Machine Learning in Embedded, Data-deprived Devices

Modern AI approaches are not suitable for small embedded systems that are disconnected from the Cloud. The must autonomously process sensor data, decide what actions to take, and learn from their experiences without the benefit of Cloud infrastructure. Common challenges include learning from only one example, making rapid decisions and taking action in milliseconds; and delivering guaranteed performance. This talk will describe a variety of domains with these challenges and some of the approaches we use to handling them.