Using machine learning to develop your hypothesis

There are many applications for machine learning, but one of the most exciting is using it to create hypothesis to test. An article in the Economist, “Computer says ‘try this’,” discusses many of the ways computers are now creating hypothesis to move medicine, farming and even cooking forward.

There are already many projects using machine learning to generate valuable and novel hypotheses. One project, BrainSCANr, suggests research topics for neuro-scientists by looking at millions of peer-reviewed papers. Research published by Baylor College of Medicine researchers used machine-learning software to review over 150,000 papers on a technique to curb the growth of cancer (proteins called kinases). The algorithm led to seven new kinases that researchers had missed. The technique is even being used to analyze search terms in Bing and Internet Explorer to determine potentially dangerous pairings of medicine.

Machine learning generates strong hypotheses

What it means for you

While machine learning is already benefitting many tech and game companies, using it to help develop hypotheses for your business is invaluable. If you are a mobile game company, think of the value of having a machine-learning system suggest that rather than focusing on your premium exchange rate, you should test changing the frequency of your free-coin bonus. Continue reading

Using machine learning to improve customer interaction

I recently read a very interesting post, How Machine Learning can Improve Customer Interaction, that does a great job of listing different ways you can leverage machine learning to communicate better with your customers. The ideas include:

Machine learning

  • A personalized approach when you visit a website. When you are on an e-commerce site or using a search engine, the host collects rich information on your behavior. Machine learning analyzes the data and transforms the website into something geared to the individual customer. Machine learning then will control what you see, what appears in a search bar, how the site communicates with you, to best meet your individual needs.
  • Making recommendations. Making recommendations relevant for the user was one of the first major consumer applications of machine learning. Virtually everyone has experienced Amazon’s recommendations, when you make a purchase it recommends products likely to resonate (and almost everyone has taken advantage of these recommendations). Automated personalization with machine learning takes information about the shopper, refines those recommendations and tailors them specifically to the individual shopper. As the article points out, “it is like having a salesperson with the customer the whole time, pointing out what products he or she thinks are right up the customer’s alley.”

Continue reading

Pushing the edge of machine learning

As part of a MOOC I just finished (Northwestern’s Content Strategy course, in which there was an interview with IBM’s SVP of the Watson Group, Mike Rhodin) that did a great job of showcasing the potential and future of machine learning. The Watson Group is probably most famous for developing the computer that won on Jeopardy! and it is now deploying that technology to push the boundaries of machine learning.

During the interview, Rhodin explained how Watson can read and understand text. As it can understand text, Watson can learn by reading or obtaining additional information that either will confirm or question a hypothesis. In the latter case, it can then seek out additional information to reach the most likely answer, including looking at historical research and results. It will then give a recommendation and confidence level based on all available information, with the supporting evidence and why the evidence is important. Watson will then analyze whether its recommendations were correct, learn from its mistakes, and effectively get smarter (e.g., why when it played Jeopardy! it got stronger as it completed a column).

Watson wins Jeopardy

Rhodin discussed how Watson is currently helping doctors make more accurate diagnoses. The doctor will tell Watson a patient’s symptoms (Watson can understand spoken English), then Watson will compare these symptoms with what it has read and what is in its knowledge base. It will then narrow down the possible causes and present that information to the doctor. Keep in mind that there is so much research being published daily (and even more historical research) that no doctor can stay on top of all of it. Once Watson has narrowed it down to a few possible causes, it will present these to the doctor with the evidence it generated. The doctor can either do their own research or it may trigger a memory of an article they read in the past. Watson has thus helped the doctor reach a diagnose faster, which often helps recovery rates and reduces treatment costs.

Continue reading

What is machine learning and why it is crucially important

A few weeks ago, I posed a question on Quora about the differences between machine learning and predictive analytics. I was surprised at the number of people who started following the question and actually liked it (although it sounded rhetorical, I was hoping to understand machine learning better). A recent article in Fast Company about The New York Times, of all companies, did a great job of explaining the fundamental value of machine learning.

My interest in machine learning was ignited recently as it has become the hot buzzword in the Bay Area; some have argued if you add machine learning to your PowerPoint you can add a zero to the end your company’s valuation. While that claim is obviously an exaggeration, investors are among the savviest businessmen, so their interest in machine learning shows it is a crucial emerging space.

The article discusses Chris Wiggins, a biologist The New York Times just hired as its Chief Data Scientist. Wiggins’ mandate is to build and lead the Times’ machine learning team. In Fast Company’s interview with Wiggins, it became clear exactly what machine learning is, how it is different than predictive analytics and why it is important.

What is machine learning

According to Wiggins, “Machine learning sits at the intersection of data engineering and mathematical modeling. The thing that makes it different from statistics traditionally, is far more focus on building algorithms.”

University of Utah image for machine learning

Also, while statistics is traditionally focused on explaining data, machine learning is geared to building predictive models. When Netflix or Amazon make product recommendations to you, they are using machine learning to predict what you would be interesting in experiencing. Continue reading