Case injected genetic algorithms for learning across problems

Abstract

Genetic algorithms (GAs) augmented with a case-based memory of past design problem-solving attempts are used to obtain better performance over time on sets of similar design problems. Rather than starting anew on each design, a GA's population is periodically injected with appropriate intermediate design solutions to similar, previously solved design problems. Experimental results on configuration design problems: the design of parity checker and adder circuits, demonstrate the performance gains from the approach and show that the system learns to take less time to provide quality solutions to a new design problem as it gains experience from solving other similar design problems.

Solving Incremental Optimization Problems via Cooperative Coevolution

“…The Optimizer function is any optimizer of choice which optimizes the component κ for a certain number of iterations. Once the component κ is optimized, its contribution is updated and set to the magnitude of improvement before and after optimization (lines [12][13][14]. This exploitation strategy is broken with probability p e to give all components a chance to be optimized, which is done by setting the contribution of all components to ∞.…”

“…The key issue in experience-based optimization is how to make use of historical information to guide future optimization, either implicitly or explicitly. For example, Sushil Louis borrowed the idea from the case-based reasoning, where the previously obtained solutions are injected into the search process to improve the performance on syntactically similar problems [13]. This can be seen as an implicit way to use historical information.…”

Case-DeSC: A system for case-based design with soft computing techniques

“…This module is similar to the framework proposed in CIGAR (Louis, 2004), but the methodology of case retrieval and case injection in the GA population are different.…”

“…Adaptation is performed through thematic abstraction (generation of new solution through a single case) and composition (splits or merges case pieces generating new solutions). CIGAR (Louis, 2004) addresses design as an optimization problem whose solution is extracted by a GA augmented by a case-based memory that periodically injects similar design solutions to the GA population Rosenman (2000) utilizes an evolutionary algorithm for performing adaptation tasks in a CBR system and evaluates his method in 2-D spatial design of houses. In the context of learning knowledge that is used in retrieval and the adaptation tasks in CBD systems, researchers have been investigating the impact of specificto-general and general-to-specific learning on adaptation knowledge (Wiratunga, Craw, & Rowe, 2002).…”

Knowledge reuse in genetic programming applied to visual learning

“…In the context of GP, knowledge reuse is often connected with knowledge encapsulation [11,25,3,5], which is however not used in the approach presented here. Among reported research, the work done by Louis et al most resembles our contribution [18,16,17]. In particular, in Case Injected Genetic Algorithms (CIGAR) described in [17], the experience of the system is stored in a form of solutions to problems solved earlier ('cases').…”