Marijuana Business Magazine September 2018

first tests, it could accurately distin- guish flower from leaves only 50% of the time. That led to a “horrifying” amount of lost flower, Gowa said. Bloom Automation improved the robot’s accuracy by introducing a con- volutional neural network to the new- est beta model and using more than 5,000 images of cannabis to improve the beta machine’s algorithm. A neural network is a mathematical model that works similarly to neurons in the brain. It takes an input – in this case, images of cannabis – and uses learned clues to identify parts of the plant. Bloom Automation’s team marked each image of cannabis with flower in one color and sugar leaf in another color. Then, they loaded the original image and the marked image into the machine for training. The more images they collected, the more often the computer correctly determined what was labeled flower. Typically, automation utilizes more conventional “machine vision,” Gowa said, which does a good job recogniz- ing known shapes, such as squares, circles, patterns, letters and numbers. For image recognition in agriculture, which isn’t always uniform, more sophisticated techniques, including artificial intelligence and machine learning, are needed, he said. Forming Partnerships Relationships with cannabis cul- tivators and processors have been critical to advancing Bloom Automa- tion’s technology. “A big challenge we faced entering the industry was getting exposed to operations,” Gowa said. His initial research about cultiva- tion operations was done online, and he used what he learned to reach out to licensed cultivators in California and Massachusetts. Now, Bloom Automation works with its licensed partners to test its product and gather images of various strains of cannabis, which continuously improves the robot’s algorithm. The result of the improvements – and Gowa’s networking – is that Bloom Automation’s beta model of the robot is now able to distinguish flower from leaves with 97% accuracy in plants the machine has tested, Gowa said. “Every customer we are working with now is enormously important to the quality and accuracy of the robot,” Gowa said. “Without our partners, it would be impossible to test, evaluate and improve the robots. We’ve found the industry very welcoming to innovation, which has been crucial to our pro- gress. By joining up with customers from Step One, we hope to produce a product that will integrate seamlessly with their operations.” ◆ IS A SEED-STAGE ACCELERATOR A GOOD FIT FOR YOUR ANCILLARY STARTUP? S eed-stage business accelerators and venture funds – such as Col- orado’s Canopy Boulder – can provide necessary networking, mentorship, education and funding opportunities to ancillary cannabis businesses that need a jump-start. Generally, funding and services are provided in exchange for equity. In Canopy Boulder’s case, it invests up to $80,000 per team for 6%-9.5% ownership in every company. Bloom Automation participated in Canopy Boulder’s spring 2017 program with 10 other ancillary startups that ranged from web-based application developers to retail ad management software companies and geolocated cannabis social networks. Overall, Bloom Automation’s experience at Canopy Boulder was extremely ben- eficial, according to CEO and founder John Gowa.The ready access to mar- ket information and mentors who help participants navigate research was also critical in Bloom Automation’s formative stages, Gowa said. “It didn’t hurt that the people at Canopy Boulder were so rooted in the industry that when you were looking for market information, not only was it readily available, you had individu- als there to guide you through the information: what regulations meant, or what licensing categories meant in particular states,” he said. Canopy Boulder’s team encouraged startups to have due diligence documents – business plans, financial projections, a certificate of incorporation and purchase orders or letters of intent frombuyers – on hand to better court investors. “An accelerator like Canopy Boul- der can be immensely important to jump-start ideas into more formal, fundable businesses,” Gowa said. “It gave us the tools and the founda- tion to really begin to build steam, in funding and in product and market development.” However, accelerators might not be the right fit for every startup, Gowa said. “A company that has already started to generate substantial, sustained rev- enue is likely beyond the stage where an accelerator can be helpful,” Gowa said. “It is important to weigh the ben- efits of a program that is more aimed at launching businesses from an early stage … for example, from ideation and prototyping to incorporated and pro- ducing product. We found the typical participant at Canopy was not yet a revenue-producing company.” – Joey Peña 92 • Marijuana Business Magazine • September 2018