Machine Learning

The subset of AI based on inductive approach

Something really important to notice is that ML is not for everything:

• If we have a known function, even really complex, or a method or algorithm (even complex) that can give us the result from the input, don't use ML. Why if you already have a solution?

• You must have a good dataset. Good in quality and quantity. If not, don't believe in miracles

• There must be a pattern in the input data in order to get the output, even if this pattern is highly complex, but there must be a pattern. If not, sorry, no ML technique will help you.

• ML is also a good strategy for solving large search space problems

• ML techniques also allows us to develop highly adaptive and dynamic solutions (just working with new data and/or adding new features, etc.)

• One of the most incredible things we are living in AI is that, using some ML techniques, we are discovering unpredictable patterns in input data, opening a new world of possibilities and beaten human beings.

From a deductive approach to an inductive one

ML changes how we solve problems. No deductive approach, an inductive one. From examples to rules, then the generated rules will be able to solve new instances of the problem:

When the model is created, then it's time to make predictions:

By Machine Learning (ML) we understand a family of algorithms and techniques that try to learn automatically from the data provided

Its objective it's to generalize behaviors from data provided as example

It's a process of knowledge induction

ML has also many problems. One is interpretability. But there are others such as bias.