Study Shows how Mobile Devices can Conduct Machine Learning

According to a new EPFL/INRIA research, it’s now possible for our smartphones to conduct machine learning as part of a distributed network.

One of the compromises of accessing the internet is data privacy.

Big techs can access behavioral data every time you’re online — whether reading news or searching for the closest coffee shop. What’s more, companies such as Facebook and Google claim that they collect user data to improve services.

That’s because machine learning models use your history to compile news articles that they think will interest you. Similarly, the models could offer videos that are similar to the ones you watched in the past.

In other words, tech giants use behavioral data to create a more personalized web experience. But there’s a flipside.

These companies also use your data to create targeted advertising. It’s also possible to share the information with third-parties — a reason digital privacy issues are so prevalent.

However, having machine learning in your pocket could change all that.

For the first time, a French research team has shown that machine learning can run on mobile devices in real-time. What’s more, there won’t be any compromise in functionality or sharing of data with tech giants.

In a statement to the press, one of the study author, professor Anne-Marie Kermarrec said:

“What we have shown is that if we put all our phones together they start constituting big computing power to match the likes of Google and that gives people alternatives to relying on centralized, powerful computer farms.”

So, how does it work?

Conducting Machine Learning in Your Pocket

The researchers introduced FLeet, a result of advancement in Federated Learning.

Federated Learning is a global model trained with updates that mobile devices compute while keeping the data local. Despite its privacy benefits, the model also has a low energy and performance impact.

As a result, it’s unsuitable for applications that require frequent updates, such as news recommendations.

Meanwhile, FLeet combines the privacy of standard Federated Learning with the precision of online learning. And that’s due to two core components, which include:

I-Prof: a profiler that predicts and controls the impact of learning tasks on mobile devices.
AdaSGD: an adaptive learning algorithm that is resilient to delayed updates.

As it turns out, our current smartphones have both the data and battery power to enable distributed machine learning.

In a statement about the model, study author and EPFL professor Rachid Guerraoui explained:

“With FLeet it is possible, while you are using your mobile phone, to use some of its spare power towards machine learning tasks without having to worry that your call or internet search will be interrupted.”

Currently, FLeet is only a prototype. The next step is to develop a usable end product and research other aspects of FLeet’s design.

The team also hopes to make the system more secure against possible attacks.

The researchers published their study in the ACM Digital Library.