Making machine learning in education easier for every day users

Last week I wrote “Two ways to use Azure Machine Learning in education”, which started exploring the use of algorithms, alongside cloud-based machine learning in education to solve some of the key challenges facing education institutions. The problem is that it all sounds so very geeky. Hey, I just wrote “algorithms” and “machine learning” in the first sentence, which kind of proves the geekiness. Although this kind of technology is making huge differences to our online lives (like protecting us from spam email and giving us just the 3 out of 100 emails that aren’t spam) it’s also something that has been the domain of technical wizards. To make a difference, machine learning in education has to be simpler.

But we’re moving into a world where we’re going to be able to use this technology to solve real-world problems that don’t involve huge numbers of data scientists, and where the real knowledge sits inside the heads of business users in our organisations. Not the IT department and the data analysts.

So how do we make it easier for every day users to be able to apply their expertise to analyse their own data?

Part Two: Making it easier for every day users to use intelligent analytics and machine learning

Missed Part One: Building and sharing algorithms? Here it is.

If we’re recognising that there’s just a bit too much rocket surgery involved in today’s work with data, how do we make it easier to work with, for mere mortals like you and I? Well, there’s some smart teams working on that across Microsoft.

Patrice Simard, a Microsoft Distinguished Engineer, is leading a new machine teaching research project at Microsoft Research, which plans to focus on how to make the tools and UI possible for non-experts to create helpful and valuable machine learning capable systems – rather than just focusing on how to make machine learning algorithms more accurate – through a project call ‘machine teaching’. Before you react with shock, this isn’t about machines teaching, but about users teaching machines!

As Patrice says “No one has really built a machine learning tool for the layman” – and as more uses are found for machine learning, there’s a growing deficit between demand and the availability of data scientists with the right skills. There just aren’t enough people with machine learning expertise to do all the projects businesses and organizations want.

You can read more about this work on the next evolution of machine learning: Machine teaching here

But there are already some practical examples that you can look at to see what the future of Machine Learning could resemble for every day users, in Project Oxford, revealed earlier this year. Project Oxford allows developers to create smarter apps, which can do things like recognise faces and interpret natural language even if the app developers are not experts in those fields.

Project Oxford currently includes four main components:

  • Face recognition: This automatically recognises faces in photos; groups faces that look alike; and verifies whether two faces are the same.
  • Speech processing: This can recognise speech and translate it into text, and vice versa. A developer might use it for hands-free tools such as the ability to dictate text or to have an automated voice read out instructions or other necessary functions
  • Visual tools: This can analyse visual content to look for things like inappropriate content or a dominant colour scheme. It also can detect and understand text in photos, such as a team name, and can sort photos by content, such as pictures of beaches, animals or food.
  • Language Understanding Intelligent Service (LUIS): This enables applications to understand what users mean when they say or type something using natural, everyday language. Using machine learning, in which systems get better at predicting what the user wants based on experience, it then figures out what people want the app to do. For example, in an exercise app the system might learn that when the user says “I want to start my run,” “begin a run” or even “go for a run,” it all means that it should begin tracking the person’s distance, and that the type of activity is a “run”.

If you have basic development skills, or you can team with somebody who has, then the Project Oxford website is the place to start.

So far, so good. But what about real uses of this technology? And what about the simplification angle? – sparking a viral use of machine learning


A couple of months ago something called the #HowOldRobot went viral globally. It was the work of 3 Microsoft engineers who were applying machine learning systems to the challenge of working out how old somebody looks. And they made it very simple. You go to and either upload your own photo, or find one on the internet, and it will estimate how old everybody in the photo looks.

Of course, it’s not 100% accurate, but it’s a powerful demonstration of simplifying the use of machine learning – for a start, it’s trying to guess how old you look, not how old you are!

Users are simply posing a problem, and letting the technology start solving it.

I found it disturbingly accurate. For example, a month before my 50th birthday, it tagged me as 50, and then got me as 48 for a photo taken when I was 48.


But, then again, when I tried it with a photograph of me in my 30’s, it completely messed up, and pronounced me as 15 years old (better than prematurely ageing me)


You can read a lot more about the #HowOldRobot, and the technology behind it, here, and there is also an excellent behind the scenes look at the viral growth of the #HowOldRobot here (imagine building something you thought would be used by 50 users, and getting 35,000+ people using it within three hours).

Inspired by this project, another Microsoft engineer Mat Velloso built a service in just a few hours to compare two photos of people and rate their similarity, with a ‘twin rating’. With, it’s the same simple user interface, and process of hiding the complexity of machine learning – with just a few hours coding. Again it uses the Project Oxford work.

You should try it for yourself, but here’s the result for the two most twin-like members of our Australian education team.image

How can simplifying machine learning in education help?

If we can use machine learning algorithms for (arguably) trivial things, and make it very simple to use, where can it be applied in education?

Carnegie Mellon University are already using it work out how to cut campus energy usage by 20%.

Helping student retention in universities

One of the examples that is easy to see (and currently very difficult to solve) is the problem of students dropping out of universities. In Australia, one in five students drops out of their course in the first year, with the majority dropping out of the university altogether. In some universities, this is as high as one in three students.

And yet, there is a strong bank of research from different universities which identifies the key factors that are associated with students dropping out (across six different studies, there are four factors which feature in the top five of over half the studies). Some projects have identified over 30 factors to monitor and analyse. It is the perfect scenario to use machine learning, because instead of spending a year or two analysing the factors, you can analyse the data every night for every student, and help identify the students at risk of dropping out. And plan your proactive intervention and support in response to what you can predict will happen, rather than reacting once you’ve discovered something has happened.

The beauty of using machine learning to do this is that the system can manage the model itself – it learns as it goes along, rather than you having to keep using an out-dated idea of what the causes of drop-out are. That’s just one of a few key reasons why you find telco’s using it to forecast customer churn, and online retailers using it to suggest additional products to buy.

What to do next

If you’ve made it this far, hopefully you can see that there’s some value in keeping an eye on machine learning. So what can you do next? Here’s three resources and ideas for next steps:

  1. Find a Microsoft partner with the Data Analytics competency, and experience in education, and see what ideas they have to help
    > Go to Pinpoint, the Microsoft website for finding partner solutions
  2. Learn more about Azure Machine Learning yourself, or talk to your analytics team internally to try an idea out
    > Machine Learning documentation and tutorials are here
  3. Read the Microsoft Machine Learning blog
    > You’ll find it on TechNet here
  4. Keep an eye out for events which include Azure Machine Learning, and especially use cases in education
    > There’s the first ever Cortana Analytics workshop in Seattle in September
    > Closer to home, S1 Consulting are running a workshop to launch their Student Retention module, which uses Azure Machine Learning, in Brisbane on 10th August
  5. Or share this article with friends and colleagues, and see what they have to add!
    > Share on Twitter
    > Share on LinkedIn
    > Share on Facebook