Salman Avestimehr (left) and Chaoyang He’s new startup promises to make Federated Machine Learning possible anywhere (PHOTO CREDIT: USC Viterbi)
For three years, Salman Avestimehr, USC Viterbi Dean’s professor of electrical and computer engineering, and just-graduated PhD student Chaoyang He have been conducting cutting-edge research into the nascent field of federated machine learning. Now, the two collaborators have leveraged that experience into a new startup that promises to empower communities to build and connect AI anywhere at any scale. FedML, the Federated Learning/Analytics and Edge AI Platform just closed a (pre)seed fundraising of approximately $2 million. The startup is backed by Plug-and-Play, GGV Capital, Miracle Plus, Acequia Capital, and several other top-tier venture capital firms.
Federated learning, a field Avestimehr and He helped pioneer, is a solution to the data-sharing concern that the public and enterprise have (due to privacy, security, regulations, etc.) regarding the machine learning-powered technologies that modern society increasingly relies on. These technologies depend on huge, centralized data sets of sometimes sensitive information to make decisions on, say, patient health. This type of confidential information is susceptible to security breaches, making everyday people distrustful of the technology.
Fortunately, federated learning allows AI technologies to use these personal datasets without centralizing or transferring the data. It enables machine learning from decentralized data at various nodes without concentrating any data in the cloud (i.e., “learning without sharing”), which provides maximum user privacy, complies with regulations, and reduces development costs.
But FedML takes that revolutionary approach significantly further. “Our new platform brings zero-code, lightweight, cross-platform, and provably secure federated learning out of the lab and into the hands of businesses everywhere,” said Avestimehr. The company’s technology provides users with the tools necessary to simplify decentralized machine learning and easily apply it to their data, whether in finance, healthcare, insurance, or more. “Our technology allows anyone to transform their data into secure and trusted intelligence with minimum effort,” Avestimehr adds.
“FedML goes beyond federated machine learning. We view federated as a broad concept that connects all AI Elements, including data, models, training methods, and computing resources. Thus, FedML also stands for Foundational Ecosystem Design for ML,” said He. “Our mission is to build a community building and sharing AI anywhere at any scale. Essentially, we are willing to be the connector of AI productivity, a collaborative AI marketplace that connects developers, AI enterprise, and data infrastructure in a social, secure, and scalable manner.”
To date, FedML has more than 1000 users worldwide and has already been adopted by several industrial and academic institutions. A video introduction of the FedML platform can be seen here.
Published on May 2nd, 2022
Last updated on May 2nd, 2022
