Presentation
Machine Learning Sensors: Building Blocks for AI-enabled Systems
DescriptionMany built objects could benefit from emerging AI capabilities, including person sensing, gesture recognition, even voice interfaces. Currently integrating these into systems is a software and machine learning problem. Engineers have to gather training data, train ML models, and deploy complex software libraries on embedded systems with tight power, cost, memory, and compute constraints. They face challenges around protecting their users' privacy, since the cameras and microphones involved can collect very sensitive personal data. Unsurprisingly, most such proposed projects fail. Machine learning sensors are a new approach to system design. This involves wrapping commonly needed capabilities like voice, person, and face recognition in small, low-cost hardware modules with a simple interface like I2C that provides only actionable metadata from the subsystem. This abstraction protects user privacy by ensuring the module's guarantees can be verified by third parties, and greatly simplifies the system development process by providing a very small API with a protocol that engineers are already familiar with. This talk will discuss existing commercial and open-source examples of these sensors, and how this idea is likely to evolve and grow over the next few years.
Event Type
Special Session (Research)
TimeWednesday, July 12th1:30pm - 2:00pm PDT
Location3001, 3rd Floor
AI
