SANTA CLARA, Calif., — The IoT DevCon and Machine Learning conference, held last week at the Santa Clara Convention Center, was mashup of advice and trends on how to develop products in the Internet of Things (IoT), machine learning and artificial intelligence (AI). The three will likely be intertwined in future products.
Analyst firm Tirias Research chaired the conference this year. Previously Marcus Levy, the chairman of benchmarking organization EEMBC and longtime president of the Multicore Association, chaired the conference. When he joined semiconductor company NXP in late 2017, Levy enlisted Tirias Research partners Kevin Krewell and Jim MacGregor to chair the conference.
Some key takeaways from the conference:
IoT will be a large market
Everyone has their target number for how many IoT devices will be connected by a certain date — 2020 to 2025. “The thing we do know is the market is going to be large,” said Krewell in his opening remarks. IoT went through the hype cycle, but Krewell reassures us that it will be big, with IIoT (the Industrial Internet of Things, for factories and manufacturing) as the largest sector of IoT — in the beginning at least.
Low-power, lasting for many years
It’s common knowledge that most IoT devices, especially those in remote or difficult-to-reach locations, need to be designed to consume low amounts of power and be energy self-sufficient for years, at least 10 years. This means the battery needs to last.
Security and privacy are very important to IoT:
“Without security, IoT will fail,” said Krewell. People care about security and privacy now.
“There is no pure security in IoT, or any devices,” said Krewell, but there are things you can do to make devices more secure. Without proper security, devices become targets for exploitation by malicious hackers. Devices can become botnets. Therefore, security is an ongoing process that needs to start from day one of your design, said Krewell. Device design decisions need to be informed by security. Adding security later won’t work.
Responsibility for security doesn’t end after product ships
“When you ship product, your responsibility doesn’t end,” said Krewell. You need to be sure your product stays secure over time.
Machine learning: large, proprietary datasets are gold
If you don’t have one, get one: a large dataset owned, groomed and managed by your company. Continue to grow the dataset. In machine learning, you are measured by your large, proprietary dataset. Why? You have to train your machine learning engine to make inferences by giving it enough data for an inference model.
Developers — unless you are in a large OEM or software company — are struggling with having the resources to deal with machine learning. Krewell says Apple and IBM are the first companies to offer solutions that include data and training resources.
Voice is new user interface
Alexa and Siri are here now, so consumers are already being exposed to AI. “Voice will become more important,” said Krewell. Other speakers at the conference agreed, including Geoff Lees from NXP and Vinay Iyengar, venture capital Investor from Bessemer Venture Partners, who both said voice was an important new user interface that was natural to users. Krewell said voice will be in IoT systems as well as AI.
Machine learning chips
Machine learning chips will become very pervasive in system design. “You will see more and more cores designed specifically to run machine learning,” said Krewell. “We are in very early stages of processing paradigm.” More work has to be done, but no one machine learning framework will win.
