Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Explore how small and medium enterprises can implement edge generative AI through an innovative laboratory mouse monitoring system in this 17-minute conference talk. Discover how researchers developed a smart cage system that predicts mouse behavior and intentions using just 16 frames of VGA-resolution video, moving beyond basic monitoring of water and food availability to anticipate potential harm and enable timely intervention despite clean-room protocol constraints. Learn about the technical optimization journey from a 240-million parameter model down to 170 million parameters while improving accuracy, and see how the system achieves sub-300 millisecond inference times on a Raspberry Pi 5 without hardware acceleration, with potential for 30ms real-time performance using specialized hardware. Understand the three-component generative neural network pipeline combining video-to-text models, OPT transformers, and text-to-speech capabilities for natural interaction, and gain insights into scaling generative AI appropriately for resource-constrained edge devices without requiring supercomputers or billion-parameter models.
