Machine learning has been through numerous slumps, each following a spasm of technological over-enthusiasm. But it may be the only way to deal with the incredible complexity of situations that embedded-systems designers now face – something that is now being reflected in the strategies of embedded-processor designers.
Rene Haas, IP products group lead at ARM, says: “We believe machine learning is one of the most significant change that is hitting our computing landscape. We believe that years from now people will not be looking at machine learning as a unique category where computers learn but rather it will be native to everything that computers do.”Although people tend to associate machine learning with image and audio recognition through services like Apple’s Siri, the systems already in the market send a lot of data to the cloud to be processed. That is changing rapidly, says Haas.
“What is rising is the amount of inference and training that can take place at the edge. The level of analytics, the level of learning, level of sophistication performed locally has moved much faster than I think everyone has anticipated,” Haas explains, pointing to bandwidth and server-capacity problems as chief concerns.