Roundup 11/9/2016

Machine learning (ML) and deep learning (DL) content from the past 24 hours.

Fundamentals

— On the Algorithmia blog, Charlie Crawford writes an excellent introduction to deep learning. He clearly understands the relationships among ML/DL/AI, presents a good intro to neural networks and lists some of the top DL frameworks. My only criticism is that he omits Microsoft’s CNTK/Cognitive Toolkit from the list. (h/t Matt Kiser)

— Kristin Harris lists five ways the machine learning will turbocharge your workforce. It’s a pretty good list:

  1. Automation lifts productivity
  2. Machines eliminate human error
  3. Leave monotony to the machines
  4. Create self-serving economy
  5. Fill specialist roles with robots

Michael Guta asks: what is machine learning? His answer is, well, unique.

Issues

— In TechCrunch, Ben Dickson tries to explain why it’s hard to create unbiased artificial intelligence. He muddles the analysis.

Methods and Techniques

— In Harvard Business Review, James Hodson offers four tips for “making your company machine learning ready.” Step #1: “catalog your business processes.” In other words, hire the sort of consultants who write articles for HBR.

Software

— Cloudera commissions a Spark survey. Lo and behold, the survey finds that most respondents sourced Spark from Cloudera.

Hardware

— Intel publishes an advertorial promoting optimized Caffe for deep learning on Intel Xeon Phi Processors.

Applications

— On the Google Research blog, Malay Haldar et. al. explain how they built a deep neural network to better understand searches on Google Play.

— James Vincent asks: can deep learning help solve lip reading? Researchers at Oxford University’s AI lab publish a paper that documents LipNet, software that does just that.

— Researchers at Cincinnati Children’s announce that they used machine learning analysis of spoken or written words to classify patients as suicidal, mentally ill but not suicidal or both. The researchers claim 93 percent accuracy in correctly classifying suicidal patients. A freelance writer goes gaga over this claim. But keep in mind that a simple rule that says “all patients are suicidal” will correctly classify 100 percent of the true suicides; the acid test is how many false positives the method produces.

— Fangping Wan et. al. use deep learning to identify compound-protein interactions in computer simulations to understand how drugs work.

— Facebook demonstrates the use of Caffe2go deep learning for video style transfer on Android and iOS devices.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s