Pilar Martínez Ortigosa

Hardware description of multi-layer perceptrons with different abstraction levels

Abstract This paper presents different hardware implementations of a multi-layer perceptron (MLP) for speech recognition. When defining the designs, we have used two different abstraction levels: a register transfer level and a higher algorithmic-like level. The implementations have been developed and tested into reconfigurable hardware (FPGA) for embedded systems. We also present a comparative study of the costs for the two considered approaches with regards to the silicon area, speed and required computational resources. The research is completed with the study of different implementation versions with diverse degrees of parallelism. The final aim is the comparison of the methodologies applied in the two abstraction levels for designing hardware MLP’s or similar computational structures.

Solving the multiple competitive facilities location and design problem on the plane

A continuous location problem in which a firm wants to set up two or more new facilities in a competitive environment is considered. Other facilities offering the same product or service already exist in the area. Both the locations and the qualities of the new facilities are to be found so as to maximize the profit obtained by the firm. This is a global optimization problem, with many local optima. In this paper we analyze several approaches to solve it, namely, three multistart local search heuristics, a multistart simulated annealing algorithm, and two variants of an evolutionary algorithm. Through a comprehensive computational study it is shown that the evolutionary algorithms are the heuristics that provide the best solutions. Furthermore, using a set of problems for which the optimal solutions are known, only the evolutionary algorithms were able to find the optimal solutions for all the instances. The evolutionary strategies presented in this paper can be easily adapted to handle other continuous location problems.

Heuristics for the facility location and design (1|1)-centroid problem on the plane

A chain (the leader) wants to set up a single new facility in a planar market where similar facilities of a competitor (the follower), and possibly of its own chain, are already present. The follower will react by locating another single facility after the leader locates its own facility. Fixed demand points split their demand probabilistically over all facilities in the market in proportion to their attraction to each facility, determined by the different perceived qualities of the facilities and the distances to them, through a gravitational model. Both the location and the quality (design) of the new leader’s facility are to be found. The aim is to maximize the profit obtained by the leader following the follower’s entry. Four heuristics are proposed for this hard-to-solve global optimization problem, namely, a grid search procedure, an alternating method and two evolutionary algorithms. Computational experiments show that the evolutionary algorithm called UEGO_cent.SASS provides the best results.

On success rates for controlled random search

Controlled Random Search (CRS) is a simple population based algorithm which despite its attractiveness for practical use, has never been very popular among researchers on Global Optimization due to the difficulties in analysing the algorithm. In this paper, a framework to study the behaviour of algorithms in general is presented and embedded into the context of our view on questions in Global Optimization. By using as a reference a theoretical ideal algorithm called N-points Pure Adaptive Search (NPAS) some new analytical results provide bounds on speed of convergence and the Success Rate of CRS in the limit once it has settled down into simple behaviour. To relate the performance of the algorithm to characteristics of functions to be optimized, constructed simple test functions, called extreme cases, are used.

UEGO, an Abstract Clustering Technique for Multimodal Global Optimization

In this paper, UEGO, a new general technique for accelerating and/or parallelizing existing search methods is suggested. The skeleton of the algorithm is a parallel hill climber. The separate hill climbers work in restricted search regions (or clusters) of the search space. The volume of the clusters decreases as the search proceeds which results in a cooling effect similar to simulated annealing. Besides this, UEGO can be effectively parallelized; the communication between the clusters is minimal. The purpose of this communication is to ensure that one hill is explored only by one hill climber. UEGO makes periodic attempts to find new hills to climb. Empirical results are also presented which include an analysis of the effects of the user-given parameters and a comparison with a hill climber and a GA.

Two- and three-dimensional modeling and optimization applied to the design of a fast hydrodynamic focusing microfluidic mixer for protein folding

We present a design of a microfluidic mixer based on hydrodynamic focusing which is used to initiate the folding process (i.e., changes of the molecular structure) of a protein. The folding process is initiated by diluting (from 90% to 30%) the local denaturant concentration (initially 6 M GdCl solution) in a short time interval we refer to as mixing time. Our objective is to optimize this mixer by choosing suitable shape and flow conditions in order to minimize this mixing time. To this end, we first introduce a numerical model that enables computation of the mixing time of a mixer. This model is based on a finite element method approximation of the incompressible Navier-Stokes equations coupled with the convective diffusion equation. To reduce the computational time, this model is implemented in both full three-dimensional (3D) and simplified two-dimensional (2D) versions; and we analyze the ability of the 2D model to approximate the mixing time predicted by the 3D model. We found that the 2D model approximates the mixing time predicted by the 3D model with a mean error of about 15%, which is considered reasonable. Then, we define a mixer optimization problem considering the 2D model and solve it using a hybrid global optimization algorithm. In particular, we consider geometrical variables and injection velocities as optimization parameters. We achieve a design with a predicted mixing time of 0.10 μs, approximately one order of magnitude faster than previous mixer designs. This improvement can be in part explained by the new mixer geometry including an angle of π/5 radians at the channel intersection and injections velocities of 5.2 m s−1 and 0.038 m s−1 for the side and central inlet channels, respectively. Finally, we verify the robustness of the optimized result by performing a sensitivity analysis of its parameters considering the 3D model. During this study, the optimized mixer was demonstrated to be robust by exhibiting mixing time variations of the same order than the parameter ones. Thus, the obtained 2D design can be considered optimal also for the 3D model.

A robust and efficient algorithm for planar competitive location problems

Abstract In this paper we empirically analyze several algorithms for solving a Huff-like competitive location and design model for profit maximization in the plane. In particular, an exact interval branch-and-bound method and a multistart heuristic already proposed in the literature are compared with uego (Universal Evolutionary Global Optimizer), a recent evolutionary algorithm. Both the multistart heuristic and uego use a Weiszfeld-like algorithm as local search procedure. The computational study shows that uego is superior to the multistart heuristic, and that by properly fine-tuning its parameters it usually (in the computational study, always) find the global optimal solution, and this in much less time than the interval branch-and-bound method. Furthermore, uego can solve much larger problems than the interval method.

A two-level evolutionary algorithm for solving the facility location and design (1|1)-centroid problem on the plane with variable demand

In this work, the problem of a company or chain (the leader) that considers the reaction of a competitor chain (the follower) is studied. In particular, the leader wants to set up a single new facility in a planar market where similar facilities of the follower, and possibly of its own chain, are already present. The follower will react by locating another single facility after the leader locates its own facility. Both the location and the quality (representing design, quality of products, prices, etc.) of the new leader’s facility have to be found. The aim is to maximize the profit obtained by the leader considering the future follower’s entry. The demand is supposed to be concentrated at n demand points. Each demand point splits its buying power among the facilities proportionally to the attraction it feels for them. The attraction of a demand point for a facility depends on both the location and the quality of the facility. Usually, the demand is considered in the literature to be fixed or constant regardless the conditions of the market. In this paper, the demand varies depending on the attraction for the facilities. Taking variable demand into consideration makes the model more realistic. However, it increases the complexity of the problem and, therefore, the computational effort needed to solve it. Three heuristic methods are proposed to cope with this hard-to-solve global optimization problem, namely, a grid search procedure, a multistart algorithm and a two-level evolutionary algorithm. The computational studies show that the evolutionary algorithm is both the most robust algorithm and the one that provides the best results.

Parallel algorithms for continuous competitive location problems

A continuous location problem in which a firm wants to set up a single new facility in a competitive environment is considered. Other facilities offering the same product or service already exist in the area. Both the location and the quality of the new facility are to be found so as to maximize the profit obtained by the firm. This is a hard-to-solve global optimization problem. An evolutionary algorithm called Universal Evolutionary Global Optimizer (UEGO) seems to be the best procedure to cope with it, but the algorithm needs several hours of CPU time for solving large instances. In this paper, four parallelizations of UEGO are presented. They all are coarse-grain methods which differ in their migratory policies. A computational study is carried out to compare the performance of the parallel algorithms. The results show that one of the parallelizations always gives the best objective function value and has an almost linear speed-up for up to 16 processing elements for large instances.

FPGA Implementation of Multi-layer Perceptrons for Speech Recognition

In this work we present different hardware implementations of a multi-layer perceptron for speech recognition. The designs have been defined using two different abstraction levels: register transfer level (VHDL) and a higher algorithmic-like level (Handel-C). The implementations have been developed and tested into a reconfigurable hardware (FPGA) for embedded systems. A study of the two considered approaches costs (silicon area), speed and required computational resources is presented.