Computational ‘breakthrough’ enables AI training on mobile devices

Computational ‘breakthrough’ enables AI training on mobile devices

Technology News |
By Rich Pell

The new method of training AI, says the company, enables AI training on devices such as cell phones, tablets, self driving cars, and any of the worlds 25 billion connected devices.

“Advancement in the capabilities of artificial intelligence is hindered by the current paradigm for training,” says Max Trokhimtchouk, CEO and founder of Recurrent Dynamics. “This involves acquisition of data, and training on centralized cloud infrastructure, typically provided by Amazon Web Services, Google Compute Cloud, or Microsoft Azure. The huge up-front cost and complexity of this paradigm mean that only very well funded companies can advance AI capabilities. We are changing that.”

By eliminating the need for expensive cloud infrastructure, says the company, its method promises “to bring the era of ubiquitous, human-like artificial intelligence a decade closer.”

“Our computational breakthrough enables AI to be trained on small cheap devices such as cell phones and tablets rather than expensive cloud infrastructure,” says Trokhimtchouk. “If you can run your AI model on a device, you can now train your model on the device. It is a powerful new paradigm that puts the advancement of AI capabilities on a new accelerated evolutionary path.”

The company says it foresees several impacts from this breakthrough:

  • AI will become significantly more capable, in less time: This is due to dramatically faster prototyping and larger scale training. Robotic devices and self-driving cars stand to benefit greatly.
  • There will be an explosion in practical applications of AI: This is because the new paradigm of training AI at the edge avoids the huge up-front costs of centralized training in the cloud. Millions more developers can now participate in advancing AI solutions.
  • We will see the emergence of the AI “collective mind”: This is because training can be coordinated between devices using the IoT (Internet of things) resulting in a subordinate role for cloud infrastructure going forward.

Trokhimtchouk, a PhD in mathematics from Berkeley, has 12 years experience in machine learning infrastructure. The company did not provide further details of its training approach, although its name – Recurrent Dynamics – suggests the method involves recurrent neural networks (RNNs) – a class of artificial neural networks where connections between nodes form a directed graph along a temporal sequence, allowing it to exhibit temporal dynamic behavior.

Recurrent Dynamics

Related articles:
Google brings machine learning to the edge
AI at the edge to significantly impact semi industry
Intel, Baidu partner on AI training processor
Baidu launches ‘no code’ AI model training platform
Edge AI startup receives more Chinese backing

If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles