What’s the Next Big Thing in Artificial Intelligence?
“But this technology has been around for years now!” That’s the refrain I commonly hear these days when talking to more seasoned AI practitioners about the resurgence of AI and machine learning.
The origins of many of the scaled machine learning techniques in vogue today can be traced all the way back to the advent of computing in the late forties.
Breakthroughs in scaling Artificial Neural Networks (ANNs) have led to a resurgence of interest and much wider application of these techniques, now better known as Deep Learning.
So, what other technologies are primed for a resurgence?
How about Evolutionary Algorithms?
Like ANNs, Evolutionary Algorithms (EAs) are techniques inspired by models of biological intelligence, applied to real-world problem solving. In the case of EAs, the approach is population-based.
Here’s how it works: A model of the desired outcome is used for producing candidate solutions otherwise known as genes. For example, let’s say we are given ten numbers and the task is to order them from smallest to largest.
The model for the desired outcome is a list of size ten that includes all ten numbers. In other words, as long as we have a full list of ten numbers, in any particular order they may be, we have a candidate.
The system then generates a population of such candidates, all of which comply with the model, but that differ with one another. To generate this population, the easiest thing to do is to throw dice, and generate the candidates randomly (step 1).
It is likely that none of the candidates generated in the first go round are our solution, because none are ordered from smallest to largest. However, some are closer to the solution than the others. In other words, some seem to be more ordered. —> Read More