CrowdFlower Announces First Round Winners of $1 Million AI For Everyone Challenge
Kiva Data Scientist And Stanford Researcher Honored For Facial Recognition And Disease Identification Work
CrowdFlower, the essential human-in-the-loop Artificial Intelligence platform for data science and machine learning teams, announced the first-round winners of its $1 million “AI for Everyone” Challenge. These first winning proposals are computer vision projects that will label millions of images to build the largest collection of training data libraries for facial recognition and living-cells.
Announced in May of this year, the “AI for Everyone” Challenge was created to help advance cutting-edge Artificial Intelligence projects. The challenge is granting eight awards to companies, organizations or individuals using AI to solve critical problems.
“The potential of AI to solve real problems is predicated on the quantity and quality of the training data that can be used to teach a machine learning model how to work in the human world,” said Robin Bordoli, CEO of CrowdFlower. “The goal of this challenge is to offer resources to ambitious data scientists everywhere to help jump start their ground-breaking projects that otherwise might take years to get started. Today’s winners Melissa and David and their bold projects targeting major societal contributions are examples of the type of work we want to support.”
Kiva.org engineer Melissa Fabros’ winning submission centers around the creation of the world’s largest, most diverse set of training data for facial recognition. One of the biggest problems with facial recognition today is the limited amount of training data available to teach an algorithm how to process the image. As a result, algorithms struggle to accurately process the faces of people across a wide range of skin colors or if the image isn’t perfectly clear or well lit. Kiva, which has been focused on crowdfunding micro-loans in across 80 countries has amassed a database of more than 900,000 human faces from varying global ethnicities.
As part of Kiva’s crowdfunding process, photos and descriptions of borrowers are reviewed by hundreds of volunteers each month before being posted to Kiva.org. Using the CrowdFlower Human-in-the-Loop AI platform, Kiva hopes that their volunteers will have a strong tool to help catch mistakes and make recommendations. This is turn will assist volunteers in reviewing more loans each month. Kiva will convert those raw images into detailed training data sets and make them available to academics researching machine learning algorithms and facial recognition capabilities of AI.
The second winning proposal centers around developing enough training data to enable machine learning platforms to better help medical researchers looking for cures to cancer and infectious diseases. Today, these researchers study the behavior of living cells both individually and collectively over time using a microscope. The challenge however is the work is manually intensive, extremely complex and time consuming.
David Van Valen, M.D., Ph.D, a Postdoctoral Fellow at Stanford University who will be starting his own research group at Caltech next fall as a new assistant professor believes AI can augment these researchers’ quest to rid the world of these diseases. To that end, Van Valen’s work at Stanford has shown that AI and deep learning systems can speed up researcher’s efforts, but success requires a massive amount of training data to teach a machine learning algorithm how to best determine the location, identity, and state of a cell.
Using the CrowdFlower platform, Van Valen’s project will label and catalogue thousands of images of mammalian cells as they change over time. By annotating at a pixel level, Van Valen will create a massive library of training data that can be used to train machine learning algorithms. These annotated datasets and trained algorithms will allow scientists to perform experiments that were previously impossible and enable them to make important discoveries about human disease states.
Finalists were selected by a group of distinguished judges including members of CrowdFlower’s Scientific Advisory Board: Barney Pell, founder at Moon Express; Pete Warden, Staff Research Engineer at Google; Monica Rogati, independent data science advisor; Adrian Weller, Senior Research Fellow at the University of Cambridge; Jack Clark, Director of Strategy and Communications at OpenAI and Lukas Biewald, founder at CrowdFlower. Selection is based on the innovation of the project, its importance to the advancement of AI and the overall potential impact of the proposed initiative.