Biomimicry-based Algorithms and Their Lack of Generalization

Dean Klopsch. Biomimicry-based Algorithms and Their Lack of Generalization. Bachelor’s thesis, University of Bern, February 2021. Details.


Biomimicry has received much attention in engineering, and many breakthrough discoveries have been guided by a solution found in nature. However, many biomimicry-based proposals apply to a specific problem, provide limited context, and lack implementation details. That makes it unnecessarily hard for practitioners to find relevant literature for their problems. To investigate this problem, we performed a literature review on 111 publications related to biomimicry and extracted several characteristics, e.g., meta-data, the solution, and the investigated species. In particular, we were interested in whether the proposed algorithms could be used for other use cases. Our results indicate a structural issue: publications related to new or adapted algorithms very prominently emphasize on a specific use case, instead of the generalized problem category, e.g., clustering. We found that 38% lack generalization at least in one of the introductory elements (i.e., title, abstract, and introduction), and that 53% of them lack generalization entirely. Moreover, 40% of the proposed algorithms lack at least one major characteristic, e.g., code samples or benchmarks against state of the art algorithms. We motivate the found generalization problem with our adapted implementation of an algorithm proposed for load scheduling. Moreover, the artifacts of this study can support practitioners in finding more efficiently existing solutions across research domains.

Posted by scg at 17 February 2021, 4:15 pm link
Last changed by admin on 21 April 2009