Thrust 1: Internet-of-Things and Scalable Databases for Digital Agriculture

Deployment in this space typically involves three types of platforms that have different computational resources and degrees of latency:

1. The sensors that are often inexpensive and need to be deployed effectively and programmed to be energy-aware for running on their battery power;
2. The edge devices that may be servers on the farm (say in the farm building) or tablet carried near to the farm operations; and
3. The cloud platform to which these lower-resourced devices connect to from time to time to transfer data for more sophisticated computation to occur, such as with the intent to connect networks of small and mid-sized farms.

Questions we ask in this space are along the following lines:

1. Can we distribute computation effectively across the three platforms of computation to optimize energy consumption and latency. In order to optimize the computation at the sensor level, we use approximate forms of computing such as trimming down the size of a neural network to work within the bounds of accuracy specified by the user.
2. We are interested in designing the middleware for more efficient computation on the sensor and at the edge. Edge computing essentially brings storage, networking, and computing closer to the consumer. Thus, edge computing is more amenable to scenarios where latency is a concern, such as when real-time decisions have to be taken.
3. Finally, we are interested in optimization such that database performance (whether locally or on cloud instances or cloud virtual machines (VMs)) is maximized for bounded dollar costs.

To read about some of our work in this area, refer to our papers in ACM-Middleware (Rafiki, 2017) and Usenix-ATC (Sophia, 2019) and ASABE (Iris, 2019).