Google Coral – Build Intelligent Systems with Local AI

Google Coral – Build Intelligent Systems with Local AI

Google launches Coral Dev Board which enables you to develop AI-based intelligent systems using fast and secure local neural networks on smart devices.

On the 6th of March, 2019, Google announced the launch of Coral (Beta), the platform which is all set to significantly transform the field of machine learning. As it is, lately, the world has woken up to the multitude of opportunities which the developments in AI have made possible.

As the leaders of the Coral team, Billy Rutledge (Director) and Vikram Tank (Product Manager), have themselves pointed out, “AI can be beneficial for everyone, especially when we all explore, learn, and build together”. True to this notion, Google has already developed inclusive and user-friendly platforms in the past, such as TensorFlow and AutoML. By launching Coral, Google has indeed taken yet another long stride forward in this direction.

In more ways than one, Coral will drastically improve the scope for developers to represent their ideas in terms of codes, programs, algorithms and so on.

What is Google Coral?

So, first thing first, let us deal with the obvious question. Well, to put it very plainly, Coral is a platform which will allow developers to create intelligent systems with the help of local AI.

In other words, if a developer wants to make an Internet of Things (IoT) hardware from scratch, she can now use Google Coral to include AI locally, and without the need for any cloud platform. It also allows developers to not only create content but also train and run local neural networks (NNs) on a local device.

Owing to the complete local AI toolkit provided by Coral, the process through which an idea moves from prototype to production will become much more tangible than before.

With Coral, Google’s aim is to significantly accelerate NNs locally. In doing so, Google has, obviously, considered speeding up the Neural Network performance of their product, as well as, has ensured optimum privacy. Among other things, both of these have been possible because of the complete localization of the process.

Moreover, due to its local model, Coral is very power-efficient. Coral has been designed in a way that enables the developers to actualize their ideas into prototypes while increasing scalability on production lines.

What’s New to Coral?

The Coral Dev Board is basically a single-board-computer (SBC), which can rightly be seen as a successor to innovations at Raspberry Pi Foundation, UK. Yet, as expected of Google, they have incorporated much newer elements into their SBC. This has resulted in a significant forward movement from its predecessors while transforming the field of AI development.

Priced at ~$149, the target group of Google’s product is not amateur developers but advanced ones. This, I believe, is the essential difference between the Coral Dev Board and other existing SBCs in the market.

What’s in the Coral Dev Board?

In this section, let us highlight some of the exciting specifications of the Coral board.

First, the board comes with an Edge TPU co-processor, a microchip designed to run and develop IoT using the Edge framework. Despite being the hardware itself, the chip is pre-installed with a set of AI programs and algorithms. Consequently, it allows developers to train algorithms, as well as, run trained data locally on any kind of smart device. This rules out the need to store this information on the cloud, and, also the dependence on the internet. However, the fact remains that developers can combine the local NN with the services from Google Cloud IoT.

Second, the System on Module (SoM) also comes with additional features like Wi-Fi, Bluetooth and USB connectivity along with onboard RAM and eMMC memory.

Third, “[to] make prototyping computer vision applications easier”, Google is also offering a camera unit, which works with the board over a MIPI interface. Coral also has the option to connect with an audio device through the dedicated audio jack slot.

Fourth, in order to make the development process lightweight, Coral works with TensorFlow Lite and not the traditional TensorFlow platform.

Fifth, in case there’s a need to speed up the algorithm training process even further, a USB accelerator can be used with the Coral board on a Linux system. In doing so, the ML algorithms are enabled to draw statistical conclusions, by a process known as Inference.

What are the Possible Downsides?

Well, as we already know, nothing is or can be absolutely perfect. Despite its many benefits, Google’s Coral Dev Board also has its fair share of downsides. At least, these shortcomings can turn out to be significantly problematic in certain situations.

Coral’s specific focus on machine learning and IoT, somewhat limits its scope to a certain degree. Although it can be connected to a monitor and keyboard, Google itself has warned that this might hamper the board’s overall performance. So, Coral cannot be appropriately used as a desktop, unlike some other SBCs in the market.

That said, it has to be granted that the purpose of designing Coral was to assist machine learning and not to make another SBC-cum-desktop device. In this regard, Google’s product does seem to stand apart from all others and can be used by businesses of all kinds, small, medium and large. Also, as Coral comes into popular use, it’ll obviously incorporate further developments and innovations, which will make the system even better.

Prateek Arora

Contributing Editor at Wimoxez. Apart from this, I'm big into books and love reading books in different niche.

wimoxez: Data, Insights and Intelligence

Data, Insights and Intelligence media platform and bring the best resources to explore valuable technologies which will shape tomorrow.