicon wafers. The DSMC results are found in excellent agreement with our measurements, such as evaporation rates of yttrium and titanium measured in-situ by quartz crystal resonators, deposited film thickness distribution measured by Rutherford backscattering spectrometer (RBS) and surface profilometer and deposited film molar ratio distribution measured by RBS and inductively coupled plasma atomic emission spectrometer (ICP-AES). This can be taken as an indication that a combination of DSMC method with elaborate measurements may be satisfactory for predicting and designing accurately the transport process of EBPVD at the atomic level.

Experimental Investigation Of Particle Velocity Distributions In Windblown Sand Movement

With the PDPA (Phase Doppler Particle Analyzer) measurement technology, the probability distributions of particle impact and lift-off velocities on bed surface and the particle velocity distributions at different heights are detected in a wind tunnel. The results show that the probability distribution of impact and lift-off velocities of sand grains can be expressed by a log-normal function, and that of impact and lift-off angles complies with an exponential function. The mean impact angle is between 28 degrees and 39 degrees, and the mean lift-off angle ranges from 30 degrees to 44 degrees. The mean lift-off velocity is 0.81-0.9 times the mean impact velocity. The proportion of backward-impacting particles is 0.05-0.11, and that of backward-entrained particles ranges from 0.04 to 0.13. The probability distribution of particle horizontal velocity at 4 mm height is positive skew, the horizontal velocity of particles at 20 mm height varies widely, and the variation of the particle horizontal velocity at 80 mm height is less than that at 20 mm height. The probability distribution of particle vertical velocity at different heights can be described as a normal function.

Distributions of temperature, velocity and viscosity at symmetrical axis of cylindrical plume in earth's mantle with variable viscosity

The parameters at the symmetrical axis of a cylindrical plume characterize the strength of this plume and provide a boundary condition which must be given to investigate the structure of a plume. For Newtonian fluid with a temperature-and pressure-dependence viscosity, an asymptotical solution of hydrodynamic equations at the symmetrical axis of the plume is found in the present paper. The temperature, upward velocity and viscosity at the symmetrical axis have been obtained as functions of depth, The calculated results have been given for two typical sets of Newtonian rheological parameters. The results obtained show that the temperature distribution along the symmetrical axis is nearly independent of the theological parameters. The upward velocity at the symmetrical axis, however, is strongly dependent on the rheological parameters.

Egalitarian distributions in coalitional models: The Lorenz criterion

The paper presents a framework where the most important single-valued solutions in the literature of TU games are jointly analyzed. The paper also suggests that similar frameworks may be useful for other coalitional models.