Flower Labs

Flower is at DISC 2024! Chong Shen Ng, Ph.D. , Research Engineer, and Adam Narożniak, Machine Learning Engineer, are in 🇪🇸 Spain where they gave a hands-on tutorial on federated learning and using the Flower framework! Yesterday's session covered core topics, including: privacy and security in FL, training large language models with Flower, and featured a hands-on demo of an FL environment 🚀 If you're at DISC 2024, feel free to connect with Chong and Adam today! https://buff.ly/4gYoghR

Fine-tuning LLMs with FlowerTune helps create domain-specific LLMs, maintain data privacy, and distribute the compute load. Flower Labs has now launched the FlowerTune LLM Leaderboard where AI devs and researchers can try to improve the core FlowerTune library and share everything open-source. Check it out!

🔔 FlowerTune LLM Leaderboard is Live! ⚡️ Let's work 🤝 together to improve and simplify methods for fine-tuning LLMs using federated learning. Build task-specific LLMs safely using private data. 😎 We offer 4 challenges to explore: general NLP, finance, medical, and code. Thank you to our partners on this exciting initiative: 🧠 Mistral AI and 🏫 DeepLearning.AI. 🤖 Submissions are already underway. It is time to get started, and see how well your solution ranks globally! 🌍 📺 Video detailing how to participate: https://lnkd.in/e_w5wgKa ➡️ Complete launch details: https://lnkd.in/ef2Ud8Fr 🦙 Visit the leaderboard: https://lnkd.in/efavqnX3 👨🔬 FlowerTune LLM Leaderboard Champion: Yan Gao -- Research Scientist at Flower Labs; contact him for more details

🎉 Excited to share that I have completed the "Intro to Federated Learning" course by DeepLearning.AI, taught by the inspiring Andrew Ng and Daniel J. Beutel! Thank you both for an enlightening experience that has opened new doors in my understanding of this emerging field. Federated Learning represents a new paradigm in AI, where data privacy is at the forefront. Instead of gathering data in one place, this technique enables models to be trained across decentralized devices, safeguarding sensitive information while allowing organizations to leverage data-driven insights. This approach promises to reshape industries reliant on sensitive data, from healthcare to finance, creating more secure and privacy-conscious AI applications. Looking forward to applying these concepts and exploring this innovative frontier further. 🌐🔐

Another 🌼 Flower Milestone: 5K stars on GitHub! 🎉 Big thanks to our amazing community and contributors for helping us get here! 🙌 https://buff.ly/44xnvVz

October 🌼 Flower Monthly recordings are available now: https://buff.ly/3Yg8MNr 👉🏽 “Federated Learning in Insurance” by Haoyuan Harry Loh -- ERM Actuary; Yung-Yu Michelle Chen -- Capital Actuary at AIG; Scott Hand -- Actuarial Analyst at Legal & General and Dylan Liew, Head of Technical Pricing & Data -- Bupa https://buff.ly/3YlOKRX 👉🏽 “Protecting Federated Learning from Extreme Model Poisoning Attacks via Multidimensional Time Series Anomaly Detection” by Edoardo G. -- PhD Student at Sapienza Università di Roma (Sapienza University of Rome) https://buff.ly/3UjJGfm Next Flower Monthly 6th Nov 16:00 UTC https://buff.ly/4fGVIbJ

Last week, every member of team 🌼 Flower traveled to Nice, France 🇫🇷 to plan and further develop our next big steps in federated AI. As a growing remote-first AI organization, people came from all parts of the 🌎 globe for a fun and productive week of in-person collaboration. We also found time to enjoy and explore this wonderful region, we left even more energized for the journey ahead! Join us by becoming part of the Flower community -- https://buff.ly/3PjWcsY -- and help us build an AI future for everyone!

🌼 Flower Datasets 0.4.0 is released! Enjoy the benefits from the updates right now: `pip install -U flwr-datasets[vision]` Exciting new features include: ✔️ New Partitioner: GroupedNaturalIdPartitioner 🧑🤝🧑 - group the dataset with natural ids into groups (e.g., to transform your dataset from cross-device to cross-silo) ✔️ New Partitioner: SizePartitioner 📊 - introduce size skew in any imaginable way ✔️ Docs Improvement: 100% 🥳 of the code has an Examples section in the docstring ✔️ Tool to dynamically create a federated dataset (without coding) and visualize results ✔️ Improved tutorial 📔 on working with local data + clearer error messages 💣 Check out the full release notes: https://buff.ly/3YhDWnx 🚀 Join the 🌼 Flower community: https://buff.ly/3YONpTy

✨ Today, I had the honor of presenting our work at the 17th International Workshop on Selected Topics in Wireless and Mobile Computing, held at CNAM Paris. 📍 Our paper, titled Efficient Federated Intrusion Detection in 5G ecosystem using optimized BERT-based model, tackles the challenges of intrusion detection within the 5G ecosystem, leveraging an optimized BERT model and a federated learning approach. Thanks to Flower Labs, whose framework significantly contributed to the success of our simulations and greatly enhanced their efficiency.🌍💻 Special thanks to my co-authors Moez Esseghir (HDR) and prof Leila Merghem Boulahia, whose mentorship and invaluable insights have played a pivotal role in kickstarting this exciting journey.🙌 This is just the first of many to come in what promises to be a long and exciting journey. 🚀 #research #5G #federatedlearning #AI #BERT #flowerlabs #cybersecurity #CNAMParis #UTT.

Let's take federated LLM fine-tuning mainstream!