- author = "Leonardo Trujillo and Omar M. Villanueva and Daniel E. Hernandez",
- title = "A Novel Approach For Search-Based Program Repair",
- journal = "IEEE Software",
- year = "2021",
- volume = "38",
- number = "4",
- pages = "36--42",
- month = jul # "-" # aug,
- keywords = "genetic algorithms, genetic programming, genetic improvement, SBSE, automatic program repair, APR, GenProg, computer bugs, search problems, maintenance engineering, standards, linear programming, statistics, space exploration, Novelty Search",
- ISSN = "1937-4194",
-
DOI = "
doi:10.1109/MS.2021.3070552",
- size = "7 pages",
- abstract = "In search-based software engineering, automatic bug repair methods use use an objective function that measures a patch quality to guide the search. This work presents results of the first study that focuses on solution novelty instead. Novelty search works under the assumption that most real-world problems are intrinsically difficult. By shifting how program patches are evaluated, away from quality and towards novelty, this technique increases a bug repair system's ability to explore the solution space, produce more viable patches and repair more bugs.",
-
notes = "'By shifting how program patches are evaluated, away
from quality and toward novelty, the novelty search
technique increases a bug-repair system ability to
explore the solution space, produce more viable
patches, and repair more bugs.'
p38 'to apply NS to an evolutionary algorithm, only the fitness function has to be modified.'
p41 'NS-GenProg out performs standard GenProg in terms of total bug repairs'
Tried on: ManyBugs (gmp gzip libtiff lighttpd python php).
3 mutations: delete, add, swap.
Fig 2 population diversity
Tecnlogico Nacional de Mexico/IT de Tijuana, Tijuana, Mexico",
