Blog Viewer

Permalink

TechTalk Blog - Machine Learning - 1st in a Series

By Christopher Mishler posted 02-02-2017 08:50 PM

  

Machine Learning - 1st in a Series

 Since I just got my new “Machine Learning for Dummies” by John P. Mueller and Luca Massaron, you will be treated to a serial book review as I work my way through each chapter.  Why will you be subjected to this regular installment?  Machine Learning (ML) is getting increasing amounts of press as Big Data gets bigger and more imaginative applications of Artificial Intelligence (AI) come to fruition.  Some are rather dull-sounding (“Aren’t we doing that already?”) and others are inching closer to our expectations from watching futuristic movies or reading science fiction.  I got this book because I wanted to see what the fuss was all about and whether it has strong applicability to accountants and our kin in analytics, auditing and operations.  We already can guess or know that the marketing functions are all over the concepts and solutions that pertain under this topic, but what about the rest of the organization?  Should we care, be concerned or be excited?  The answer is Yes to all three options!

Let’s get a couple things straight:

Machine Learning and Artificial Intelligence are not the same thing.  Machine Learning is characterized by the use of algorithms (sets of rules) to manipulate data. 

Artificial Intelligence uses ML as one tool to improve its output from data inputs. Other disciplines include:

  • Natural language processing to take regular language and convert it to a form useable by the computer
  • Natural language understanding is acting on that language based on its meaning
  • Knowledge representation puts information into a structure that speeds retrieval
  • Planning (goal-seeking) ability is needed to draw conclusions from stored data to take almost immediate actions
  • Robotics may be involved in case the output needs to take a physical form

 

One thing users of ML have to be on the hunt for is the right algorithm for the desired output.  You can have a whole box full of tools, but the art/science is in choosing the appropriate technique to answer the business questions, either known or in an exploratory fashion.  A few other early terms to include in the ML milieu are Statistics and Big Data.  A lot of the statistical math involved is intended to compute probability of certain outcomes.  More (bigger) data increases the likelihood that the predictions or outputs produce the most helpful information for decisions – either in an automated system or as part of a more human-based decision model.

 

To wrap up this scintillating introduction, Mueller and Massaron give a few examples of existing AI system reliant on ML and a few potential applications.  Currently available:

  • Fraud detection – as in systems used by credit card companies to find anomalies in the use of a card or credit account. My wife got a call last week about unusual transactions and was able to put a stop to them the same day they were charged through this very type of AI, saving us the hassle of fighting for over $700 in fraudulent purchases.
  • Resource scheduling, such as a hospital might use in caring for a patient, to decide the best location, doctors and tests.
  • Complex analysis as found in diagnosing tricky illnesses, since the number of factors involved could be staggering. In the last year, studies have been conducted on drug interaction AI applications for reducing harmful drug interactions (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4904823/)
  • Better Automation may come from AI in allowing the process to address unexpected events that would otherwise bring the system crashing to a stop or hurt someone.
  • Customer service – who wouldn’t want an improvement in the online and phone systems with which we interact to solve our consumer complaints?
  • Safety first – one example is the improvement in automotive systems such as automatic braking systems that can take more factors into account to improve safety.
  • Machine efficiency saves money through AI optimizing the speed at which a system runs, dealing with constraints and minimizing inputs.

 

Think about some of these possible applications too, our authors list:

  • Access control – like smartcards which know our roles and therefore whether a certain resource or room should be accessible
  • Animal protection can be improved in the ocean by informing vessels of animals likely to cross their path in a harmful way by analyzing the sounds and behavior patterns.
  • Those wait time predictions for service can get more accurate through AI taking more factors into consideration, such as staff availability, loads, complexity of problems in the queue ahead of us.

 

Next posting – Machine Learning in the Big Data Age

(Isn’t this fun?)

 

0 comments
109 views