Raj Subbu

Evolutionary algorithms + domain knowledge = real-world evolutionary computation

We discuss implicit and explicit knowledge representation mechanisms for evolutionary algorithms (EAs). We also describe offline and online metaheuristics as examples of explicit methods to leverage this knowledge. We illustrate the benefits of this approach with four real-world applications. The first application is automated insurance underwriting-a discrete classification problem, which requires a careful tradeoff between the percentage of insurance applications handled by the classifier and its classification accuracy. The second application is flexible design and manufacturing-a combinatorial assignment problem, where we optimize design and manufacturing assignments with respect to time and cost of design and manufacturing for a given product. Both problems use metaheuristics as a way to encode domain knowledge. In the first application, the EA is used at the metalevel, while in the second application, the EA is the object-level problem solver. In both cases, the EAs use a single-valued fitness function that represents the required tradeoffs. The third application is a lamp spectrum optimization that is formulated as a multiobjective optimization problem. Using domain customized mutation operators, we obtain a well-sampled Pareto front showing all the nondominated solutions. The fourth application describes a scheduling problem for the maintenance tasks of a constellation of 25 low earth orbit satellites. The domain knowledge in this application is embedded in the design of a structured chromosome, a collection of time-value transformations to reflect static constraints, and a time-dependent penalty function to prevent schedule collisions.

Fuzzy logic controlled genetic algorithms versus tuned genetic algorithms: an agile manufacturing application

This paper presents a comparison of the performance of a fuzzy logic controlled genetic algorithm (FLC-GA) and a parameter tuned genetic algorithm (TGA) for an agile manufacturing application. In the FLC-GA, fuzzy logic controllers dynamically schedule parameters of the object-level GA. A fuzzy knowledge-base is automatically identified and tuned using a high-level GA. In the TGA, a high-level GA is used to determine an optimal static parameter set for the object-level GA. The object-level GA supports a global evolutionary optimization of design, manufacturing, and supplier planning decisions for manufacturing of printed circuit assemblies in an agile environment. We demonstrate that high-level system identification or tuning performed with small object-level search spaces, can be extended to more elaborate object-level search spaces. The TGA performs superior searches, but incurs large search times. The FLC-GA performs faster searches than a TGA, and is slower than the GA that utilizes a canonical static parameter set. However, search quality of the FLC-GA is comparable to that of the GA which utilizes a canonical static parameter set.

Multiobjective financial portfolio design: a hybrid evolutionary approach

A principal challenge in modern computational finance is efficient portfolio design - portfolio optimization followed by decision-making. Optimization based on even the widely used Markowitz two-objective mean-variance approach becomes computationally challenging for real-life portfolios. Practical portfolio design introduces further complexity as it requires the optimization of multiple return and risk measures subject to a variety of risk and regulatory constraints. Further, some of these measures may be nonlinear and nonconvex, presenting a daunting challenge to conventional optimization approaches. We introduce a powerful hybrid multiobjective optimization approach that combines evolutionary computation with linear programming to simultaneously maximize these return measures, minimize these risk measures, and identify the efficient frontier of portfolios that satisfy all constraints. We also present a novel interactive graphical decision-making method that allows the decision-maker to quickly down-select to a small subset of efficient portfolios. The approach has been tested on real-life portfolios with hundreds to thousands of assets, and is currently being used for investment decision-making in industry.

Multicriteria decision making (mcdm): a framework for research and applications

We view Multicriteria Decision Making (MCDM) as the conjunction of three components: search, preference tradeoffs, and interactive visualization. The first MCDM component is the search process over the space of possible solutions to identify the non-dominated solutions that compose the Pareto set. The second component is the preference tradeoff process to select a single solution (or a small subset of solutions) from the Pareto set. The third component is the interactive visualization process to embed the decisionmaker in the solution refinement and selection loop. We focus on the intersection of these three components and we highlight some research challenges, representing gaps in the intersection. We introduce a requirement framework to compare most MCDM problems, their solutions, and analyze their performances. We focus on two research challenges and illustrate them with three case studies in electric power management, financial portfolio rebalancing, and air traffic planning.

Fast meta-models for local fusion of multiple predictive models

Fusing the outputs of an ensemble of diverse predictive models usually boosts overall prediction accuracy. Such fusion is guided by each models local performance, i.e., each models prediction accuracy in the neighborhood of the probe point. Therefore, for each probe we instantiate a customized fusion mechanism. The fusion mechanism is a meta-model, i.e., a model that operates one level above the object-level models whose predictions we want to fuse. Like these models, such a meta-model is defined by structural and parametric information. In this paper, we focus on the definition of the parametric information for a given structure. For each probe point, we either retrieve or compute the parameters to instantiate the associated meta-model. The retrieval approach is based on a CART-derived segmentation of the probes state space, which contains the meta-model parameters. The computation approach is based on a run-time evaluation of each models local performance in the neighborhood of the probe. We explore various structures for the meta-model, and for each structure we compare the pre-compiled (retrieval) or run-time (computation) approaches. We demonstrate this fusion methodology in the context of multiple neural network models. However, our methodology is broadly applicable to other predictive modeling approaches. This fusion method is illustrated in the development of highly accurate models for emissions, efficiency, and load prediction in a complex power plant. The locally weighted fusion method boosts the predictive performance by 30-50% over the baseline single model approach for the various prediction targets. Relative to this approach, typical fusion strategies that use averaging or globally weighting schemes only produce a 2-6% performance boost over the same baseline.

Modeling and convergence analysis of distributed coevolutionary algorithms

A theoretical foundation is presented for modeling and convergence analysis of a class of distributed coevolutionary algorithms applied to optimization problems in which the variables are partitioned among p nodes. An evolutionary algorithm at each of the p nodes performs a local evolutionary search based on its own set of primary variables, and the secondary variable set at each node is clamped during this phase. An infrequent intercommunication between the nodes updates the secondary variables at each node. The local search and intercommunication phases alternate, resulting in a cooperative search by the p nodes. First, we specify a theoretical basis for a class of centralized evolutionary algorithms in terms of construction and evolution of sampling distributions over the feasible space. Next, this foundation is extended to develop a model for a class of distributed coevolutionary algorithms. Convergence and convergence rate analyses are pursued for basic classes of objective functions. Our theoretical investigation reveals that for certain unimodal and multimodal objectives, we can expect these algorithms to converge at a geometrical rate. The distributed coevolutionary algorithms are of most interest from the perspective of their performance advantage compared to centralized algorithms, when they execute in a network environment with significant local access and internode communication delays. The relative performance of these algorithms is therefore evaluated in a distributed environment with realistic parameters of network behavior.

Evolutionary optimization of transition probability matrices for credit decision-making

Statistical transition probability matrices (TPMs), which indicate the likelihood of obligor credit state migration over a certain time horizon, have been used in various credit decision-making applications. A standard approach of calculating TPMs is to form a one-year empirical TPM and then project it into the future based on Markovian and time-homogeneity assumptions. However, the one-year empirical TPM calculated from historical data generally does not satisfy desired properties. We propose an alternative methodology by formulating the problem as a constrained optimization problem requiring satisfaction of all the desired properties and minimization of the discrepancy between predicted multi-year TPMs and empirical evidence. The problem is high-dimensional, non-convex, and non-separable, and is not effectively solved by nonlinear programming methods. To address the difficulty, we investigated evolutionary algorithms (EAs) and problem representation schemas. A self-adaptive differential evolution algorithm JADE, together with a new representation schema that automates constraint satisfaction, is shown to be the most effective technique.

Evolutionary Decision Support for Distributed Virtual Design in Modular Product Manufacturing

Superior design-manufacturing-supplier decisions are critical to the survival of modern enterprises that seek to compete in a dynamic global marketplace. An evolutionary decision support system (virtual design environment) that supports such decision-making is introduced. The VDE framework utilizes evolutionary agents as program entities that generate and execute queries among distributed and heterogeneous computing applications and information resources. Evolutionary agents support a global optimization of design-manufacturing-supplier planning decisions by retrieving appropriate information from distributed information repositories, and fostering the systematic selection of planning alternatives that reduce cost and increase throughput. This paper introduces a general formulation of the design-manufacturing-supplier planning decision problem that is applicable to a variety of assembly-oriented design-manufacturing domains. We present a description of the prototype VDE that has been used to examine plann...

Locally Weighted Fusion of Multiple Predictive Models

Fusing the outputs from an ensemble of models in an effective way can often boost overall model accuracy. This paper presents a novel method, called locally weighted fusion, which aggregates the results of multiple predictive models based on local accuracy measures of these models in the neighborhood of the probe point for which we want to make a prediction. While we demonstrate the method in the context of multiple neural network models, the concepts may be applied to other predictive techniques as well. This fusion method is applied to develop highly accurate models for emissions, efficiency, and load prediction in a complex real-world power plant. The locally weighted fusion method boosts the predictive performance by 20-40% over the baseline single model approach for the various prediction targets. Relative to this approach, fusion strategies which apply averaging or globally weighting only produce a 2-6% performance boost over the baseline.

A virtual design environment using evolutionary agents

The virtual design environment is an information architecture to support design-manufacturing-supplier planning decisions in a distributed, heterogeneous environment. The approach utilizes evolutionary intelligent agents as program entities which generate and execute queries among distributed computing applications and databases. The evolutionary agents support a global evolutionary optimization process in which successive populations systematically select planning alternatives which reduce cost and increase throughput. A prototype of the virtual design environment has been implemented using CORBA as a principal distributed systems programming tool. The prototype has been used to examine design-manufacturing-supplier decisions for a real commercial electronic circuit board product (Pitney Bowes Inc.) and to explore plans in controlled experiments with alternative manufacturing facilities.