Elroy can Counter the Challenge of Private Data Silos in Greenhouse Farming

Quality data is the lifeblood of AI, and the lack thereof remains a challenge when developing AI models. Data deficiencies remain whether you tackle the problem right away by using publicly accessible data or dedicate resources to acquire large amounts of data from private sources. Regardless, data shortages force many AI companies to develop models on a small amount of real-world data.

"Over 1/3 of AI and Analytics Projects in the Cloud Fail Due to Data Quality"

-SAN FRANCISCO, Jan. 23, 2020 (GLOBE NEWSWIRE)

The Problem with Private Data Silos

For too long the majority of AI companies continue to develop the technology in the same antiquated way: Gather and Analyze; they gather data on their cloud servers, conduct an analysis and then develop models from the data. While this works fine for some industrial sectors, data from greenhouse facilities poses a unique challenge. This Gather and Analyze approach depends on gathering large amounts of data to produce AI solutions that will work in diverse greenhouse farms and indoor growing facilities.

"The January 2019 estimate of the World Greenhouse Vegetable area is 496,800 ha. (1,228,000 acres), according to Cuesta Roble."

-Produce Grower, January 7th, 2019 (Cuesta Roble)

By approximating the amount of greenhouse space that currently employs data-driven and data-collection technology, it becomes obvious that AI companies are still using a small proportion of greenhouse data available to train and develop AI models. If an AI company, for example, were to acquire data from 10 greenhouses of 10 ha in size, data generated from all these facilities would represent only 0.02% of the data being used right now in the greenhouse farming industry. Such paltry amounts of data are a far cry from being representative of the industry as a whole.

Complicating matters further, most data from greenhouse farming is private, stored in company files away from public view. Consider the example of greenhouse climate control systems, being a major data source. Since most climate control systems keep the data onsite in rural areas, it gets really complicated really fast for urban tech companies to access a steady stream of climate control data needed to build AI models to optimize greenhouse operations.

Being developed with limited datasets from individual greenhouses with unique growing environments and operations, we often get underperforming algorithms with unreliable accuracy. Moreover, an algorithm made for one location will not work in another greenhouse.

The key to AI model development for the indoor agriculture sector is to acquire large amounts of data from multiple indoor growing facilities. This is a hefty challenge with the current Gather and Analyze approach, where AI companies know all-too-well how the current system throws up multiple barriers to data migration, collecting sufficient data, and abiding by legal contracts and data security policies ( such as GDPR). There’s got to be a better way.

Data opens new possibilities

Over the past decade, AI innovation continues to develop exponentially thanks to new tools like Graphic Processing Units (GPUs) that speed up AI model training and new techniques that enable us to gather more insights from datasets. Data transformation, however, has played a pivotal role in our abilities to achieve prominent AI milestones. As an example, Deep Blue wouldn't have been able to beat chess grandmaster Kasparov without the 21 million chess games--published back in 1991--it used to master playing the game; IBM Watson’s Jeopardy win was only successful thanks to the millions of articles correctly structured by Wikipedia in 2009 that became a valuable dataset like no other.

Combining high-quality data with numerous improved AI tools, data scientists are now at the cusp of innovation and can access machine learning research at their fingertips. Such resources enable the quick adoption of techniques so that any data scientist can quickly test out new concepts within AI systems.

Despite the growth in advanced AI tools, we still see the limits of what our AI systems can do. Limited data hindered achieving almost every AI milestone, and this is definitely the major obstacle holding back AI progress in agriculture.

We unlock the doorway to private data

TensoAI's approach to data challenges starts with establishing infrastructure that supports AI development and deployment in agriculture. We focus on a "Compute and Gather" approach that delivers models to the data source as a means to increase the speed of AI development.

To accomplish this, we send greenhouses a small compact AI device we call Elroy. This non-intrusive hardware plugs into existing systems to perform AI tasks in the background. We make it as easy as possible for companies to adopt AI solutions.

TensoAI will break down data silos

Gaining access to private silo data is the answer we've been searching for and is exactly what Elroy enables greenhouse growers to provide with ease. Unlocking a portal to vast amounts of private data, coupled with advanced AI techniques and Elroy's compute power, data is no longer a limiting factor. We believe this will open the gates for quality AI development, enabling data scientists to create AI models that can provide greater efficiencies to multiple indoor grow facilities.

Having more universally validated models, with datasets at a ready-state for training or assessment, greenhouses gain the ability to choose which AI models best serve the current needs of their farms. This will ultimately increase efficiencies for greenhouses and reduce their production costs.

This is how TensoAI is supporting technology to make indoor grow facilities more efficient. We do this in a safe, secure and scalable way by running AI models at the data source--your greenhouse. Beyond efficiencies, our process is ideal for protecting the private data from your greenhouse business. That will be the focus of our next article. Stay tuned!