Defining the future of Machine Learning with Google and TensorFlow Lite
There are more than 20 million developers around the world. Developer experience and adoption plays a pivotal role in the success or failure of any software. Apple when it opened their platform for developers in 2008, it changed the entire mobile industry, now there are more than 2 million apps in the appstore.
Machine learning is a difficult and complex discipline. Implementing a machine learning project involves significant effort and deep understanding of underlying principles. Thanks to Google’s TensorFlow this is far less daunting for a developer. TensorFlow makes it easy for acquiring, training, serving and building machine learning solution. Currently TensorFlow is the most popular framework for developing machine learning solutions.
What is TensorFlow Lite?
TensorFlow is an open source numerical computation library created by Google brain team to distribute machine learning for large scale deployment. This framework can be used to run machine learning code on a large cluster of machines easily. Thus freeing developers and researchers to focus on solving machine learning problems than distributed systems.
Tensorflow Lite is the light weight solution of TensorFlow built specifically for mobile devices, it abstracts away the optimisation and performance enhancements essential for running a machine learning model on a mobile device where resources are constrained and vary significantly from user to user. With more than 2.1 billion mobile devices worldwide, there is a very strong need to create an engaging experience.
Innovation @ YML
At YML we have an open innovation culture & we constantly keep exploring machine learning in various aspects like Age & Gender classification, Exploring Text Region & Face Detection, Hand Gesture Recognition and more. When Google approached YML to partner and work with their TensorFlow Lite developer experience we were a natural fit.
Initially we brainstormed and came up with a collection of ideas. Working with the Google team narrowed down and decided on 4 examples. Our goal was to simplify the problem broadly for 3 kind of developers. Android developers, iOS developers and ML engineers. The breadth of knowledge required to conceptualize and implement such a solution included machine learning, frontend, iOS, Android. Not only was the proficiency in machine learning essential, but team needed to have strong cross functional expertise in all the technologies to execute well.
Our work specifically involves Gesture recognition, Image classification, Object detection and Speech recognition.
Once we started to work together, we quickly adapted to collaborate and create multiple iterations of the solution. Agility and iteration was key to creating a long lasting experience and deliver on the promise of developer experience.
Be sure to checkout the TensorFlow Lite and learn the technology which would shape the next generation of Machine Learning products.
The story is far from over, there is still lot more to be achieved and lot more work to be done. Follow our publication to get more udpates.