New frontiers of ML- Transfer Learning & Federated Learning

ML use and success in industry has so far been mostly driven by Supervised Learning. Fueled by advances in Deep Learning, more capable computing utilities, and large labeled data sets, supervised learning has been largely responsible for the wave of renewed interest in AI, funding rounds and acquisitions, and the applications of machine learning that have become part of our daily lives.

Business applications of ML in real world are too complex and have its own set of issues and challenges. It is to overcome such challenges that new frontiers of ML are created. In this article I bring together two forms that are driving the next wave of ML- Transfer Learning & Federated Learning.

Refer here for my article on Supervised, Unsupervised and Reinforcement Learning 

Transfer Learning

In the classic supervised learning scenario of machine learning, if we intend to train a model for some task and domain, a critical ingredient is the labeled data for the same task and domain. The task is what our model aims to perform e.g. recognize pedestrians in images, and domain is where our data is coming from e.g. images taken on Bangalore roads. To train the model for such scenario, one would need a lot of data (photos) from Bangalore roads with pedestrians tagged in each of them.

The supervised learning paradigm breaks down when we change the task or domain and do not have sufficient labeled data for the new task or domain. For example, if we want to train a model to detect pedestrians on night-time images (different domain). We would now need tons of labeled images taken during night time. Model performance from daytime data will collapse as the model does not know how to generalize to a new domain (night time). Or, if we want to train a model to detecting bicyclists (different task), we cannot even reuse the existing model, as the labels between the tasks differ.

Transfer learning allows us to deal with these novel scenarios and builds on the learning from one task & domain with good data (e.g pedestrian-day time), to other where data might not be available in plenty (e.g. cyclist- day time or pedestrian- night time). This now enables production-scale use of machine learning that goes beyond tasks and domains were labeled data is plentiful.

According to Andrew Ng, Transfer Learning is going to lead the next wave of commercial ML.

Federated Learning

Standard machine learning approaches require centralizing the training data on one machine or in a data-center. But with mobile and IoT, data resides scattered in hundreds of thousands of devices. It is to cater to this distributed ‘small data’ that federated learning comes handy. Models here are trained from user interaction with mobile devices.

Federated Learning enables mobile phones to collaboratively learn a shared prediction model while keeping all the training data on device, decoupling the ability to do machine learning from the need to store the data in the cloud. This goes beyond the use of local models that make predictions on mobile devices (like the Mobile Vision API) by bringing model training to the device as well.

It works like this: your device downloads the current model, improves it by learning from data on your phone, and then summarizes the changes as a small focused update. Only this update to the model is sent to the cloud, using encrypted communication, where it is immediately averaged with other user updates to improve the shared model. All the training data remains on your device, and no individual updates are stored in the cloud. 

Federated Learning allows for smarter models, lower latency, and less power consumption, all while ensuring privacy. And this approach has another immediate benefit: in addition to providing an update to the shared model, the improved model on your phone can also be used immediately, powering experiences personalized by the way you use your phone.

These advances in Machine Learning are pushing the envelop for AI. Sky is the limit!

References

(1) Transfer Learning

(2) Federated Learning