Whenever we have the ground truth data, we can make use of Machine learning in these kinds of scenarios but there is actually a very little obvious correlation of elements that produce a ground truth. We can take the example of flowers. For a normal human these are just the normal pictures of flowers. But let’s say you want to feed these pixels to a program that can distinguish these pictures from this one.
To a normal person, a picture is not a flower obviously. Therefore it is possible to write a program to tell the flowers from the umbrella because you are able to do it. But how will you do it? What kind of algorithm will you use.? There are plenty of features in a graphics image, and so many abstract concepts like a line vs. a curve that coding a fixed algorithm for it would be problematic. Therefore the main idea is that you build a huge database of images. You flag the ones that are flowers as being flowers, and those that don’t have flowers in them as not being flowers. Then you save it in a machine learning algorithm and let it decide how to tell them apart. It will definitely create a lot of data and a long long time.
So at what point does machine learning not make sense? Whenever you are available with very little data and ground truth. If you don’t have enough cases for the program to train on, it is unlikely that it will be able to generalize the features needed to be successful.
There are plenty of other problems whose features are simple and obvious. Making a camera that can look at a fixed table in which a two-color coin is being flipped on and identifying what colour is face up is so simple that you probably wouldn’t involve a machine learning program to deal with it.
So these are some of the instances where we can and we cannot use Machine Learning.
If you want to become an expert in Machine Learning, check out this Machine Learning Certification Course offered by Intellipaat.
See this Machine Learning Course for more information :
Back