Coral, Google’s artificial intelligence initiative without the need for the cloud

With the launch of the Coral hardware and software platform, Google allows individual and enterprise users to develop their products.

Artificial intelligence allows machines to perform all human tasks on their own. Need to set up quality control on the factory production line? Or you need to set up a camera equipped with artificial intelligence to detect defects. What do you think about the interpretation of medical data? Machine learning can scan a physician to identify potential tumors and flag them for a doctor.

These apps are useful as long as they are safe and fast. An AI camera that takes a few minutes to process images is not useful for the factory environment, and no patient wants to risk exposing their medical data to the cloud . All of the above are problems that Google is trying to solve with a program called Coral. According to Vikram Tank, product manager of Coral:

The data of artificial intelligence devices are sent to large computing samples deployed in centralized data centers. In these centers, machine learning models operate at high speeds. Coral, a platform that includes Google’s hardware and software components, helps build AI-enabled devices. One of the benefits of this platform is the increased speed of the neural network hardware.

Coral Products

Coral products, like the development board, are used to build prototypes of new AI devices.

You may not have heard of Coral before (this product was released in beta last October). Coral is part of the growing field of artificial intelligence. According to market analysts, in year 2, more than 2 million chips and PCs will be sold, up from $ 1.5 billion by year 6.

Although most AI chips are install on user devices such as mobile phones, they can also be used by enterprise and industry customers such as automotive and healthcare.

Coral offers two main products to meet customer needs: accelerators and development boards used to prototype new ideas, and modules that help brainstorm AI-producing devices such as sensors and smart cameras. In both cases, Google’s Edge TPU is the heart of the hardware. The Edge TPU is an ASIC chip designed to run machine learning style algorithms (the younger brother of the water cooled TPU used in Google’s cloud servers).

 

While engineers can use Coral hardware to build fun projects (for example, Coral provides guidance on how to make a confectionery machine and feed intelligent birds), the hardware’s long-term focus is on corporate clients in industries such as the automotive and healthcare industries. For example, consider a car scenario that uses car vision to identify objects on the street. According to Tank:

“A car with a speed of 4 km / h (2 km / h) takes about 2 meters in 2 milliseconds, so any delay in processing due to a slow mobile connection increases the risk of risk.” Stop sign or traffic light, analyzes should be performed by the device itself. The tank points to similar advantages in privacy :

Suppose the manufacturer of medical devices wants to analyze ultrasound imaging using image recognition technology. Sending images to the cloud weakens the connection for hackers, but doctors and patients by analyzing instant images on the device itself make sure that data processing doesn’t get out of control.

edge tpu

Google’s Edge TPU, a small processing chip for artificial intelligence that is at the heart of most coral products.

Although organizations are the primary target of Coral, this project is root in Google’s AIY, a machine learning tutorial set. AIY packages launched in the year 5 supported by PC legacy PCs allow users to build their own speakers and cameras. This project was a great success in the STEM toy market. The AIY team quickly realized that while some customers build toys solely on instructions, others want to use the project’s hardware to prototype their devices, so Coral came up to meet the needs of those customers.

With the Coral platform you can build prototypes of hardware products

Google’s problem is that there are dozens of companies with similar performance to Coral. These companies range from startups like Seattle’s Xnor (maker of solar-powered optical AI cameras) to big giants like Intel , which unveiled its first USB accelerators in year 6 and raised $ 5 billion in December last year. Chip maker Habana Labs is dedicate to improving the latest AI products.

Despite the high number of competitors, the key to the Coral differentiation is hardware integration with Google’s ecosystem AI services. This product suite (including chips, cloud training, development tools) is a key strength of Google’s AI. Coral has a library of AI models that are tailor-made for its hardware, as well as a suite of AI services on Google Cloud that are integrated with standalone coral modules such as environmental sensors.

Coral is heavily ties to Google’s artificial intelligence ecosystem. Google’s Edge TPU hardware is only compatible with TensorFlow, Google’s machine learning framework. This is one of the limiting criteria in the growing market for artificial intelligence. According to an artificial intelligence spokesman for Kneron:

Coral products have a proprietary platform while our products support
all the major frameworks and models of artificial intelligence available in the market. ”(Kneron says there is no downside to evaluating them. Is).

Coral’s precise performance cannot be talk about today. Google is more focused on AI cloud services and does not share sales statistics and goals with Coral. According to an informed source, most coral orders are for standalone units (for example artificial intelligence accelerators or development boards) while a few customers have orders of 5,000 units. Coral’s appeal to Google is not necessarily monetization, but its primary goal is to learn more about artificial intelligence.

Leave a Reply

Your email address will not be published. Required fields are marked *