Researchers at the University of Johannesburg (UJ) have designed artificial intelligence (AI) algorithms based on biological processes that could improve AI functionality.
In recent years, bio-inspired algorithms have received increasing attention by the research community as well as the industry.
Scientifically, the field of evolutionary algorithms is a highly relevant research area, as new approaches to artificial intelligence spring from the idea that intelligence emerges as much from cells, bodies, and societies as it does from evolution, development, and learning.
Traditionally, artificial intelligence has been concerned with reproducing the abilities of human brains; alternate approaches take inspiration from a wider range of biological structures that are capable of autonomous operation.
A collaboration led by Professor Duncan Coulter, deputy-head: Computer Science and Michael Cilliers within the Academy of Computer Science and Software Engineering at UJ has designed algorithms based on biological processes that have inspired new optimisation methods.
The UJ team has developed a new evolutionary algorithm based on DNA methylation driven epigenetics/genetic processes to better adapt to changes in the environment and enhance computational intelligence.
Experiments show a joint computational-biological approach to study the algorithmic properties of biological processes across all levels of life (molecular, cellular, and organism).
The findings are published in the Springer’s lecture notes in computer science.
In 2019, Prof Coulter and Cilliers investigated the emergent behavior of many evolutionary algorithms where during their runtime, the diversity of the population starts out high and then rapidly diminishes as the algorithm converges. The diversity directly influences the algorithm’s ability to perform effective exploration of the problem space. In most cases if exploration is required in the latter stages of the algorithm, there may be insufficient diversity to allow for this.
This means that, when working with evolutionary algorithms, one of the key factors that must be considered is the balance between exploration and exploitation. If the algorithm focuses on exploration, large portions of the search space will be evaluated without thoroughly checking specific areas.
A focus on exploration will thus be required to find the portions of the search space which contain good candidate solutions.
Conversely, if the algorithm focuses on exploitation, small areas of the search space will be checked thoroughly. The required exploration and exploitation will be dependent on the distribution of good solutions in the search space.
In response, Prof Coulter and Cilliers evaluated the proposed algorithm and compared it to the standard genetic algorithm against current techniques for maintaining exploration. The developed algorithm was able to maintain diversity slightly longer than the standard genetic algorithm and was able to maintain a higher level of diversity after the population diversity stopped diminishing.
The UJ researchers then came to a conclusion that the developed algorithm can effectively make use of the available diversity to adapt to changes in the search space.
Prof Coulter comments: “Maintaining diversity is as important to computational intelligence as it is to society as a whole. This work looks at exploring a variety of biologically inspired approaches to preserving diversity in a pool of possible solutions in the facing of a changing environment.”