Qualcomm announced a new division that would unify all its fundamental artificial intelligence research. Qualcomm A.I. Research gives form to what was largely an amorphous effort inside the company, which is focused on moving inference out of the cloud and into devices installed with its mobile chips, such as smartphones, warehouse robots, cars and security cameras.
The reorganization reflects Qualcomm's doubling down on embedded artificial intelligence, which it argues can improve privacy for applications like voice-controlled speakers and save energy wasted sending information to the cloud. Taking artificial intelligence – a blanket term that includes machine learning – out of the cloud would also lower latency, which is important in mission-critical devices that require fast reaction times, like driverless cars.
Accordingly, the company is focused on model compression and efficient hardware to squeeze as much processing as possible from embedded devices constrained by power and heat. Qualcomm A.I. Research is also targeting more data efficient models in machine learning, as well as system architecture problems – like sensor fusion and multimodal learning – and device personalization.
While other companies design custom accelerator chips, Qualcomm has not abandoned existing hardware, making the most of its conventional CPU, GPU and DSP cores. The company uses software tools to evaluate algorithms and the cores available in Qualcomm’s Snapdragon chips, and coordinate everything to run inference tasks as quickly and efficiently as possible.
That strategy could put artificial intelligence at the fingertips of companies worldwide. Qualcomm, the world’s largest supplier of smartphone chips, said that it could enable artificial intelligence in “hundreds of millions of products” using its chips every year. “Our goal is to make on-device A.I. ubiquitous,” said Jim Thompson, chief technology officer of Qualcomm, in a statement.
Helping Qualcomm is its growing ranks of artificial intelligence researchers. Last year, the company acquired a software startup founded by Max Welling, a professor at the University of Amsterdam and former student of Geoffrey Hinton, the founder of Google’s deep learning research team. The company, Scyfer B.V., designed algorithms that can spot manufacturing defects in products or generate insights from bank transactions.
Welling, currently vice president of technology at Qualcomm Technologies Netherlands, is not changing roles. But a spokesperson said the formation of the research unit “helps to expands Max’s ability to drive and guide the A.I. research work already going on with Qualcomm.” The spokesperson declined to comment on the headcount of the new division.
Qualcomm A.I. Research mirrors labs launched by major internet companies like Facebook, Baidu and Google. In 2016, Uber Technologies launched a research lab led by Gar Marcus, a professor at New York University and chief executive of startup Geometric Intelligence, which it acquired the same year. This month, Google renamed its whole research division to Google A.I. to emphasize the importance of artificial intelligence to its business.