International Business Machines Corporation

Renaud Dumeur

Jean-Francois Puget

Paul Shaw

Embodiments relate to approximate assignment in a constraint based evolutionary search. An aspect includes providing a genome representing a collection of variable assignment preferences encoded as genes. Another aspect includes reducing the domain until a unit sized domain is reached, the unit sized domain being an approximation to a value V. Another aspect includes searching for a first assignment of the value V that is less than or equal to the unit sized domain and a second assignment of the value V that is greater than the unit sized domain. Another aspect includes responsive to a first assignment and a second assignment being found, assigning the value V of one of the first assignment and the second assignment having the least distance from the unit sized domain to a variable X.

Approximate assignment operator for constrained based evolutionary search

The present invention relates to methods and systems for applying evolutionary algorithms to generate robust search strategies for problems including decision variables. In one aspect, the invention encodes genomes of at least one triplet comprising a variable, assignment priority, and assigned value. The genome may later be decoded to determine a partial or complete assignment of values to variables. If a partial assignment is reached, a search strategy may be applied to generate a complete or more complete assignment. The genomes may also be evolved to produce offspring genomes. One type of evolutionary operator, called the Lamarckian operator is introduced, wherein the similarities, differences, and unbound variables resulting from the decoding of two or more parent genomes are collected. These collections are then used to encode an offspring genome, building upon the strengths of the parents.

Growing community of inventors

Paris, France

Renaud Dumeur