Search

The purpose of this article is to develop and study a decentralized strategy for Pareto optimization of an aggregate cost consisting of regularized risks. Each risk is modeled as the expectation of some loss function with unknown probability distribution, while the regularizers are assumed deterministic, but are not required to be differentiable or even continuous. The individual, regularized, cost functions are distributed across a strongly connected network of agents, and the Pareto optimal solution is sought by appealing to a multiagent diffusion strategy. To this end, the regularizers are smoothed by means of infimal convolution, and it is shown that the Pareto solution of the approximate smooth problem can be made arbitrarily close to the solution of the original nonsmooth problem. Performance bounds are established under conditions that are weaker than assumed before in the literature and, hence, applicable to a broader class of adaptation and learning problems.

This work introduces the particle-intensity channel (PIC) as a model for molecular communication systems and characterizes the capacity limits as well as properties of the optimal (capacity-achieving) input distributions for such channels. In the PIC, the transmitter encodes information, in symbols of a given duration, based on the probability of particle release, and the receiver detects and decodes the message based on the number of particles detected during the symbol interval. In this channel, the transmitter may be unable to control precisely the probability of particle release, and the receiver may not detect all the particles that arrive. We model this channel using a generalization of the binomial channel and show that the capacity-achieving input distribution for this channel always has mass points at probabilities of particle release of zero and one. To find the capacity-achieving input distributions, we develop an efficient algorithm we call dynamic assignment Blahut-Arimoto (DAB). For diffusive particle transport, we also derive the conditions under which the input with two mass points is capacity-achieving., Comment: arXiv admin note: text overlap with arXiv:1705.08040

Signal processing applications of semidefinite optimization are often rooted in sum-of-squares representations of nonnegative trigonometric polynomials. Interior-point solvers for semidefinite optimization can handle constraints of this form with a per-iteration-complexity that is cubic in the degree of the trigonometric polynomial. The purpose of this paper is to discuss first-order methods with a lower complexity per iteration. The methods are based on generalized proximal operators defined in terms of the Itakura–Saito distance. This is the Bregman distance defined by the negative entropy function. The choice for the Itakura–Saito distance is motivated by the fact that the associated generalized projection on the set of normalized nonnegative trigonometric polynomials can be computed at a cost that is roughly quadratic in the degree of the polynomial. The generalized projection is the key operation in generalized proximal first-order methods that use Bregman distances instead of the squared Euclidean distance. The paper includes numerical results with Auslender and Teboulle's accelerated proximal gradient method for Bregman distances.

Integration of a large number of flexible consumers in a smart grid requires a scalable power balancing strategy. We formulate the control problem as an optimization problem to be solved repeatedly by the aggregator in a model predictive control framework. To solve the large-scale control problem in real-time requires decomposition methods. We propose a decomposition method based on Douglas–Rachford splitting to solve this large-scale control problem. The method decomposes the problem into smaller subproblems that can be solved in parallel, e.g., locally by each unit connected to an aggregator. The total power consumption is controlled through a negotiation procedure between all cooperating units and an aggregator that coordinates the overall objective. For large-scale systems, this method is faster than solving the original problem and can be distributed to include an arbitrary number of units. We show how different aggregator objectives are implemented and provide simulations of the controller including the computational performance.

A recent generalization of the Kalman-Yakubovich-Popov (KYP) lemma establishes the equivalence between a semi-infinite inequality on a segment of a line or circle in the complex plane and a linear matrix inequality (LMI). In this technical note we show that when the data are real, the matrix variables in the LMI can be restricted to be real, even when the frequency range is asymmetric with respect to the real axis. © 2006 IEEE. ispartof: IEEE Transactions on Automatic Control vol:56 issue:12 pages:2940-2944 status: published

This paper presents a new method for fitting a convex piecewise-linear function to a given set of data, which can serve as an empirical modeling framework for circuit optimization via geometric programming. The method iteratively solves a series of linear optimization problems to minimize the fitting error. To reduce the fitting error in each iteration, an extra plane is added in the region where the largest error occurs. For verification, we apply the method to create transistor-level models in 90-nm complementary metal-oxide-semiconductor technology. Numerical results indicate that the proposed method can generate process-dependent transistor-level models with reasonable modeling accuracy.

In this paper, we propose a distributed parallel support vector machine (DPSVM) training mechanism in a configurable network environment for distributed data mining. The basic idea is to exchange support vectors among a strongly connected network (SCN) so that multiple servers may work concurrently on distributed data set with limited communication cost and fast training speed. The percentage of servers that can work in parallel and the communication overhead may be adjusted through network configuration. The proposed algorithm further speeds up through online implementation and synchronization. We prove that the global optimal classifier can be achieved iteratively over an SCN. Experiments on a real-world data set show that the computing time scales well with the size of the training data for most networks. Numerical results show that a randomly generated SCN may achieve better performance than the state of the art method, cascade SVM, in terms of total training time.

We discuss a method for multidimensional FIR filter design via sum-of-squares formulations of spectral mask constraints. The sum-of-squares optimization problem is expressed as a semidefinite program with low-rank structure, by sampling the constraints using discrete cosine and sine transforms. The resulting semidefinite program is then solved by a customized primal-dual interior-point method that exploits low-rank structure. This leads to a substantial reduction in the computational complexity, compared to general-purpose semidefinite programming methods that exploit sparsity.

We consider the problem of evaluating the probability of error of binary communication systems in the presence of additive noise and intersymbol interferences whose statistics are inexactly known due to the estimation errors of the channel coefficients. We present a new method using semidefinite programming to evaluate tight bounds on the error probability based on the upper and lower bounds on the moments of those interferences. Numerical results are provided and compared with a previously published technique.

Several important problems in control theory can be reformulated as semidefinite programming problems, i.e., minimization of a linear objective subject to linear matrix inequality (LMI) constraints. From convex optimization duality theory, conditions for infeasibility of the LMIs, as well as dual optimization problems, can be formulated. These can in turn be reinterpreted in control or system theoretic terms, often yielding new results or new proofs for existing results from control theory. We explore such connections for a few problems associated with linear time-invariant systems.

Conventional methods for optimal sizing of wires and transistors use linear resistor-capacitor (RC) circuit models and the Elmore delay as a measure of signal delay. If the RC circuit has a tree topology, the sizing problem reduces to a convex optimization problem that can be solved using geometric programming. The tree topology restriction precludes the use of these methods in several sizing problems of significant importance to high-performance deep submicron design, including for example, circuits with loops of resistors, e.g., clock distribution meshes and circuits with coupling capacitors, e.g., buses with crosstalk between the wires. In this paper, we propose a new optimization method that can be used to address these problems. The method is based on the dominant time constant as a measure of signal propagation delay in an RC circuit instead of Elmore delay. Using this measure, sizing of any RC circuit can be cast as a convex optimization problem and solved using recently developed efficient interior-point methods for semidefinite programming. The method is applied to three important sizing problems: clerk mesh sizing and topology design, sizing of tristate buses, and sizing of bus line widths and spacings taking crosstalk into account.

A number of important problems from system and control theory can be numerically solved by reformulating them as convex optimization problems with linear matrix inequality (LMI) constraints. While numerous articles have appeared cataloging applications of LMIs to control system analysis and design, there have been few publications in the control literature describing the numerical solution of these optimization problems. The purpose of this article is to provide an overview of the state of the art of numerical algorithms for LMI problems, and of the available software.