Collaborative artificial intelligence startup FedML Inc. announced today that it has closed a $6 million seed funding round that will help it bring enterprises and developers together to build machine learning models anywhere, across thousands of people Train, deploy, and customize edge and cloud-hosted nodes.
Today’s round was led by Camford Capital and included participations from Plug and Play Ventures, AimTop Ventures, Acequia Capital and LDV Partners.
Although FedML has just closed its first round of funding, it has already created an open source community, an enterprise platform, and various software tools that make it easy for people to collaborate on machine learning projects. They can do this by sharing data, models and computing resources, the company explained.
FedML’s mission is to create an ecosystem that meets enterprise needs for custom AI models. It says there are a number of companies that want to train or tweak AI models with their own data so they can use them for more specific tasks like business automation, customer service, product design, and so on. However, this data is often extremely sensitive and regulated or siled, making it difficult to use cloud-based AI training systems.
To overcome this, FedML created a federated learning platform that allows developers to train AI models together with private or isolated data at the edge without having to move that data elsewhere. FedML calls this approach “learning without sharing”. For example, a retail company could build models for personalized shopping recommendations without disclosing a customer’s private data. A healthcare company would be able to build an AI model that can detect rare diseases by training it on scarce and extremely sensitive healthcare records, potentially spread across multiple hospitals.
According to Salman Avestimehr, co-founder and chief executive of FedML, the future application of AI will depend on such collaborations. “We want to create a community that trains, serves and promotes the best AI models,” he said. “For example, we allow data owners to contribute their data to a machine learning task, and they can collaborate with AI developers or training specialists to create a custom machine learning model, and everyone is rewarded for their contributions.”
FedML not only brings the concept of federated learning to AI, but believes its collaborative approach will help overcome the cost and complexity of large-scale AI development. OpenAI LP, the company that developed ChatGPT, has spent millions of dollars training this model.
Of course, many companies don’t have that much cash to invest in AI training, which means the best models are reserved for only the biggest tech companies. AI training is not only expensive, but also very complex and requires significant expertise that not every company has. FedML believes these challenges can be overcome with its collaborative, open-source AI development community.
“We enable people to train anywhere and serve anywhere, from the edge to the cloud, enabling more cost-effective and decentralized AI development that is accessible to all,” said Chaoyang He, the other co-founder and chief technology officer of FedML.
Launched in March 2022 after three years of development, FedML’s platform has already surpassed Google LLC’s TensorFlow Federated as the most popular open source library for federated machine learning projects. Additionally, the company has created an MLOps ecosystem for training machine learning models anywhere at the edge or in the cloud. This ecosystem has 1,900+ users who have deployed 3,500+ edge devices with FedML and trained 6,500+ models.
The startup has also signed 10 enterprise agreements spanning industries such as healthcare, financial services, retail, logistics, smart cities, web3 and generative AI.
Andy Thurai, vice president and principal analyst at Constellation Research Inc., told SiliconANGLE that FedML has gained significant traction since its release last year thanks to its open-source libraries and more affordable pricing model. However, he said it had little impact on the entire machine learning lifecycle. “Enterprises are increasingly looking to full-cycle MLOps platforms because without them, it’s difficult to bring the best ML models to market,” he explained.
However, Thurai thinks FedML has a lot of potential, especially if the concept of training smaller models with private datasets works. He said the benefit of FedML is that it allows for model training without the need for data sharing, which can be extremely useful in regulated industries and regions where privacy is of particular concern, like the EU.
“If the concept of model training at the edge works with localized data, FedML can have a big impact on that,” Thurai said. “Right now, LLMs and ChatGPT-type models are crazy, and most companies are looking to bigger and better AI models, so changing that mindset will take time.”
While there is still work to be done, Ali Farahanchi, Partner at Camford Capital, said he was impressed by FedML’s compelling vision and unique technology that will enable open and collaborative AI at scale. “In a world where every company must leverage AI, we believe FedML will drive both corporate and community innovations that democratize AI adoption,” he said.
Image: FedML Show your support for our mission by joining our community of experts, Cube Club and Cube Event. Join the community that includes Andy Jassy, CEO of Amazon Web Services and Amazon.com, Michael Dell, Founder and CEO of Dell Technologies, Pat Gelsinger, CEO of Intel, and many more luminaries and experts.
