SCAiLE stands for SCalable AI for Learning at the Edge and the consortium is working with the Japanese authorities to review opportunities for the 2020 Olympics, including video-based event detection and response capability.
“It is an open group and we have other companies expressing an interest to join,” Sylvain Dubois, vice president of marketing and business development of Crossbar, told eeNews Europe.
Crossbar is a US semiconductor memory IP company; Gyrfalcon is a machine learning processor company; mtes Neural Networks is a Japanese IoT equipment company and Robosensing a software company with use cases and datasets for IoT. As a result, the four companies bring a broad platform perspective and can provide full system-level solutions.
These could include smart lighting systems for a railway station, airport or city or smart car parking and traffic systems, Dubois said.
Through its four founding members the organization will combine acceleration hardware, resistive memory (ReRAM) and optimized neural networks to create ready-made, power-efficient solutions with unsupervised learning and event recognition capability.
The consortium foresees that the growth of IoT systems with thousands of remote edge devices – such sensor-equipped cameras – will create a torrent of unstructured information that cannot be handled effectively by classification alone.
The SCAiLE solution will be to accelerate neural networks at the edge to analyze and respond to multi-modal information including video, images, speech, keywords and sensor feeds. Gyrfalcom AI acceleration architectures and high bandwidth very wide ReRAM from Crossbar will allow rapid searching across multi-modal datasets, Crossbar said.
“We have been working closely with Gyrfalcon Technology Inc., mtes Neural Networks Corporation and Robosensing, and expect rapid progress in designing the platform,” said George Minassian, CEO of Crossbar, in a statement.
“The large volume of new kinds of information cannot be handled by mere classification,” said Matt Kobayashi, CEO of Robosensing Inc., in the same statement. “We need new ways of handling unstructured data at the edge, and the planned SCAiLE platform can help us get rid of the ‘tyranny of data classification’ through power-saving self-learning devices that use clustering to detect and interpret events.”
For now Crossbar is offering up straightforward digital non-volatile memory with broadly parallel access although it is possible that at some point in the future Crossbar ReRAM could be used for processing-in-memory [PIM], Dubois said. “PIM is a separate thing. At some point it may be part of SCAiLE but it’s not there yet.
The next step will be for the SCAiLE consortium to set up and demonstrate its AI platform.
Related links and articles: