Rapid AI Development for Greenhouses using Private Silo Data

In our last article, we discussed two main issues: 1) the lack of accessible data needed to make AI innovations in the agriculture sector; and 2) how TensoAI’s onsite AI device, Elroy, aims to circumvent many of the challenges we face in the development of AI in agriculture, such as by building the connections required for data scientists to access data securely at farming regions.

Our discussion will now focus on how methods to protect data privacy enables more rapid development cycles when producing AI solutions. We will end by noting why this is of particular importance to introducing AI technology into greenhouse production

Private data silos at greenhouses impede AI innovation

In today’s greenhouse industry, it is typical to find multiple independent private “data silos” stemming from technologies that service indoor growing facilities. Each silo could be thought of as a “data service provider” and include systems like climate-control systems that collect data with sensors and plant vision systems that provide image data from cameras. Each system in the greenhouse comes with unique security issues and legally-binding agreements in terms of data sharing. AI models produced today in greenhouse agriculture often use one of these private data silos in conjunction with publicly accessible data to develop and train the models.

The greenhouse market is replete with services and hardware that help greenhouse operators be more productive. All these different vendors employ their unique methods for data collection and interoperability between computer systems. Both issues pose challenges when trying to implement new technology, such as AI, into pre-existing systems. Isolated private data silos and the variability of interoperability of data-service providers make it challenging for new AI technologies to scale and remain secure across multiple greenhouse facilities.  

TensoAI’s Ecosystem

TensoAI overcomes this challenge by making it possible to share AI systems and data exchange within a greenhouse’s diverse systems and data-service providers. In the case depicted in the image above, predictions from AI systems remain exclusive to a given greenhouse client whereby no other greenhouses can benefit from that client’s data and AI technology. Since all greenhouse producers strive for greater efficiencies in their production, the current technology-exclusivity disadvantages the industry as a whole by inhibiting collaboration in developing and accessing efficiency-promoting technologies. Our current systems seem to do little more than protect intellectual property and private data. 

TensoAI provides an alternative system based on information exchange. By facilitating frictionless communication between data-service providers, our system enables greenhouse systems to share AI predictions and reinforce the accuracy of these AI technologies. Such exchange and communication ensure a more holistic operation of a greenhouse, where all data-service systems benefit from a data-driven feedback loop used to train and validate AI models as well as provide fundamental insights about the greenhouse as a whole.   

Consider the example of automated pest detection made possible by AI. This new technology is increasingly available from companies providing climate control and computer vision services. Both companies provide similar AI pest-detection services but implement different AI models with different datasets, where one uses digital image data, the other is measures of environmental growing conditions. In the past, validating the accuracy of AI models would generally consist of collaborating with the head grower of a greenhouse to impart her domain expertise to verify that the model indeed provides an accurate representation of growing practices. Getting such expert feedback is a slow process that involves greenhouse operators to take time away from their busy schedules. With TensoAI’s infrastructure, we can reduce the need for domain expert feedback by tapping into the strength of AI models operating on a data-driven feedback loop, which funnels results back to machine-learning engineers to enable faster development of AI models.

Developing models using data you can’t see

TensoAI is focused on data privacy as a top priority. With basic knowledge of cultivars and vendor systems in use at a greenhouse, data scientists can train AI models using fully anonymized data. Data scientists do this by using privacy-preserving tools to access useful statistical information that prevents the extraction of identifying client data. Protecting privacy, in this case, involves tools that introduce privacy-protecting “noise” into the data, where TensoAI uses the noise to control the visibility of statistical information, which is stipulated under a trust agreement between the greenhouse and data scientists. 

When running models and gathering statistical information at the greenhouse, TensoAI utilizes Elroy, our unique AI-hardware device. Elroy has sufficient computation power to ensure models operate safely and securely. Thus data scientists and greenhouses no longer face common barriers such as insufficient cloud computing power, poor data migration and unreliable data streams.

(If you would like us to demo this process with Elroy, simply send a request to hello@tensoai.com)

TensoAI has your solution

TensoAI’s infrastructure enables engineers and data scientists to tackle actual growing issues with real-world data. This is important since it helps ensure that production-ready AI models are possible and sets a foundation to provide novel AI solutions to meet the future needs of greenhouses. Challenges faced by greenhouse producers are diverse and creating solutions that address these diverse issues will ultimately reduce production costs and increase both the quality and quantity of harvests.

When we develop AI models with data privacy as a top priority, we create an infrastructure that enables quicker development and deployment of AI models. Hence, we are building not for today, but for forever.