To the uninitiated, machine learning — getting computers to act without being explicitly programmed — may sound like science-fiction.

But in reality, the technology is so widespread that you probably use it several times a day without thinking about it.

Machine learning makes web searches more effective, powers the facial recognition software that invites you to tag your friends in photos, and helps your smartphone’s virtual assistant recognise and understand your voice.

Recent breakthroughs in computer science have given machine learning a huge boost, and it’s increasingly being put to work in more advanced applications, ranging from language translation to medical diagnostics.

Here are five things to learn about the technology that is poised to take artificial intelligence to new heights.  

1.Show, not Tell

Traditionally, programmers have given computers explicit, step-by-step instructions. But in machine learning, the programmer becomes more like a trainer than a drill sergeant.


You can train a computer to learn how to recognise faces in a crowd.

“While a fixed set of predetermined rules works for some problems, it is not the way humans learn. It is also inefficient, since the computer cannot complete a task if we do not know the rules,” said Assistant Professor Robby Tan, a machine learning expert at Yale-NUS College, in an interview with TechNews.

“The key idea of machine learning is to create computer programs that can learn by themselves from data.”

Let's say you want a computer to recognise human faces. Rather than telling it what humans look like, you’d instead keep showing it thousands of photos of faces, until it eventually figures things out for itself.

The technology isn’t perfect, though.

But the computer is in effect learning — it can get smarter and better at its job over time, and can thus perform tasks that require more than just brute-force calculations.

Prof Tan, for example, is using machine learning to improve computer vision — how computers acquire, analyse and understand digital images — under bad weather and low-light conditions. His work could potentially be used to develop better driverless cars and surveillance systems.

2. It’s just gotten much more powerful

 The concept of machine learning isn’t new—humans have been trying to train computers for several decades. But recent advances have made the technology much more powerful.


Deep neural networks allow computers to beat the best human players at the game of Go.(Photo: Sean Welton)

“Machine learning has become more popular and impactful, primarily because of the availability of big data, and advances in high-performance computing for large-scale machine learning,” Steven Hoi, Associate Professor of Information Systems at the Singapore Management University, told TechNews.  

One such advance is the use of deep neural networks — massively distributed computing systems that mimic networks of neurons in the brain.

This technology underpins AlphaGo, Google's computer programme that in 2016 famously beat world number two Go player Lee Sedol.  

3. Anyone can try it today

Machine learning is fast becoming more accessible.

Part of this has to do with the availability of open source algorithms: in 2016, for example, Google made parts of its machine learning software TensorFlow free to the public.

The code lets users — companies, researchers and basement tinkerers alike — train neural networks for any application, and is already being used to track sea cow populations off the coast of Australia, analyse traffic patterns in London and sort cucumbers in Japan.

Machine learning also requires a lot of computing power, which in the past limited its use to big companies or research institutions that can afford the necessary infrastructure.

But now that companies such as Amazon, Google and Microsoft have started to offer machine learning as a cloud computing service, the technology is finally within reach of smaller outfits and even individuals.

4. It’s changing how we interact with computers

While machine language translation and virtual personal assistants are useful services, anyone who’s tried them out knows that translation and voice recognition errors remain common (and are often the cause of much human amusement).


Machine learning is improving how virtual assistants such as Apple’s Siri and Amazon’s Alexa understand our questions.

Over the past year, however, advances in machine learning have helped computer language translation take a huge leap forward.

Google’s Neural Machine Translation, rolled out in September 2016, runs entirely on neural networks. It works with text at a higher level, translating entire sentences rather than breaking them up piecemeal.

Likewise, machine learning is also improving how virtual assistants such as Apple’s Siri and Amazon’s Alexa understand our questions and provide answers in return.

The ability to carry out a real conversation, however, still lies squarely within the domain of humans.

For now, anyway.

5. It’s transforming healthcare

It’s unlikely that computers will ever completely replace doctors.

But machine learning, which confers on computers the ability to absorb and learn from huge amounts of data, has already started to transform medical care.

In 2016, for example, IBM’s Watson supercomputer helped doctors diagnose a patient's rare leukaemia by cross-referencing mutations in her DNA with data gleaned from tens of millions of cancer research papers—all in a matter of minutes.


Researchers are training computers to, for example, recognise images of pasta faster, using machine learning.

Machine learning could also help people lead healthier daily lives.

One of Prof Hoi’s research interests, for example, is training computers to recognise images of food.

“Food image recognition enables smart food logging—people can capture photos of their daily food intake and monitor their eating behaviours.”


Photo Credit:

1. Creative Commons Go Board Game by Sean Welton licensed under CC BY-NC-SA 2.0. Image has been cropped and edited.