Universität Paderborn » SFB 901 » Projects » Project Area B » B2 » Abstract

Subproject B2

Configuration and Rating

Abstract

This project will perform research on the configuration and rating of OTF services. Starting with experiences in technical domains resource-, model- and case-based methods for (automatic) configuration will be investigated. Adequate modeling levels shall be used to increase efficiency when processing configuration knowledge and checking composition preconditions. For controlling the configuration process, ratings based on hard and soft criteria will be investigated. These criteria will be estimated by adaptation and learning methods and aggregated by homeostatic techniques.

Goals and Challenges

Figure 1: Conflicting priorities (click to enlarge).

The main goal of subproject B2 is to support OTF providers with respect to

the search, selection, composition and rating

of services in large-scale OTF markets. In this context, subproject B2 moves between the conflicting priorities of three general dimensions: expressive power of description formalism, fitting accuracy of services, as well as the efficiency and level of automation (see Figure 1).

Basic Configuration:

  • Selection, parameterization and aggregation of services within an automated configuration process.
  • Quality-based service selection as a global or local decision-making problem.

Matching and Configuration Approaches:

  • Matching of functional service properties considering the two contradicting dimensions of accuracy and efficiency.
  • Case-based configuration approaches based on domain knowledge for automatic service configuration.

Learning of Value Functions:

  • Learning approaches for automatic service configuration (e.g., Reinforcement Learning).
  • Control of learning approaches in order to improve convergence behavior with respect to OTF service configuration.

Efficiency of Methods and Approaches:

  • Complexity of decision-making problems within the configuration process.
  • Convergence behavior of learning approaches as a function of different influencing variables.

Example: Case-based Configuration

Figure 2: Case-based configuration (click to enlarge).

Example: Configuration Controlled by Learned Value Functions

Figure 3: Exemplary configuration process, which is controlled on the basis of learned value functions (click to enlarge).
  1. User request for a (complex) service
  2. Identification of alternatives:
    • case-based approach
    • matching of specifications
  3. Rating of identified alternatives:
    • aggregation, rating and comparison of quality values
    • prioritized list of alternatives
  4. Selection of one alternative (e.g. with highest priority) by configurator
  5. Update quality values of previously selected alternative dependent on currently selected alternative and feedback
  6. Proceed with 2. until service is completely configured (composed)