Léo Françoso Dal Piccol Sotto
ABSTRACT
Whereas evolutionary computation usually solves problems from scratch, organisms evolve under changing environments and possess flexibility, adapting from being good at one task to being good at a related task. There is abundant evidence that there are general properties that promote flexibility in nature, such as hierarchy, modularity, exploratory behavior, and degeneracys or neutrality.
Our interest is to understand if such properties can also be identified for non-biological systems. We thus study if a controller evolved by a genetic algorithm for one pole balancing task can be adapted to a different pole balancing task, and if this saves training time compared to evolving a new controller from scratch. Moreover, we investigate how diversity and degeneracy in the controllers population affect adaption efficiency by promoting high quality solutions that are both structurally and behaviorally diverse, concluding that it can potentially decrease the adaption cost.
- Timothy Atkinson, Detlef Plump, and Susan Stepney. 2020. Horizontal gene transfer for recombining graphs. Genetic Programming and Evolvable Machines 21 (2020), 321--347.Google Scholar
Digital Library
- J. Scott Brownlee. 2005. The pole balancing problem: a benchmark control theory problem.Google Scholar
- Antoine Cully, Jeff Clune, Danesh Tarapore, and Jean-Baptiste Mouret. 2015. Robots that can adapt like animals. Nature 521, 7553 (2015), 503--507.Google Scholar
- Gerald M Edelman and Joseph A Gally. 2001. Degeneracy and complexity in biological systems. Proceedings of the National Academy of Sciences 98, 24 (2001), 13763--13768.Google ScholarCross Ref
- Marc W Kirschner and John C Gerhart. 2008. The plausibility of life. Yale University Press.Google Scholar
- John R. Koza. 1992. Genetic Programming: On the Programming of Computers by Means of Natural Selection. MIT Press, Cambridge, MA, USA.Google ScholarDigital Library
- Sean Luke and Liviu Panait. 2002. Fighting Bloat With Nonparametric Parsimony Pressure, Vol. 2439.Google Scholar
- Merav Parter, Nadav Kashtan, and Uri Alon. 2008. Facilitated variation: how evolution learns from past environments to generalize to new environments. PLoS computational biology 4, 11 (2008), 1--15.Google Scholar
- Léo Françoso Dal Piccol Sotto, Paul Kaufmann, Timothy Atkinson, Roman Kalkreuth, and Márcio P. Basgalupp. 2021. Graph representations in genetic programming. Genetic Programming and Evolvable Machines 22 (2021), 607--636.Google ScholarDigital Library
- Léo Françoso Dal Piccol Sotto, Franz Rothlauf, Vinícius Veloso de Melo, and Márcio P. Basgalupp. 2022. An Analysis of the Influence of Noneffective Instructions in Linear Genetic Programming. Evolutionary Computation 30, 1 (2022), 51--74.Google ScholarCross Ref
- Masanori Suganuma, Masayuki Kobayashi, Shinichi Shirakawa, and Tomoharu Nagao. 2020. Evolution of Deep Convolutional Neural Networks Using Cartesian Genetic Programming. Evolutionary Computation 28, 1 (2020), 141--163.Google ScholarDigital Library
- Kay Chen Tan, Liang Feng, and Min Jiang. 2021. Evolutionary Transfer Optimization - A New Frontier in Evolutionary Computation Research. IEEE Computational Intelligence Magazine 16, 1 (2021), 22--33.Google ScholarDigital Library
- Sebastian Thrun and Lorien Pratt. 1998. Learning to learn. Springer Science & Business Media.Google Scholar
- James M Whitacre. 2010. Degeneracy: a link between evolvability, robustness and complexity in biological systems. Theoretical Biology and Medical Modelling 7, 1 (2010), 1--17.Google ScholarCross Ref
- Darrell Whitley. 1998. A Genetic Algorithm Tutorial. Statistics and Computing 4 (1998).Google Scholar
- Qiang Yang, Yu Zhang, Wenyuan Dai, and Sinno Jialin Pan. 2020. Transfer learning. Cambridge University Press.Google Scholar
- Shengxiang Yang. 2013. Evolutionary computation for dynamic optimization problems. In Proceedings of the 15th annual conference companion on Genetic and evolutionary computation. 667--682.Google ScholarDigital Library
Index Terms
- The pole balancing problem from the viewpoint of system flexibility
Recommendations
Fitness landscapes and graphs: multimodularity, ruggedness and neutrality
GECCO '12: Proceedings of the 14th annual conference companion on Genetic and evolutionary computationOne of the most commonly-used metaphors to describe the process of heuristic search methods in solving combinatorial optimization problems is that of a fitness landscape. The landscape metaphor appears most commonly in work related to evolutionary ...
Difficult features of combinatorial optimization problems and the tunable w-model benchmark problem for simulating them
GECCO '18: Proceedings of the Genetic and Evolutionary Computation Conference CompanionThe first event of the Black-Box Discrete Optimization Benchmarking (BB-DOB) workshop series aims to establish a set of example problems for benchmarking black-box optimization algorithms for discrete or combinatorial domains. In this paper, we 1) ...
A tunable model for multi-objective, epistatic, rugged, and neutral fitness landscapes
GECCO '08: Proceedings of the 10th annual conference on Genetic and evolutionary computationThe fitness landscape of a problem is the relation between the solution candidates and their reproduction probability. In order to understand optimization problems, it is essential to also understand the features of fitness landscapes and their ...
Comments