ARTIFICIAL INTELLIGENCE, MACHINE LEARNING AND DEEP LEARNING IN PERSPECTIVE: WHAT’S ALL THE FUSS ABOUT?
by William E Hamilton@ Iman Fiqrie, Lecturer, CPLP, Ph.D. Candidate
What’s All the Fuss About?
What’s all the fuss about with artificial intelligence (AI), machine learning (ML), and deep learning (DL)? The truth is, ML and DL have been around conceptually for a long time—literally! If you look at figure 1., below one can see, thanks to Moore’s Law, that we’ve come a long way in processing speeds, computing power, and storage capacity since the 1980s.
Deep Learning is Big Data and Intelligence
Make no mistake about it, Deep Learning is about Big Data! Fast forward to 2018 in the aftermath of the “technology tsunami”, the potential intelligence gains by utilizing Big Data and its application to potentially help solve many global and humanitarian issues—and, of course make money. How does it work?
Recommender, Predictor, and Prescription Machines
An organization or potential user gets Big Data, hooks up a machine, and potentially solves big and deep problems with AI, ML, and DL; e.g., recommender, prediction, and prescription systems. When I say “hook up a machine”, make no mistake about it—before any Data Scientist (DS) or person knowledgeable in ML and DL will do such a thing—he or she is going to go through a meticulous process of data cleansing or data preprocessing as it’s called in the community. A sure-fire way to kill the process is require the DS to use certain variables; recommend, supply and support, but if forced to use garbage, you’ll get garbage.
Data Cleansing and Preprocessing
Data pre-processing (DP) involves narrowing down unneeded variables through a process referred to as “Step-wise Regression” and score comparison of significance levels, usually 5% (0.05); dealing with the categorization and manipulation of non-numeric variables, i.e., text variables, and creation of dummy variables (consideration the “dummy variable trap”); importing of several Python or “R” libraries that deal with these concerns; splitting of the DP into a training set and test set for the machine to train on, learn and then predict with; fitting of the multitude of machines to the previously mentioned datasets; and interpretation of the results. If the variable set is let’s say 50 – 100 variables, it will be difficult to explain which variables have an impact on anyone! Therefore, DP is paramount.
Big Data is Compact, Tight, Clean, and to the Point
Long story short, none of these machines are going to help an organization if their mission, vision, and organizational strategy (and by default, data strategy) doesn’t support a rock-solid mission and vision with great data points. What variables give operational excellence, customer intimacy, and product leadership? These would be your dependent variables usually put in the last column of a machine learning scenario dataset (Excel spreadsheet.csv); compact, tight, clean, and to the point.
Anyway, I’ve gone on too long about all the fuss about AI, ML, and DL. Check out my last paper as linked here for the full persuasive argument. I’m sure you’ll find it interesting if you love to hear more about deep learning.
Persuasive Argument for Deep Learning
Please follow and like us: