It can be very difficult to ensure that your farming equipment uses the latest and greatest machine learning models. Connectivity can be sparse and expensive when you’re using satellites, says Jason Campbell, Wallaroo’s director of architecture.
His company therefore introduced the Air Gap Edge Deploy functionality to make it easier for companies to deploy and manage machine learning models at the edge in environments without IP connectivity. Think of derricks, gas pipelines and transmission lines for energy and utilities; and autonomous equipment in smart manufacturing and smart agriculture.
We spoke to Jason Campbell about the air gap solution and how it could benefit farmers.
Why was the Air Gap Edge Deploy feature developed?
“There are many reasons why companies (not just farmers) could use machine learning over Air Gap. For example, devices might be far from Internet connectivity at the edge, such as in oil rigs, gas pipelines, or in agriculture, perhaps a combine harvester in a rural area far from a cell phone tower. Additionally, the rise in cybercrime has led some organizations to explore the possibility of air-gapping to protect their systems. By isolating their networks from outside networks, they can prevent the vulnerabilities that come with those connections, like data breaches and ransomware attacks that can cause billions of dollars in losses.”
How does the Air Gap Edge Deploy feature work?
“You can train a machine learning model anywhere, in the cloud or on-premises. But from there you need a remote connection (e.g. to the actual farming equipment you want to use the model on. We provide more details on how it works here.”
Why is it relevant for plant breeders?
“Agricultural machines generate more data than ever. John Deere’s CTO said in an interview with The Verge that their farming equipment has essentially become “mobile sensor suites with computing power.” By combining this sensor data with AI, plant breeders can use just the right amount of water, fertilizer, pesticides, etc. for each individual plant. In addition, plant breeders rely more on robots to pick different plants with computer vision and so on.
What data plus AI essentially brings you as a farmer is a higher yield while reducing your input costs. However, it is very difficult to ensure that your devices use the latest and greatest machine learning models. And you also have to think about the flow of data back to the data scientists who trained the model. All of this sensor data can be gigabytes or even terabytes of data per day.
Connectivity is sparse and can get expensive if you use satellite. It is therefore much more cost-effective for the equipment owner or servicer to use something like this air gap solution to deploy the machine learning model in the equipment and take the production data so you can ensure your models stand still to be accurate and good to go function. More information on how this data flow comes in and out can be found here.”
What is the cost for a generator using this air gap solution?
“As you see more of a shift towards Equipment as a Service (EaaS), more of these costs will be borne by the owners and service providers of the farming equipment, who will then sell them to farmers as part of larger smart farming services. So a farmer pays for AI services, which include edge machine learning.”
And what are the financial benefits for the breeders?
“The adoption of machine learning in agricultural production has become a necessity given the need to increase food production while balancing sustainability. ML has already started to make an impact, providing insights that help increase productivity, use less water, fertilizers, pesticides, etc. The data processing capabilities afforded by ML have meant that early adopters have expanded crop growing opportunities into previously unsuitable areas and increased overall yield.”
Can you give a practical example?
“We already have examples of farming equipment that automatically adjust water, nutrients and other chemicals used down to the plant level using sensor data combined with AI. Regarding our own air gap solution, we don’t have any examples to share, but we do have them in other industries, especially in manufacturing. In fact, we’re working with the US Space Force to bring machine learning to their fleet of satellites.
