Google Explains Machine Learning And Deep Learning; Plus: Short Takes From Educause 2016

November 2, 2016 Bob Nilsson

 

Machine Learning is an important concept in computer science and for higher education in general that is developing rapidly. Greg Corrado, a senior research scientist at Google, described the ML basics that educators and IT managers in higher education all need to be aware of.

Although machine learning is not entirely new, it has gotten much more attention since last March, when it was used to defeat Lee Sedol, the Go world champion. But even before that, ML has been powering apps like Google photos, speech recognition, text-to-speech converters, and face recognition. The reason it is coming to the forefront now is that the computational resources that it requires have become readily available. This is analogous to powered flight becoming feasible in the Wright brothers’ era only when the mass-to-power ratio of engines achieved the right level.

Whereas early artificial intelligence machines were programmed to be clever, ML-based AI systems learn to be clever. The original spam-filtering programs had to be explicitly programmed. Now they are ML-based and constantly improving.

ML code tends not to be very lengthy. The algorithms require a set of parameters, basically a table of numbers, and a learner program that tunes the parameters by feeding back information on how accurate the model is as it goes through training on the input data. There are four ingredients necessary for ML.

  1. Computational resources
  2. Algorithms and tools
  3. Training examples (in the millions or billions)
  4. Human creativity and ingenuity

Google provides algorithms and tools like tensorflow as open source. Computational resources are available in the form of CPUs, graphics processing units (GPUs), tensor processing units (TPUs), and mobile. The rate limiting step today is human creativity and ingenuity.

Deep learning takes machine learning another level. Instead of analyzing inputs in one step and providing one answer, deep learning breaks the problem down into multiple steps or pieces, similar to building a structure out of Legos. While ML would look at a whole photograph to determine what it is, deep learning finds and analyzes an eye, then an ear, and legs, and further determines how these pieces fit together to conclude it is a cat image, for example.

Google’s ready-made tools and systems are available for courses specifically on ML, or to use in an applied way in any classes.

How Analytics Relate to Student Success

Analytics continues to be a much-discussed topic when education CIOs get together. During the session, Comprehensive Approach to Student Success Using Analytics, the different types and goals of analytics were categorized. Ultimately, we’d like to have metrics that truly predict and measure learning. In the meantime, these are five types of analytics:

  • Classroom engagement metrics and alerts
  • Retention analysis
  • Student risk and persistence scoring
  • Adaptive learning technologies
  • Faculty engagement

Common data required for successful analytics initiatives

The presenters noted the high value, as well as the challenges, in using analytics. Too often anecdotal data and folklore, which is actually non-reflective of the actual learning dynamics, can drive behavior.

Short Takes From Educause 2016

There’s always news and discussion of the latest trends at the annual Educause conference. Here are some things talked about this year among attendees and at the sessions specifically on wireless local area networking and network management.

Many (most?) universities now have 1 GB of bandwidth to the Internet. The highest in the networking session was 20 GB. About two-thirds still provide wired network outlets in the res halls. Even newly-constructed halls are getting wired networks.

Most schools provide eduroam on their campuses and report getting very positive comments from visitors for doing so.  

IBM Watson is getting into education. A company representative talked about making this the “next moon shot after healthcare” for Watson. They are busy merging capabilities with Blackboard and have a significant partnership with Pearson to offer tutoring capability.

Students are getting older. Still plenty of 18-22 year olds. But it was surprising how many discussions began with, “our average student is 35 years old.”

Google is offering educational grants for students and faculty to use their cloud platform in higher education. For now, this is a US-only program for computer science or a related subject area.

 

Educause 2016 attendees queue up to hear about new products from Extreme Networks.

 

Educause 2016 was held in Anaheim, CA.

The post Google Explains Machine Learning And Deep Learning; Plus: Short Takes From Educause 2016 appeared first on Extreme Networks.

Previous Article
Extreme Looks to Crush the Competition and Set Partners Up For Success
Extreme Looks to Crush the Competition and Set Partners Up For Success

At last week’s Extreme Global Partner Summit, I had the privilege of addressing our growing partner ecosyst...

Next Article
Coding, Gaming, IoT, and Wearables: Educause 2016 And The Increasing Presence Of Virtual Reality In Education
Coding, Gaming, IoT, and Wearables: Educause 2016 And The Increasing Presence Of Virtual Reality In Education

Coding, Gaming, IoT, and Wearables: Educause 2016 And The Increasing Presence Of Virtual Reality In Educati...