WTF is Machine Learning Anyway?

In a world where we might think is being ruled and controlled by tech geeks and data scientists, during meetings and phone calls with customers I’m still, often, being hit with honest and candid questions about any given topic about the data and analytics and give my personal take on them.

In virtue of this, I’ve decided to take a shot and a series of posts to answer, as plainly as I possibly can, common questions I receive in my day-to-day life as a consultant and analyst.

Starting with my most popular question nowadays: WTF is machine learning?
So, here we go...

Machine Learning in a Tiny Nutshell

The discipline of  machine learning evolved as part of larger disciplines including data mining and artificial intelligence (AI) and, in many ways, evolving side by side with traditional statistics and data mining and other mathematical disciplines.

So, simply put, machine learning cares about the development of mathematical models and algorithms with the ability to “learn” from data input, adapt  and subsequently, improve the outcome. The concept of "learning" in machine learning, yet far from simple in practice, starts with a simple definition:

• Learning​ ​=​ ​Representation​ ​+​ ​Evaluation​ ​+​ ​Optimization

In which:

• Representation​ is a data element called classifier represented in a formal language that a computer can handle and interpret
• Evaluation​ consists of a function that distinguish or evaluate the good and bad classifiers; and
• Optimization​ which represents the method to be used to search among these classifiers within the language to find the highest scoring ones

From the previous idea, machine learning can be done by applying specific learning strategies, including:

• Supervised strategy or learning, to map the data inputs and model them against desired outputs
• Unsupervised strategy or learning, to map the inputs and model them to find new trends

Of course, derivative ones that combine these have appeared, such is the case for the combined semi-supervised learning strategy and others. Opening the door onto a multitude of new approaches to machine learning and the incorporation of diverse data analysis disciplines to its arsenal, such is the case for predictive analytics as well as pattern recognition.
(post-ads)
As approaches and algorithms emerge, they have been frequently organized in taxonomies and classified after different criteria, including the type of input, and output, required and its use in different situations and use case scenarios.

Some of these approaches include (in alphabetical order):

• Association rule learning
• Artificial neural networks
• Deep learning
• Bayesian networks
• Clustering
• Decision tree learning
• Genetic algorithms
• Inductive logic programming
• Reinforcement learning
• Representation learning
• Rule-based machine learning
• Learning classifier systems
• Similarity and metric learning
• Sparse dictionary learning
• Support vector machines

(post-ads)
Then, What is a Machine Learning Software Solution?

A perfect combination of factors, like the evolution of machine learning approaches and algorithms, as well as the continuous improvement in software and hardware technologies have enabled machine learning software to be applied for solving more types of problems and being adopted in increasingly number of business processes.

In essence, an machine learning software solution is simply a software piece ingrained with specific machine learning functional features aimed to solve both specific or general issues where machine learning is applicable so, we can see machine learning software evolving in two main ways:

• As standalone applications supporting machine learning functionality and other advanced analytics approaches, this is the case for solutions such in the likes of: Ayasdi, BigML, Dataiku, DataRobot, Emcien, Skytree and others.
• Evolved from open source based initiatives, key for its mainstream adoption and implementation, these initiatives include projects including: R, Hadoop Mahout or Spark MLlib
• As part of a big software provider software portfolio like: Amazon’s AWS Machine Learning Services, IBM Data Science Experience, Microsoft Azure Machine Learning Studio, SAP Leonardo Machine Learning Foundation, SAS Visual Data Mining and Machine Learning.
• As software embedded within larger enterprise solutions to enhance or transform the way they perform its original functions, some examples include: BMC’s Cognitive Service Management, Infor's GT Nexus Commerce Network or Oracle’s New Autonomous Data Warehouse Cloud.

So, today its likely that we, as information workers or as common users of a given software are in one way or the other, consuming software resources which actually use some form of machine learning technique.

Then, How Can I Use Machine Learning in My Organization?

As the adoption of machine learning increases, so does the use cases, a brief list describes some uses of machine learning applied in different industries and lines of businesses:

• Recommendation systems. Probably its most common use case for, machine learning algorithms are deployed to analyze the online activity of an organization’s customer base to determine individual and/or collective buy or choosing preferences, enabling the system to increasingly learn about customer’s behavior to increase the system’s prediction accuracy. Companies including Amazon, Netflix or BestBuy
• Marketing Personalization. Today, some organizations apply machine learning techniques to learn and understand better its customers and consequently to improve its marketing campaigns. From learning customers behavior, organizations can personalize, for example, which email campaigns a customer must receive and/or which direct mailings or coupons, or offerings that will likely have more impact if showed “recommended”.
• Fraud Detection. Companies like Paypal are now using Machine Learning software solutions that analyze all their transactions, learn and identify fraudulent transactions from legitimate ones while increasing accuracy over time.

These of course are just a couple of examples of a wide set of uses cases in different industries including, healthcare, data security, healthcare and many others.

So...?

On one hand, today it is not hard to find use cases for machine learning, and it keeps growing, so if you are looking into adopting a machine learning solution, there is a good chance you will find one that fits your current needs for improving your organization’s analysis capabilities. Also, given it is possible to find many types of machine learning solutions in the market, both commercial and open source, it might not be cost prohibited to embark at least in the evaluation of some of these available options to get a sense of the benefits of having machine learning capabilities within your organization.

On the other, it is important to note, as with any other type of software, you will need to do the legwork and ensemble a coherent approach for the adoption of a machine learning initiative for your organization to get the best of a machine learning initiative, including a clear definition, scoping and evaluation of your actual needs that will help you define the best solution of choice in the market.

Small advise, don’t look for a vanilla solution, look for the most convenient for your organization.

You can find another example (pun intended) of the use ef machine learning and other technologies on Google’s latest product: The bad joke detector.

Finally, you are welcomed to leave a comment in the box below or download our very first DoT Industry Note report here.