The Limitations of Public Data in AI
Daniel Beutel, co-founder of Flower and a researcher at the University of Cambridge, highlights the constraints of relying solely on public data for AI training. He emphasizes that while public data forms only a small fraction of global data, distributed data, which is often trapped in devices or organizational silos, offers a more comprehensive and vast resource. However, this distributed data remains largely untapped in the current AI landscape.
Flower’s Vision for Decentralized AI Training
Founded in 2020 by Beutel, Taner Topal, and Nicholas Lane, Flower aims to decentralize the AI training process. The platform employs federated learning, allowing developers to train AI models on data distributed across numerous devices and locations without directly accessing the data. This approach ensures enhanced privacy and compliance, especially in scenarios where data sensitivity is paramount.
The Power of Federated Learning
Federated learning is not a novel concept, but its implementation has gained traction in recent years. The technique involves training AI algorithms on local data samples without exchanging the actual samples. Flower’s platform embodies this principle, ensuring that data remains at its source during training. Only the training results are transmitted and merged, keeping the data itself secure.
Flower recently unveiled FedGPT, a federated approach to training large language models (LLMs) akin to renowned models like ChatGPT and GPT-4. FedGPT enables organizations to train LLMs on globally distributed data, addressing concerns related to data privacy, leakage, and regional data movement restrictions.
Collaboration with Brave: The Dandelion Project
In a strategic partnership with the open-source web browser Brave, Flower is spearheading the Dandelion project. The initiative aims to establish an open-source federated learning system that encompasses the vast user base of Brave, which currently stands at over 50 million clients.
Flower’s Growing Influence
Flower has witnessed significant growth, with its developer community now exceeding 2,300 members. Several Fortune 500 companies and esteemed academic institutions, including Porsche, Bosch, Samsung, Banking Circle, Nokia, Stanford, Oxford, MIT, and Harvard, are among Flower’s users.
Investor Confidence and Future Plans
Flower’s innovative approach has garnered attention from investors such as First Spark Ventures, Hugging Face CEO Clem Delangue, Factorial Capital, Betaworks, and Pioneer Fund. With a recent funding of $3.6 million, Flower’s total investment now surpasses $115 million. The funds will be allocated to expand Flower’s core team, accelerate the development of its open-source software, and foster its ecosystem.
Addressing the AI Reproducibility Crisis
Beutel acknowledges the challenges faced by the AI community, particularly the reproducibility crisis in federated learning. He emphasizes the need for collaborative efforts to develop a comprehensive set of open-source federated techniques available on Flower for the global community.
In conclusion, Flower’s recent funding round underscores its commitment to revolutionizing the way AI models are trained. By championing federated learning and prioritizing data privacy, Flower is poised to lead the next wave of AI advancements.
Please email us your feedback and news tips at hello(at)dailycompanynews.com