Hyun Kim is the CEO and Co-Founder of Superb AI, a company that provides a new generation machine learning data platform to AI teams so that they can build better AI in less time. The Superb AI Suite is an enterprise SaaS platform built to help ML engineers, product teams, researchers and data annotators create efficient training data workflows.
What initially attracted you to the field of AI, Data Science and Robotics?
As an undergraduate majoring in Biomedical Engineering at Duke, I was passionate about genetics and how we can engineer our DNA to cure diseases or create genetically engineered organisms. I remember one wet-lab experiment distinctly that kept failing for like 6 months straight. The most frustrating part of it was that there was a lot of repetitive manual work, and in hindsight that was probably the root of some many potential errors.
That frustration led me to become interested in anything that has to do with automation. I basically floated around several labs at Duke until I joined one lab that was researching how machine learning can help diagnose Parkinson’s disease using brain MRI scans. Here I got a real taste for the game-changing potential of Deep Learning networks. I ended up pursuing a PhD program in computer science at Duke, and worked intensively at the Intelligent Robot Lab teaching robots how to learn things.
In 2016, you were involved in the Amazon Robotics Challenge. What were you working on and how did you enjoy this experience?
In Amazon Robotics Challenges, teams earn points by making robots autonomously pick and stow items in a given time. Robots in factories and assembly lines can be engineered to the specific object the robot is picking-and-placing, but the ARC challenges our robots to operate in very dynamic situations. I was the leader of “Team Duke” and its Motion Planning function. I designed and implemented robot motion planning methods for pick-and-place robot manipulation tasks in a realistic warehouse environment. It was an exciting learning experience as we had to put together a multitude of complex systems, from computer vision based robot perception systems to motion planning algorithms and custom-designed mechanical end-effector hardwares.
You then worked for nearly 2 years as a machine learning research engineer at SK T-Brain, what was this project?
About a year into my PhD, in March of 2016, Google’s AlphaGo defeated the Human champion, Lee Sedol, on Go. It was such groundbreaking news especially in Korea, where Go is much more popular than elsewhere.
After that event, the government and all major companies immediately started investing a ton in AI research. One of the companies was called SKT, or SK Telecom, a major Korean conglomerate. I was offered a Machine Learning Research Engineer position at their brand new AI research lab, called SKT Brain. I took a leave from my PhD, and went back to Korea to work for about 2 years.
The purpose of my team was to do research on various AI topics that could potentially spin off as a product, or create some business opportunity for the company. In those two years, I took some stabs at topics like self-driving cars, game AI (specifically StarCraft AI) and Generative Adversarial Networks or GANs.
After two years, instead of returning to school to finish my PhD program, I left to start my company, Superb AI.
What was the inspiration behind launching Superb AI?
While researching robot learning at school, and also while working at a corporate research lab, it was very apparent to me that most of my time was spent in data curation.
At school, I was spending most of my time creating simulated environments for robotics simulation data. At my previous company, I was spending time collecting and labeling datasets for self-driving and game AI.
And, sadly, that wasn’t just for me. It was the same for my colleagues, and a very common pain point for every researcher and engineer in the machine learning industry.
I wanted to fix that. As you can tell, I’m a big fan of machine learning and AI, and I think it can truly revolutionize our lives. I want tech breakthroughs to happen more quickly, and I want to see them applied to our everyday lives.
In order to make that happen, I initially researched making machine learning algorithms learn with less human input. That got me interested in things like unsupervised learning and meta-learning. After publishing a paper on GANs, I realized what I wanted to do. Instead of publishing research works, I wanted to make an actual product or service that can impact the industry and start solving the data problem right now.
How would you best describe the services that are offered by Superb AI?
Superb AI provides a machine learning data platform called the Superb AI Suite. Suite helps companies create, label and manage training data efficiently, and accelerate their ML Ops cycle.
It’s well known that the majority of machine learning teams spend over 50% of their time managing training datasets. We help engineers easily filter, search and manipulate training datasets, and integrate with their ML Ops stack, such as data storage or deep learning frameworks through a powerful SDK and API’s.
Product leaders also spend a lot of time with training data. We help make their lives easier through seamless issue tracking, data analytics and many collaboration and productivity related features.
Additionally, our auto-labeling feature, which utilizes many advanced ML techniques such as transfer learning and active learning, can assist the automatic labeling and quality control process.
What has been the most difficult aspect of building a machine learning data platform?
Building a machine learning data platform poses an interesting engineering challenge, not only because machine learning requires a tremendous amount of unstructured data, such as images and videos, but even more because the data must be constantly read and updated by numerous users around the world.
What are some companies that are using the Superb AI platform?
We have clients of varying sizes across many verticals. Large consumer electronics firms including Samsung and LG use our platform to manage data and accelerate their ML development process. Many enterprises and start-ups in the autonomous vehicles industry, as well as companies that utilize unmanned systems in applications from physical security to construction use our platform.
In addition, AR/VR and gaming companies use our training data platform to create and manage datasets that can teach ML models.
In the medical sector, research labs at renowned international universities use our platform to manage training data and more efficiently train Computer Vision models to recognize tumors within MRIs and CT scans.
Superb AI was a member of the Winter 2019 class of Y Combinator, could you describe this experience and what are some of the key takeaways that you learned?
Our two main takeaways were learning to focus on the users, and iterating quickly. YC’s motto is “make something people want”. Oftentimes startups, and especially tech startups, tend to focus on the technical innovations and neglect what the users actually need. Throughout the three month process, we learned to be extremely user-driven — talking to as many users as possible, and really trying to understand what they really need — while also iterating on product updates and messaging to fit the user needs. Since it’s impossible to nail a product or reach product-market-fit on the first try, it’s paramount to constantly talk to users and deliver what they want. It’s very obvious when you think about it, but very hard to focus on in practice.
Is there anything else that you would like to share about Superb AI?
We just beefed up our free product offering that gives users more raw asset storage and more labeling/data management/developer features and toolkits. We are on a real mission to democratize AI and want people to know that there are enterprise companies such as ourselves that are vested in providing as much as possible to further the adoption and integration of AI into our everyday lives.
Thank you for the interviews, readers who wish to learn more should visit Superb AI.
- Attention-Based Deep Learning Networks Could Improve Sonar Systems
- Cerebras CS-1 System Integrated Into Lassen Supercomputer
- Deepfaked Voice Enabled $35 Million Bank Heist in 2020
- Facebook: ‘Nanotargeting’ Users Based Solely on Their Perceived Interests
- IBM Announces AI-Driven Software for Environmental Intelligence