Serendipity35

When I first think of federated learning, what comes to mind is something like a college federated department. For example, the history faculty at NJIT and Rutgers University-Newark are joined in a single federated department offering an integrated curriculum and joint undergraduate and graduate degree programs.

Having worked at NJIT, it made sense to combine the two departments and collaborate. Each had its own specialties but they were stronger together.

In technology, a federation is a group of computing or network providers agreeing upon standards of operation in a collective fashion, such as two distinct, formally disconnected, telecommunications networks that may have different internal structures.

There is also federated learning which sounds like something those two history departments are doing, but it is not. This federated learning is the decentralized form of machine learning (ML).

In machine learning, data that is aggregated from several edge devices (like mobile phones, laptops, etc.) is brought together to a centralized server.  The main objective is to provide privacy-by-design because, in federated learning, a central server just coordinates with local clients to aggregate the model's updates without requiring the actual data (i.e., zero-touch).

I'm not going to go very deep here about things like the three categories (Horizontal federated learning, vertical federated learning, and federated transfer learning). As an example, consider federated learning at Google where it is used to improve models on devices without sending users' raw data to Google servers.

For people using something like Google Assistant, privacy is a concern. Using federated learning to improve “Hey Google,” your voice and audio data stay private while Google Assistant uses it.

Federated learning trains an algorithm across the multiple decentralized edge devices (such as your phone) or servers that have local data samples, without exchanging them. Compare this to traditional centralized machine learning techniques where all the local datasets are uploaded to one server.

So, though federated learning is about training ML to be efficient, it is also about data privacy, data security, data access rights and access to heterogeneous data.

MORE at analyticsvidhya.com...federated-learning-a-beginners-guide
 

A few years ago I was writing about how a lot of people were looking to save money on their TV entertainment by what was known as "cutting the cord" since it meant disconnecting from a cable service. Those services had boomed in the 1970s and 80s and had brought clear channels from local and distant services and led to the rise of services like HBO and Showtime. People are still cord-cutting, but things have changed.

We tired of $100+ per month channel bundles that included lots of channels we never watched. People wanted a cafeteria-style choice. Just pick the things you wanted. But cable companies didn't want to offer that. So, people began to drop their cable contract and replace it with streaming TV services and perhaps a TV antenna or device that offered local channels, news, and a kind of all-in-one bundle.

In 2015, I wrote about a group of people that I called "The Disconnected" and I did a presentation on how we might connect to the disconnected. The disconnections ranged from cord-cutting to ownership of things (home, cars, physical media) and possibly from education and the world. Since then, I have added other disconnected aspects of our lives.

The pandemic that forced disconnections in early 2020 has accelerated some of that. Ironically, as disconnected as we became to friends, offices, campuses and stores, most of us became more connected to media.

Cord-cutters still needed an Internet service and that connection became quite critical in these pandemic times. We needed it to continue working, learning and staying in touch with other people. Those connections are very important, but I also have been thinking about how connected we have become to those streaming services on our screens for entertainment.

The tech divide either got wider the past year or minimally became more obvious. Home Internet speeds should be at least 15Mbps (megabits per second) for each device you plan to have running at the same time. That means that those two TVs, the laptop and three smartphones and one tablet all playing at once would ideally have a connection of at least 105Mbps. That’s a lot to ask of a DSL or satellite service and from most cable company broadband services. Those people with access to fiber broadband or some other fast connection had a big advantage.

It is now almost a decade from dropping your cable connection and moving to streaming and now I am hearing more people complain about the cost of buying all the services needed to keep up with the content that all your friends are telling are essential viewing. 

What is the cost of having Netflix, HBO Max, Disney Plus, Hulu, Amazon Prime, Peacock, and others and also a bundle of live TV channels such as YouTube TV or Sling TV?

Yes, there are a bunch of free (ad-based) sources of streaming video too (Crackle, IMDb, Kanopy, Peacock, Hoopla, Pluto TV, the Roku Channel, Tubi TV, Vudu, etc.). 

You might also want a streaming device that connects to the Internet and allows you to show things on devices on bigger screens (Chromecast, Roku’s Streaming Stick, or Amazon’s Fire TV Stick. 

At one time, I could watch Disney films on Netflix, but Disney and most of the other content providers have now decided that they are better off offering their content on their own services. YouTube TV recently was removed from Roku. Battles will continue.

If you cut the cord, will you soon need to cut or narrow the streams flowing into your home?

I have been writing a lot about MOOCs since 2012. (Do I still need to explain that a MOOC is a Massive Open Online Course?) That was (as dubbed by The New York Times) the “year of the MOOC.” 

This year, the Times was saying that though MOOCs were "near-death" the COVID-19 crisis has put them back into the "trending" category. Their article is headlined "Remember the MOOCs? After Near-Death, They’re Booming."

Though MOOCs existed prior to 2012, the emergence of online learning networks was something new. While many colleges initially viewed these free online courses as a threat to their tuition systems, within a year many of the most elite colleges began to offer them. It was more than "if you can't beat the, join them." Schools, faculty and students (often on their own) discovered the value of not only MOOCs but online learning in general.

The Times article is negative on the impact of that MOOC revolution saying that "the reality didn’t live up to the dizzying hype." I agree that the hype was truly hype. It was too much. My wife and I wrote a chapter for the book Macro-Level Learning through Massive Open Online Courses (MOOCs): Strategies and Predictions for the Future and we titled it "Evolution and Revolution." The title was not meant as a question. Much of the discussion in 2012 was about the revolutionary nature of MOOCs, but we viewed them through the lens of 2015 and saw them as more evolutionary.  

Fast forward to 2020 - the "year of the pandemic" - and we see schools from kindergarten to graduate schools forced to use online learning in some way. A revolution? No. Again, an evolution that should have started for schools a decade ago but clearly has not for many of them who fond themselves unprepared in march 2020 to go fully online.

MOOCs have changed. My many posts here have shown that the open part of mOoc has become far less open both in the ability to reuse the materials and in the no-cost aspect. Companies have been formed around offering MOOC-like courses, certificates and degrees. 

The biggest criticism of MOOCs was probably that most learners (not always traditional students) never completed the courses. Completion rates in free courses of about 10% certainly sounded like a failure. Making students pay even a small fee or offering credit improved that percentage but not enough to make observers feel the revolution had succeeded.

I never worried about the completion rates because my research and my own experiences teaching and as a learner in these courses made it clear than the majority of students in them never intended to complete all of the coursework. They were there to get what they wanted to learn and get out. They didn't need to take a freshman year of requirements and prerequisites or gain admission to Stanford in order to take a course on artificial intelligence from Stanford. 

Of course, as the Times article points out, MOOCs kept going without all the hype. They evolved, and in some ways so did online learning because of them. Platforms and for-profit companies emerged and certificates, fully online MOOCish degrees, and nanodegress were offered. 

With the spotlight off them, MOOCs were able to evolve into different species - free, for-profit, accredited, for lifelong learning, massive, small, skills training, corporate, for K-12, etc. 

Sheltering and working and learning from home has given another boost to that second "O" in moOc. The providers like Coursera have signed up 10 million new users since mid-March, and edX and Udacity have seen similar surges. And that doesn't even take into account the less-visible use of big (such as Khan Academy) and small grassroots use of these courses by teachers and students.

My wife and I are now writing a journal article for this fall about online learning as a solution for some crises in higher education. 2020 has definitely a time of both crisis and opportunity for online learning. I hope the hype doesn't return to the MOOC. It did not serve it well in the past.

If you have any thoughts on the current state of MOOCs and online learning, contact me.

Serendipity35 crossed over the 14 year anniversary on February 2 this year as people were predicting the American football Supr Bowl and predicting the remaining weeks of winter based on a groundhog or the traditions of Candlemas. Usually, I look back on the previous year on this blog's anniversary, but this year I decided to look ahead.

Looking ahead and making predictions is a December and January tradition. I feel like most of the time those predictions don't come true, but we often don't look back to check them. In education and especially in technology, it's hard to predict what is coming in the year ahead. That is why I had to look at an article I saw that was titled "Five Predictions About U.S. Higher Education In 2040."

2040? It's hard enough to see ahead to the end of 2020. So the author of that piece, Sally Blount, is either crazy, has Nostradamus DNA, realizes that no one is going to check back on her predictions in 20 years, or she has analyzed real data.

She is a contributor to Forbes, former dean of New York University’s Stern Undergraduate College of Business and seems to specialize now in careers. She analyzed the data and marketplace of the U.S. post-secondary education along with Larry Shulman, senior partner emeritus at the Boston Consulting Group and they came up with 5 "not-so-crazy predictions" about how the U.S. market for four-year bachelor’s degrees will likely look in 20 years.

One prediction is that half of those college degrees will be awarded to students who have spent three years or less on a college campus. How will that happen? More courses per semester? Trimester years rather than bimester years? No. It's based on the number of college-level courses (Advanced Placement), tests and enrollments in college-level online courses that have been expanding "outside credits" by about 5-10% per year over the last decade. 

Their second prediction plays off #1. With three-year bachelor’s degrees becoming more of the new normal, private colleges and universities offering 4-year degrees will close. Less time on campus means fewer tuition dollars. Add slowing enrollment growth and more pressure to hold down tuition costs and they calculate a 30-40% contraction among private non-profits over the next 20 years. If you consider that as many as 20 private colleges closed in the past year and that trend just stays flat, for the next 20 years, that's about 500 schools disappearing.

And the third prediction follows those two by saying that the for-profit market for college education will account for 20% or more of college credits (not degrees) each year. That ties into prediction #1 about outside credits. We thought back in 2012 that MOOCs would make this happen. They peaked, leveled off, dropped some and are now coming back in a less "open" and free fashion. Some of the alternatives the article offers are quality virtual education in key content areas (for example, core courses in the social sciences), and others offering very specialized experiential learning programs (gap years; semesters abroad) with the focus on credits, not degrees.

Blount points to these firms (not colleges) as not needing the same infrastructure and other overhead typical of traditional higher education. This is not news colleges want to hear, but it is news they need to consider.

Blount points to NCES data, saying that for-profit degree providers currently register about one million students (7% of U.S. enrollments) each year. Despite a downturn among for-profit educators after a growth spurt from 1995-2010, private equity firms have been acquiring both struggling providers and long-time providers like DeVry University and the University of Phoenix and will take advantage of the market opportunity in offering credits.

Another prediction is that the gap between elites and non-elites in the college marketplace will only grow wider. Though high-demand schools won’t face the same pressure to accept outside credits, the boards and faculty at many of these schools will explicitly move to a three-year campus residency standard to create more slots for students. 

Finally, they predict that former college campuses will be bought by companies to be used as sites for-profit education, senior living facilities, healthcare centers, corporate campuses, residential learning sites, camps, training facilities, etc. In March 2018 Bloomberg reported that Chinese companies have already purchased at least four campuses in New England. 

And 5 conclusions:

1. The restructuring of the U.S. higher education system is in motion
2. Parents preparing to send children to college should make use of all opportunities available for your students to earn college credits while in high school or through a gap year experience between high school and college. 
3. Looking at a private college? Ask about their policies for granting college credit for prior coursework and early graduation options
4. Schools should consider metropolitan settings for adjunct teaching talent and employment options for dual-career faculty couples, gaining scale and sharing costs through potential mergers, roll-ups and other consortia options.
5. Be realistic about the coming headwinds and prepare to offer the best possible array of educational options for future students.