On-Device Computation for Digital Agriculture

Our lab develops algorithms for resilient data lakes for data generated from Internet-of-Things (IoT) sensors deployed in smart and connected farms, that are in various stages of adopting IoT technologies. Here are two vision papers demonstrating the emerging directions in this area and our contributions to it.

Relevant projects:

• Approximate in-sensor analytics: this is for both lightweight workloads such as anomaly detection to detect the health of IoT sensors and for heavy-bandwidth workloads, e.g., object detection workloads that require the use of neural network variants. The approximation of these algorithms will require techniques such as early exit and low-bit precision computations using neural network quantizations.
• Edge computing: here we are deploying various algorithms on edge servers, which could range from Raspberry Pi models to NVIDIA’s Jetson systems.