A Tale of Two Feature Engineers: Machine Learning vs Deep Learning
Both Machine Learning and Deep Learning summarize a methodology for the automatic detection of patterns in data and their use to predict certain aspects of novel data. They are concerned with the approximate mapping between inputs and outputs since there are no formulas (at least until today) that dictate the reality of the countless problems to be solved. Moreover, logistic regression can be thought of as a 1-layer neural network — and from this contact, it is possible to turn the neck to the left and to the right to see the immense landscape of each of these umbrella terms.
One way to distinguish Machine Learning from Deep Learning is through the way raw data is worked on to develop a solution. Raw data, structured or not, is composed of a set of initial features. In Machine Learning, the developer is the main agent for defining the best set of features to be used (within the constraints of the problem at hand). It is, therefore, a person who, iteratively, will try to use different combinations of features, create new ones, test interaction features and also reduce the amount to be used, for example. On the other hand, in Deep Learning, the chosen architecture is the main agent for defining the optimized set of features to be used. By providing raw data, the algorithm can extract features automatically, without the need for a person to specify them.
Author – João