How Biomimicry Can Improve Data Science and Machine Learning Projects
Nature-inspired algorithms can go a long way to improve ML/DS
In Janine Benyus’s book Biomimicry, she describes biomimicry as “the conscious emulation of life’s genius. Innovation inspired by nature.”
The book goes to great lengths to highlight a prominent fact that is so prevalent in our daily lives that we probably do not notice. We live in an environment that is awash with the creatures and environments that have emerged as a result of nature refining things for millions of years.
Looking to nature as an inspiration is a necessity in the world of engineering as nature has already solved many of the problems that we face. A famous example of biomimicry in design comes from Japan’s Shinkansen (bullet train). Every time it went through a tunnel, it made a loud sonic boom that disturbed both wildlife and people. The engineer responsible for solving this was able to deal with this by imitating the aerodynamic design of a kingfisher.
In computer science, the most prominent example of biomimicry or a good attempt at it is the use of neural networks. Neural networks are designed to attempt to emulate the function of a biological neuron so as to solve a relevant problem in any given field.
Sadly, more examples of biomimicry in artificial intelligence and data science may not be as prominent as neural networks, but they have the potential to be just as impactful.
Nature-Inspired Algorithms
There are numerous nature-inspired algorithms and their use cases only increase with each passing day. The rationale is that these sophisticated methods could perform desired activities efficiently.
For the reason stated above, researchers have investigated multiple natural phenomena and they have come up with curious insights. The algorithms originate from diverse sources such as fluid dynamics (water evaporation optimization) to biology. But our focus here is on bio-inspired algorithms.
A few interesting bio-inspired algorithms and their use cases include:
The Combined Artificial Bee Colony algorithm is a composition of machine learning algorithms with an optimization algorithm called artificial bee colony. The Artificial Bee Colony is an algorithm inspired by the interactions and operation of a beehive in a colony. It uses employed bees, onlookers, and scouts just as they work in real life. This algorithm thrives as an optimization algorithm as it more or less represents a natural process for bees to identify food sources in an efficient manner.
The mathematical representation of this algorithm, including some of the artificial limitations introduced, paved the way for a refined process that allows for great optimization capabilities when combined with other algorithms.
A good example of using this is highlighted in the paper titled “Combined Artificial Bee Colony Algorithm and Machine Learning Techniques for Prediction of Online Consumer Repurchase Intention.”
There are multiple machine learning problems where the Artificial Bee Colony has found widespread usage. One example is improving the quality of clustering algorithms as demonstrated in this paper. The Artificial Bee Colony has also found use in deep learning as it can be used in training artificial neural networks as an alternative for back-propagation as envisioned in this paper.
Comet Artifacts lets you track and reproduce complex multi-experiment scenarios, reuse data points, and easily iterate on datasets. Read this quick overview of Artifacts to explore all that it can do.
The Firefly Algorithm has found extensive use and has demonstrated incredible use in clustering and can be used to train neural networks. This algorithm is defined as a “metaheuristic algorithm” that attempts to emulate the behavior of a firefly. It is specifically concerned with the flashing behavior of fireflies that are used as mating signals. This algorithm, like the previous one, can be modified to perform interesting optimization problems. When it is combined with other algorithms, it performs optimally in multiple examples as it is efficient.
This paper entitled “Firefly Algorithm: Recent advances and applications” demonstrates that the use of this algorithm is not limited and is only growing in utility.
This algorithm, when combined with neural networks, can be used for time series prediction problems such as in this paper. The firefly algorithm has also elegantly been applied to improve K-means clustering.
Another algorithm that has found extensive use cases is the Flower Pollination Algorithm. This algorithm is named after the process that inspires it; flower pollination. It is similar to the other algorithms discussed as it is reduced to a few rules that can be mathematically interpreted. These rules take direct motivation from the process of pollination that a flower undergoes. On top of these, the most important fact is that this algorithm has proven to be highly computationally efficient and has found use in many engineering fields.
Two use cases that are highly relevant to machine learning are solving optimization problems and data clustering. This paper entitled “Flower pollination algorithm parameters tuning” gives multiple examples of how this remarkable algorithm is used.
Reasons for Further Development
The few examples I have mentioned demonstrate a few things about bio-inspired algorithms. Firstly, these algorithms are not needlessly complex and difficult to grasp. Second, the inherent property of not being complex allows them to be highly computationally efficient and in some domains, extremely dominant.
It is also an understatement to mention that they can and will find use in other domains of engineering. There is a good likelihood that the properties present in algorithms used for optimization could prove to be highly useful in many other domains.
By investigating the inner workings of the natural world, it has become evident that refined algorithms could no longer be taken for granted. As data science and machine learning are heavily reliant on highly efficient methods of optimization with an emphasis on using the least amount of computational resources, the need to further develop more algorithms becomes increasingly relevant.
Editor’s Note: Heartbeat is a contributor-driven online publication and community dedicated to providing premier educational resources for data science, machine learning, and deep learning practitioners. We’re committed to supporting and inspiring developers and engineers from all walks of life.
Editorially independent, Heartbeat is sponsored and published by Comet, an MLOps platform that enables data scientists & ML teams to track, compare, explain, & optimize their experiments. We pay our contributors, and we don’t sell ads.
If you’d like to contribute, head on over to our call for contributors. You can also sign up to receive our weekly newsletter (Deep Learning Weekly), check out the Comet blog, join us on Slack, and follow Comet on Twitter and LinkedIn for resources, events, and much more that will help you build better ML models, faster.
