Introduction to Machine Learning

Understanding the Power of Machine Learning some background context

“Machine Learning: Living in the Age of AI,” below investigates how people increasingly interacti with AI. Hobbyists and teens are now developing tech powered by machine learning and WIRED shows the impacts of AI on schoolchildren and farmers and senior citizens, as well as looking at the implications that rapidly accelerating technology can have. The film was directed by filmmaker Chris Cannucciari, produced by WIRED, and supported by McCann Worldgroup. 

Introducing Machine Learning concepts and motivation

Machine learning is an artificial intelligence (AI) tool used to craft applications that learn from data and improve their predictive accuracy through repeated usage and refinement.

Machine learning algorithms are 'trained' to detect patterns and features in expansive datasets in order to implement practical classifications and predictions based on new data. As the algorithm is refined,  greater accuracy can be harnessed and predictions improved. More data generally produces better algorithms and better patches to those algorithms iteratively. As a general principle, more training data leads to better performance. Tesla might just outperform rivals, not because it started with best algorithms or ideas but because it notoriety translates into more mileage driven by the faithful feeding algorithms the oxygen of data.  In the machine learned world,  first mover advantage generates a hugely positive spiral for those who launch innovation early even if mistakes are made along the way. Perhaps, this maybe why China created a digital firewall to give Chinese companies some headspace in sectors that otherwise would be dominated by silicon valley.  India's pathway looks different but scale counts.

Localizing Machine Learning in the R and Python spectrum

For those trying to break in to key concepts - we recommend Hands-on Machine Learning with R. The authors:  Bradley Boehmke & Brandon Greenwell  describe Machine Learning techniques in a way that helps develop intuition for strengths and weaknesses/limitations. Their RStudio Cloud project provides a useful testing ground for experimenters who want to get their hands dirty. For the most part, they minimize mathematical complexity when possible but also provide resources to get deeper into the details if desired. Google Developers explain some emergent powerful tools for modelling processes and behaviors. 

Putting Machine Learning Tools to work:  a second look at the titanic dataset

Here we follow a few examples from Hal Varian. Classic examples of machine learning at work. include  spam classification in email or digital persona who judiciously trawl the web and play music in response to our voice commands. Websites intelligently recommend products/movies/songs based on what the digital trail we leave behind. In the medical field,  images can be processed to spot tumors that the naked eye may have missed. A large part our future existence will be colored by these algorithms and those in a position to hone them with deeper datasets are set to dominate the digital age. Below, we apply insights garnered by Hal Varian, in part, to explain the potential of machine learning and reveal how data can be parsed through intelligently with simple commands. No formal  is really required here. To get your hands dirty with R code, please double click on Google Colab icon on the next page. 

Introduction to Machine Learning for Classification problems

The Machine Learning exercise as implemented above using the titanic3 dataset can be viewed as a classification problem.  A more mundane example would be how to classify emails as being “spam” or “not spam” based on the signature characteristics of the email. This approach, as your own experience might prove, is not infallible. But neither are human beings.  Statisticians typically use a range of tools including qualitative choice models like a logit or probit for dealing with classification problems.  In the video clip above, a decision tree was teased out using the approach described by Hal Varian . These models can be regarded as sequence of decisions stemming from branches into leaves of terminal nodes that culminate in a range of outcomes. A tree classifier has the same general form, but the decision at the end of the process is a choice about how to classify the observation. The goal ultimately is to  “grow” a decision tree that produces good out-of-sample predictions or in more common parlance ensures that the important emails rarely slip over into the spam. The decision tree method is a powerful and popular predictive machine learning technique that is used for both classification and regression. So, it is also known as Classification and Regression Trees (CART). 

Why Machine Learning?

With some EDA, we can draw some straightforward and unambiguous conclusions— being a woman, being in 1st class, and being a child all conspire to boost your chances of survival. We could write a program to predict whether a given passenger would survive the titanic or not. This could be be realized be deploying an elaborate web of nested if-else statements with some sort of weighted scoring system. In an evolving non-static dataset or population - a learning tool fits better. Machine learning is a better fit and we might consider some tools proposed by the google developers to help you embrace these techniques:

Machine Learning Foundations

Please see playlist here.