This paper describes STAGE, a learning approach to automatically improving search performance on optimization problems. STAGE learns an evaluation function which predicts the outcome of a local search algorithm, such as hillclimbing or WALKSAT, as a function of state features along its search trajectories. The learned evaluation function is used to bias future search trajectories toward better optima. We present positive results on six large-scale optimization domains.

This paper describes algorithms that learn to improve search performance on largescale optimization tasks. The main algorithm, Stage, works by learning an evaluation function that predicts the outcome of a local search algorithm, such as hillclimbing or Walksat, from features of states visited during search. The learned evaluation function is then used to bias future search trajectories toward better optima on the same problem. Another algorithm, X-Stage, transfers previously learned evaluation functions to new, similar optimization problems. Empirical results are provided on seven large-scale optimization domains: bin-packing, channel routing, Bayesian network structure-finding, radiotherapy treatment planning, cartogram design, Boolean satisfiability, and Boggle board setup.

This paper discusses the direction toward computational approaches for product variety design. Product variety design here refers to an engineering challenge for designing multiple models simultaneously. This paper defines the formal representation of product variety design, and proposes a structure of design tasks involved in product variety. The analysis reveals several characteristics of product variety design, and identifies unique and essential tasks for simultaneously designing multiple models: correlation analysis, competing against system constraints, control variable selection, coverage distribution and combination selection. The paper closes with the scenario toward computational methods and difficulties to overcome for product variety optimization. 1 INTRODUCTION Today's product market demands a wide variety of products. Companies must have a wider vision in product development; for instance, for organizing product lines against several market segments, establishing a long-...

The subject of this thesis is the A-Teams formalism. This formalism facilitates the organization of multiple algorithms, encapsulated as autonomous agents, into cooperating teams to solve difficult problems. The ATeams formalism is one of many agent-based systems, and I start by providing a taxonomy of agent-based systems that allows us to see they how A-Teams relate to other agent-based systems. A-Teams are constructed from memories that store solutions and agents that work on those solutions. ATeams are open to the addition of new memories as well as of new agents. Sets of memories and agents can also be combined in different ways to create a variety of customized A-Teams. As new memories and agents are created, they can be added to existing repositories and reused for future applications. The automatic construction of problem-specific custom A-Teams from repositories of components has been a long standing goal of research in A-Teams. Current guidelines for A-Team construction requir...

Evaluation functions are an essential component of practical search algorithms for optimization, planning and control. Examples of such algorithms include hillclimbing, simulated annealing, best-first search, A*, and alpha-beta. In all of these, the evaluation functions are typically built manually by domain experts, and may require considerable tweaking to work well. I will investigate the thesis that statistical machine learning can be used to automatically generate high-quality evaluation functions for practical combinatorial problems. The data for such learning is gathered by running trajectories through the search space. The learned evaluation function may be applied either to guide further exploration of the same space, or to improve performance in new problem spaces which share similar features. Two general families of learning algorithms apply here: reinforcement learning and meta-optimization. The reinforcement learning approach, dating back to Samuel's checkers player [ 1959 ...

An effective partnership between industry and the university resulted in the system of design tools for the layout of HVAC systems presented in this paper and illustrated with the design of a heat pump. The system provides tools to assist in the placement of components and routing of tubes between the components. Traditional tubes, tubes that have minimized length and number of bends, and those that are impossible to route in the traditional manner, are generated. The paper provides insight on both the collaborative research interaction and the resulting set of tools. 1.

Spatial layout is the problem of arranging a set of components in an enclosure such that a set of objectives and constraints is satisfied. The constraints may include non-interference of objects, accessibility requirements and connection cost limits. Spatial layout problems are found primarily in the domains of electrical engineering and mechanical engineering in the design of integrated circuits and mechanical or electromechanical artifacts. Traditional approaches include ad-hoc (or specialized) heuristics, Genetic Algorithms and Simulated Annealing. The A-Teams approach provides a way of synergistically combining these approaches in a modular agent based fashion. A-Teams are also open to the addition of new agents. Modifications in the task requirements translate to modifications in the agent mix. In this paper we describe how modular A-Team based optimization can be used to solve 3 dimensional spatial layout problems. INTRODUCTION A number of design and manufacturing problems requ...

This paper describes Stage, a learning algorithm that automatically improves search performance on large-scale optimization problems. Stage learns an evaluation function that predicts the outcome of a local search algorithm, such as hillclimbing or Walksat, from features of states visited during search. The learned evaluation function is used to bias future search trajectories toward better optima on the same problem. This paper presents the Stage algorithm; an extension, X-Stage, that transfers learned evaluation functions to new, similar optimization problems; and empirical results on seven large-scale optimization domains: bin-packing, channel routing, Bayes network structure-finding, radiotherapy treatment planning, cartogram design, Boolean satisfiability, and Boggle board setup.

This paper describes the combination of previously-developed component placement and routing algorithms into an integrated computational approach to product layout optimization. Previ