• FedML has already created an open-source community, an enterprise platform, and a variety of software tools to help people collaborate on machine learning initiatives.
  • FedML hopes that its collaborative strategy will aid in overcoming the expense and complexity of large-scale AI development.

In order to train, deploy, and customize machine learning models anywhere, across thousands of edge- and cloud-hosted nodes, FedML Inc., a collaborative artificial intelligence startup, announced that it has closed on a USD 6 million seed financing round.

Acequia Capital, AimTop Ventures, Plug and Play Ventures, and LDV Partners also participated in the round that was headed by Camford Capital.

FedML has already created an open-source community, an enterprise platform, and a variety of software tools to help people collaborate on machine learning initiatives, despite only recently closing on its first round of financing. According to the business, they can achieve this by sharing data, models, and computing resources.

The goal of FedML is to develop an environment that will satisfy business needs for unique AI models. According to the article, many companies want to develop or improve AI models using their own data so they can use them for more specialized jobs like business automation, product design, customer service, and so forth. However, using cloud-based AI training systems can be challenging because this data is frequently extremely private, regulated, or siloed.

To solve this problem, FedML has developed a federated learning platform that enables developers to jointly train AI models on private or compartmentalized data right at the edge without having to move that data elsewhere.

This strategy is known as “learning without sharing” in FedML. Thus, a retail business could create models for tailored shopping suggestions without disclosing a customer’s private information. By using sparse and highly sensitive medical records that may be dispersed across several hospitals, a healthcare firm could create an AI model that could identify rare diseases.

Salman Avestimehr, Chief Executive and Co-founder of FedML, mentioned that future AI applications will rely on such collaborations. “We want to create a community that trains, serves and mines the best AI models. 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,” he added.

FedML hopes that its collaborative strategy will aid in overcoming the expense and complexity of large-scale AI development in addition to bringing the idea of federated learning to AI. Millions of dollars were spent on training that model by OpenAI LP, the firm that created ChatGPT.

Naturally, not all businesses can afford to spend that much on AI training, so only the largest tech companies can access the finest models. Not only is AI training costly, but it’s also complicated and demands a lot of knowledge that not all businesses have. FedML believes that these difficulties can be overcome with the help of its open-source, cooperative AI research community.

Chaoyang He, Chief Technology Officer and Co-founder of FedML, said, “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.”

After three years of development, FedML’s platform was released in March 2022 and has since surpassed Google LLC’s TensorFlow Federated as the most widely used open-source library for federated machine learning applications. The business has also developed the MLOps ecosystem, which allows users to teach machine learning models anywhere, including in the cloud or at the edge. More than 1,900 users of this ecosystem have implemented FedML on over 3,500 edge devices and trained over 6,500 models.

Additionally, the company has inked ten enterprise contracts with companies in the healthcare, financial services, retail, logistics, smart cities, Web3, and generative AI sectors.

Andy Thurai, Principal Analyst and Vice President at Constellation Research reported that FedML garnered a lot of traction after its launch last year due to the affordable pricing model and open-source libraries. However, he said that FedML is yet to make it big in terms of the full ML operations lifecycle. “More and more enterprises are looking toward full-cycle MLOps platforms because it’s difficult to bring best-of-breed ML models to market without them,” he added.

Thurai believes FedML has a lot of promise, particularly if the idea of developing more refined models using exclusive datasets gains traction. He claimed that FedML’s benefit is that it allows model training without the need for data sharing, which can be very helpful in regulated sectors and areas where data privacy is of particular 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. 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,” he added.

Ali Farahanchi, a partner at Camford Capital, expressed his admiration for FedML’s innovative technology and compelling vision, which will allow open and collaborative AI at scale even though there is still much work to be done. “In a world where every company needs to harness AI, we believe FedML will power both company and community innovation that democratizes AI adoption,” Ali added.