From a Theoretical Concept to a Microsoft Acquisition

From a Theoretical Concept to a Microsoft Acquisition

Bonsai empowers professionals across a variety of disciplines to teach machine learning systems using the expertise they’ve acquired through their craft, and apply that intelligence to real world devices, processes and applications.

  • 30X

    faster
    auto-calibration

  • 45X

    faster
    learning benchmark

Problem

Identifying an Impactful, Real World Application

Mark Hammond and Keen Browne founded Bonsai with the goal of democratizing Artificial Intelligence (AI) and machine learning.

“The spark that gave rise to Bonsai was really an insight, and the insight was that no matter how sophisticated we make machine learning algorithms, by design it's a learning system. So much focus is spent on how you build the co-machine learning algorithms themselves, and the teaching part has largely been overlooked,” explains Mark Hammond, Bonsai’s co-founder and CEO.

“Bonsai is an AI company focused on enabling people who don't necessarily have expertise in AI or machine learning to nonetheless build sophisticated control systems for their machinery and other applications.”

Having realized the importance of teaching as a mechanism for building sophisticated systems, Hammond and Browne needed an entirely new paradigm for programming machine learning systems. They also needed to determine the best real-world application.

At the time, the headlines in the mainstream press focused on AI’s potential in gaming and other consumer-focused applications. But Hammond and Browne’s ambitions for Bonsai were larger than that: “Games are great from a scientific and academic perspective,”says Hammond.“But our vision was to use AI to solve the toughest optimization and automation problems. It’s hard to see the connection between that and getting really good at playing a game.”

“We’ve done pioneering work with Siemens and Next47. Together we demonstrated the first real world application of reinforcement learning, and we’ve set state of the art benchmarks for using AI in the real world with real equipment on real problems.”

Mark Hammond

Co-Founder & CEO, Bonsai

Solution

Partner with Siemens and Next47

Bonsai found a partner in Next47 early in its lifecycle. “When I met Mark and Keen almost four years ago, they didn’t have much in terms of a product but I found their vision very compelling,” said Stuart Goose, Director at Next47 in Palo Alto. “At the time, what they really needed in a venture partner was help building use cases. I had a hunch that AI could be very interesting for industrial applications but there were really no good use cases out there, so this partnership would really be a win/win for Bonsai, Next47 and Siemens.”

Next47 quickly brokered introductions within Siemens and found that interest was high across the organization, especially within Siemens’s Digital Factory business. “We were all trying to learn how AI could make industrial robotics more efficient,” says Goose.

Over several months, Bonsai and the Siemens Digital Factory team worked together to automate the calibration ofComputer Numeric Control (CNC) machines, i.e. pieces of equipment that are used to manufacture industrial parts.

“Next47 quickly brokered introductions within Siemens and found that interest was high across the organization, especially within Siemens’s Digital Factory business. We were all trying to learn how AI could make industrial robotics more efficient.”

Stuart Goose

Director, Next47

Successes to Date

By working with Siemens, Bonsai was able to train an AI model to auto-calibrate a CNC machine more than 30 times faster than an expert human operator.

“This achievement was the first time deep reinforcement learning was successfully applied to auto-calibrate real world CNC machines, which are a vital component of the manufacturing industry,” says Hammond.

Bonsai also recently established a new reinforcement learning benchmark for programming industrial control systems. To demonstrate this breakthrough, the platform successfully trained a simulated robotic arm to grasp and stack blocks on top of one another by breaking down the task into simpler sub-concepts. Bonsai’s unique technique performed 45 times faster than a comparable approach from Google’s DeepMind.
Bonsai performs 45 times faster than Google’s DeepMind.

“This achievement was the first time deep reinforcement learning was successfully applied to auto-calibrate real world CNC machines, which are a vital component of the manufacturing industry.”

Mark Hammond

Co-Founder & CEO, Bonsai

Looking Ahead

In June 2018, Bonsai was acquired by Microsoft, placing the company in an optimal position to make AI more accessible and valuable to more people. By removing historical barriers to development, Bonsai will continue to work to empower any professional—regardless of their machine learning knowledge—to be an AI developer.

“We’ve done pioneering work with Siemens and Next47,” says Hammond. “It was never just about them writing us checks, or even doing business with Siemens. Together we demonstrated the first real world application of reinforcement learning, and we’ve set state of the art benchmarks for using AI in the real world with real equipment on real problems.”