Table of Content

  1. Machine Learning
  2. Computational Intelligence
  3. Genetic Algorithms inspired by DNA Sequence
  4. Swarm Intelligence Algorithms inspired by Crowd Intelligence
  5. Firefly Algorithm inspired by Fireflies
  6. Cuckoo Search Algorithm inspired by Cuckoo bird
  7. Flower Pollination Algorithm inspired by Pollination behavior of Flower Species


We traditionally start our machine learning journey from statistical modelling techniques like linear regression or logistic regression. Next we get familiarized with computationally intensive algorithms like decision trees, random forest, bagging, boosting and support vector machines. In the prior part the emphasis is on statistical significance, hypothesis testing, goodness of fit. In the later part as we move towards machine learning we leave the terminologies as stated in the prior case.

The reason for this shift as we move from statistical learning to machine learning is because there is a premise shift. Its a gray area, regression techniques can be considered as statistical learning as well as machine learning depending upon how you use it. Similary there are some gray areas as we move from Machine Learning to Computational Intelligence. Rather than discussing the similarities and dissimilarities it makes more sense to understand what they are.


Machine Learning

Machine learning would encompass techniques used to train an existing known algorithm with input and output data. This would give us an abstraction of the learning. Most of these are specific to the learning data. Sometimes the equation or model as we call it is generalized. But for most of our data in real world these are not generalized. For instance e=mc^2 is a generalized equation. This holds true even if the location, state, pressure and other variables change. In contrast, an abstracted model or equation from the business data we feed to a machine might not be generalized. For example, using ML you may create a classification model which can identify credit defaulters. But this model may work well for one banks data but not for a NBFC. Consider another example, a vaccine developed has to be tested for different populations, genetic pool, demographics and location. The reason being that if its efficacy is good for one population but not for other, then it cannot be globally distributed.


Computational Intelligence

Computational Intelligence aims at replicating mechanics of data and information processing of humans. Below is an infographic representation of the objectives of Computational Intelligence.

Computational Intelligence

Click the image to download High Resolution Image of Computation Intelligence  Infographic

Humans have evolved from the learning perspective. Much of this is now being documented and used in Computation Intelligence. Competitive intelligence takes inspiration from human capabilities of sensing, learning, recognizing, thinking and understanding. Sensing relates to how different mechanisms work parallel to each other. Right from the skin, eyes to the hair in our ears have capabilities to pass the data from one form to another. Sensors play a very important part of AI.

From sensing to learning the patterns in data is also an inspiration in Computational Intelligence. Evolutionary Computation which is a part of Computational Intelligence follows principles of natural evolution such as Crossover, Mutation, Selection and Reproduction. Evolutionary Computation Algorithms such as Genetic Programming, Genetic Algorithms, Grammatical Evolutionand Evolutionary Algorithms are the result of same. Data Fusion Algorithms are also classified as important aspect of Computational Intelligence. The ANNs, CNNs, RNNs and the likes of many algorithms like these are somewhere the results of Computational Intelligence. Although genetic algorithms are not so closely followed by many these days, they also are a part of Computational Intelligence.

The field of Computational Intelligence has led to a host of Nature inspired algorithms. In some other blog in the future, I would cover them in detail. For now, let me quote a few amongst them with a brief understanding for further exploration.


Genetic Algorithms inspired by DNA Sequence

When we are born, we inherit a lot of features from our parents. Sometime it's the father whose features are more prominent in the offspring and sometimes it's the mother. In few cases, we observe an indirect reference of similarity is found from grandparents. The essence over here is that the eye color, the full-grown adult height, skin complexion and host of other features are inherited. Genetic Algorithms as a class of Evolutionary Alogorithms which are Nature inspired use this information decoded in genetic sequence as the building block.


Swarm Intelligence Algorithms inspired by Crowd Intelligence

Nature inspired swarm behavior is prevalent in most of the species. To highlight a few, consider bee's, ant's, bat's, fishes, and migrating animals. Be it for food, safety, hunting or any natural needs, Nature shows how the collective intelligence of swarm or crowd is superior to individuals. Bat Algorithm and Bee Algorithm would be classified under Swarm Intelligence Algorithms.


Firefly Algorithm inspired by Fireflies

Fireflies are unisex and decide on their mate depending on the best fit optimization. They flash their lights and based on the intensity judge the fitness of the mate for healthy offspring's. This is used as an Optimization Algorithm.


Cuckoo Search Algorithm inspired by Cuckoo bird

Certain species of Cuckoo birds are known to lay their eggs in host birds nest with a random probabilistic approach. The host may realize this at a certain point or may not. Based on this probabilistic approach the Cuckoo mimics certain biological traits like size and color relative to the host which is predefined. Cuckoo Search has application in Operation Research.


Flower Pollination Algorithm inspired by Pollination behavior of Flower Species

Natural dispersion of pollen within a certain geographic area is used as Optimization Solution. Flower Pollination Algorithms have application in Optimization problems with multiple objective function.


There are more than this list of metaphor based approaches in Computational Intelligence. Not all are used exhaustively but some of them are used widely. The performance in terms of accuracy would change depending on the environment and used case.

For now, I will conclude this at a summary level. A deeper understanding of each approach would need further research. You can connect with me on LinkedIn.


Authored by : Mohan Rai