Collaborative machine learning startup FedML raises $6M to train, deploy and customize AI anywhere

Collaborative artificial intelligence startup FedML Inc. said today it has closed on a $6 million seed funding round that will help it bring together companies and developers to train, deploy and customize machine learning models anywhere, across thousands of edge- and cloud-hosted nodes.

Today’s round was led by Camford Capital and saw participation from Plug and Play Ventures, AimTop Ventures, Acequia Capital and LDV Partners.

Despite only just closing on its first round of funding, FedML has already created an open-source community, enterprise platform and various software tools that make it easier for people to collaborate on machine learning projects. They can do this by sharing data, models and compute resources, the company explained.

FedML’s mission is to create an ecosystem that will meet enterprise demands for custom AI models. It says that there are a number of businesses that want to train or fine-tune AI models on their own data so they can leverage them for more specific tasks such as business automation, customer service, product design and so on. But this data is often extremely sensitive and regulated, or else siloed, making it difficult to use cloud-based AI training systems.

To overcome this, FedML has created a federated learning platform that makes it possible for developers to collaboratively train AI models using private or siloed data at the edge, without needing to move that data anywhere else. FedML calls this approach “learning without sharing.” So, for example, a retail company could build models for personalized shopping recommendations without exposing a customer’s private data. A healthcare company would be able to build an AI model that’s able to detect rare diseases by training it on scarce and extremely sensitive healthcare records that might be spread across multiple hospitals.

According to FedML co-founder and Chief Executive Salman Avestimehr, the future application of AI will depend on these kinds of collaborations. “We want to create a community that trains, serves and mines the best AI models,” he said. “For example, we enable data owners to contribute their data to a machine learning task, and they can work with AI developers or training specialists to build a customized machine learning model, and everyone gets rewarded for their contributions.”

Besides bringing the concept of federated learning to AI, FedML believes its collaborative approach will help to overcome the cost and complexity of large-scale AI development. OpenAI LP, the company that built ChatGPT, spent millions of dollars to train that model.

Of course, many companies do not have that kind of money to throw at AI training, meaning that the best models are limited to only the biggest technology firms. AI training is not only expensive, but also very complex, requiring significant expertise that not every company has. FedML reckons these challenges can be overcome with its collaborative, open-source AI development community.

“We allow people to train anywhere and serve anywhere, from edge to cloud, enabling lower-cost and decentralized AI development that’s accessible to everyone,” said FedML’s other co-founder and Chief Technology Officer Chaoyang He.

FedML’s platform was launched in March 2022 after three years of development, and has already surpassed Google LLC’s TensorFlow Federated as the most popular open-source library for federated machine learning projects. In addition, the company has created an MLOps ecosystem for training machine learning models anywhere at the edge or in the cloud. This ecosystem has more than 1,900 users, who have deployed FedML more than 3,500 edge devices and trained more than 6,500 models.

The startup has also signed 10 enterprise contracts spanning industries such as healthcare, financial services, retail, logistics, smart cities, Web3 and generative AI.

Constellation Research Inc. Vice President and Principal Analyst Andy Thurai told SiliconANGLE that FedML has gained quite a bit of traction since its release last year, thanks to its open-source libraries and its cheaper pricing model. However, he said it has barely made a dent in terms of the full machine learning operations lifecycle. “More and more, enterprises are looking towards full cycle MLOps platforms because it’s difficult to bring best-of-breed ML models to market without them,” he explained.

That said, Thurai thinks FedML offers a lot of potential, especially if the concept of training smaller models using private datasets takes off. He said FedML’s advantage is it enables model training without needing to share data, which can be extremely useful in regulated industries and regions where data privacy is of special importance, such as the EU.

“If the concept of model training at the edge using localized data takes off, then FedML can have a big impact on this,” Thurai said. “For now, LLMs and ChatGPT-type models are the craze, with most enterprises going for bigger and better AI models, so it will take some time to change that mindset.”

Though a lot of work remains to be done, Camford Capital’s partner Ali Farahanchi said he was impressed with FedML’s compelling vision and unique technology, which will enable open and collaborative AI at scale. “In a world where every company needs to harness AI, we believe FedML will power both company and community innovation that democratizes AI adoption,” he said.

Image: FedML