Labelbox•April 9, 2019
Dear Labelbox Community,
Today, we’re excited to share a significant update: we recently secured a Series A funding round of $10 million. Our lead investor for this new phase of our company is Gradient Ventures, Google’s AI-focused venture fund, and we thank them for their support. We also received a repeat investment from Kleiner Perkins and First Round Capital. Anna Patterson, founder and Managing Partner at Gradient Ventures, VP of Engineering at Google, and Square board member, is joining the Labelbox board. We’re extremely grateful to our customers, investors, advisors, partners, and the broader Labelbox community for their support in our continued success.
In this post, we’d like to share our thoughts on where Labelbox is today and our plans for the future.
The confluence of accessible GPU compute and deep learning technology has lead to a leap in what’s possible in computer vision. We’re now seeing this new capability driving innovation in every industry, from transportation and healthcare to agriculture and insurance. At Labelbox we see this happening every day. The results are striking and real.
We work closely with data science teams as they mature their deep learning systems to production. Through this experience, it’s become clear that the hard part is teaching these systems to make the right decisions for us, not just some of the time, but every time.
Training performant deep neural networks require a lot of labeled data, often hundreds of thousands to millions of labeled images. To get this done, we enlist teams of people to label. However, if the labeling efforts result in inaccurate or inconsistent training data then the model will exhibit poor decision making performance.
It turns out that not only do we need a lot of labeled data to train performant deep learning models, but that the labeled data needs to be consistent and accurate. In other words, the models we build exist to make decisions on behalf of us as humans and should reflect that with every decision, even as the model encounters new environments and edge cases. Herein lay the toughest challenge in building production deep learning powered computer vision systems. At Labelbox, we’re driven by the opportunity to solve this problem for the world.
Labelbox is building software for industrial data science teams to do data labeling and management for the training of neural networks. When we build software, we take for granted the existence of collaborative tools to write and debug code. The machine learning workflow has no standard tooling for labeling data, storing it, debugging models and then continually improving model performance.
Labelbox's mission is to enable data science teams to build great machine learning applications by creating software for data scientists to manage data and train neural networks in the same way that GitHub and text editors are defaults for software engineers.
Building great machine learning applications requires a huge amount of high quality labeled training data. In fact, eighty percent of the time spent on developing machine learning is related to data management, which slows innovation and results in long build-test cycles. This translates to 2-4 weeks of calendar time, a preposterous amount when compared to software development. There is a better way, one in which data scientists are empowered to build great machine applications through faster iteration cycles and powerful tooling that unlocks new capabilities.
We’re committed to this future at Labelbox. With this new funding, we will continue building the best and most comprehensive solution for teams to create and manage labeled training data for computer vision applications. To do this, we plan to double our headcount in 2019 by hiring additional talent to fill engineering, sales, marketing, and customer success roles. If you are passionate about computer vision and deep learning, come join us!
When humans have better ways to input and manage data, machines have better ways to learn.
Labelbox Founders – Manu, Dan, & Brian
Labelbox•December 12, 2018
How to Scale Training Data
A Guide to Outsourcing Without Compromising Data Quality In order for data science teams to outsource annotation to a managed workforce provider — also known as a Business Process Outsourcer (BPO) — they must first have the tools and infrastructure to store and manage their training data. Data management tools and infrastructure should support R&D product management teams, outsourced labeling teams, and internal labeling and review teams working together in a single centralized place with fully
Labelbox•August 26, 2018
Labelbox August 2018 Updates
Labelbox Raises $3.9M & Product Updates We are excited to introduce teams, customer success stories and $3.9M in seed funding from Kleiner Perkins [https://www.kleinerperkins.com/], Gradient Ventures [https://gradient.google/], Google’s new AI fund and First Round Capital [http://firstround.com/]. Read on to learn more. Introducing Teams Building AI often means collaboration from many different functions, including software, management, operations, domain expertise, and data science. For these