OpenMined

🎙 Upcoming Privacy Tech Talk: What Should We Trust in Trustworthy Machine Learning? Join Andrew Trask February 24th at 1:30 PM ET (GMT-4) for an exciting presentation by Aaron Roth, Henry Salvatori Professor at UPenn. #Machinelearning is impressive, but imperfect — it makes errors. When we use ML predictions to take action, especially in high stakes settings, we want to be cognizant of this fact and take into account our probabilistic uncertainty. There are many ways of quantifying uncertainty, but what are they good for? Tune in to learn more. 🎥 The event will be live-streamed on YouTube. 🔗 Links to register and submit Q&A in the comments ↓ #PrivacyTechTalkSeries

AI researchers and dataset owners — The National AI Research Resource (NAIRR) Pilot is seeking datasets to help expand AI education and skill development across the U.S. Your dataset could play a key role in shaping the future of AI learning! We’re especially looking for datasets that: ✅ Support AI education in critical fields like health, climate, and manufacturing ✅ Help build a more diverse AI research community ✅ Come with clear documentation and support resources 📅 Deadline: February 7, 2025 We strongly encourage submissions from EPSCoR jurisdictions and Minority Serving Institutions to help create a more inclusive AI research ecosystem. Learn more & submit your dataset via link in the comments ↓

We are proud to be among the 50 projects selected by the Paris Peace Forum from 770 applications across 111 countries for the #AIActionSummit. 🗓️ We'll be in Paris on February 10-11 as we showcase ethical and inclusive AI solutions focusing on our important work with the Christchurch Call Foundation 👉 Read more about it on our blog: bit.ly/4hl59h2

🎥 Webinar: Curious about Privacy-Preserving Machine Learning (PPML)? Join our FREE webinars where you'll discover how to build AI systems that protect data privacy while delivering powerful insights. Two Sessions: • December 4th (This Wednesday!) • December 18th What you'll learn: • Core concepts of #PrivacyPreservingMachineLearning • Learn the basics of #FederatedLearning, #DifferentialPrivacy, and #HomomorphicEncryption. • Real-world applications and use cases using new tools Perfect for data scientists, ML engineers, and privacy enthusiasts looking to stay ahead in the evolving AI landscape. Learn More & Register now → https://bit.ly/3ZkxLj4

🎙Upcoming Privacy Tech Talk: Reconciling Computer Science and Legal Approaches to Privacy Join Andrew Trask on November 25th at 1PM ET for an insightful talk with Kobbi Nissim, Professor of Computer Science at Georgetown University and a pioneer of #DifferentialPrivacy. Kobbi will discuss the challenges of reconciling legal and technical approaches to privacy and explore how we can bridge these gaps to create systems that meet both legal standards and technical realities. 🎥 The event will be live-streamed on YouTube. Register now to receive a link: https://bit.ly/3Z5gTgn Submit your questions for the Q&A: https://bit.ly/4fG8C9D #PrivacyTechTalkSeries

Day 1: Federated Learning + Privacy Technologies: A First Dive 🔐🤖 #30DaysOfFLCode I aim to explore Federated Learning (FL) and how it enables collaborative machine learning while preserving privacy. My goal? Understand its core concepts, tackle real-world use cases, and apply cutting-edge tools like SyftBox. What is Federated Learning, and why does it matter? Federated Learning is a paradigm shift in how we train machine learning models. Instead of centralizing data (and risking privacy breaches), FL allows models to be trained locally on user devices or organizational datasets. The magic happens when these local models share their updates (e.g., weights or gradients) to create a global model—without exposing raw data. But FL isn't without challenges:   1️⃣ Communication Bottlenecks: Sharing updates across devices must be efficient.   2️⃣ Statistical Heterogeneity: Data across nodes isn't always uniformly distributed.   3️⃣ Privacy Concerns: Even sharing updates can leak sensitive information.  The paper I read today highlights some powerful solutions to these problems:   - Federated Averaging (FedAvg): Combines updates from local models to refine a global model.   - Differential Privacy: Adds noise to updates to protect individual data.   - Communication Efficiency: Compression techniques like sparsification reduce data transfer.   - Homomorphic Encryption: Enables encrypted computations, preserving privacy during model aggregation.  Paper link: https://lnkd.in/gzr-JZ6B Enter SyftBox: A Tool for Privacy-Preserving Collaboration SyftBox by OpenMined is like a "Dropbox for data and computation." It simplifies the implementation of privacy-enhancing technologies (PETs) like FL.  Key features:   - Distributed Network: Nodes (called Datasites) store and process data locally.   - Language Agnostic: Build APIs in any language and environment.   - Secure Collaboration: Datasites share model updates, not raw data, ensuring privacy.  I set up SyftBox today and explored its potential for real-world use cases. One exciting application is fraud detection in financial transactions. Imagine banks collaboratively training a fraud detection model without ever sharing sensitive customer data—this is what Federated Learning and SyftBox enable. What I accomplished today   1️⃣ Studied a foundational paper on Federated Learning and its techniques like FedAvg and Differential Privacy.   2️⃣ Set up SyftBox and ran a demo sharing CPU usage across a distributed network.   3️⃣ Planned how to apply FL concepts to credit card fraud detection, using techniques like Homomorphic Encryption to protect sensitive financial data.  👉 Check out my GitHub repo: https://lnkd.in/gDGcAm9Y

OpenMined recently completed a test pilot using GPU enclaves in the evaluation of an AI model across two organizations. Details here → https://bit.ly/3OiigTT

Very excited to be starting a 30-day course on Federated Learning (FL) with OpenMined. Having worked on the development of various privacy-enhancing technologies (PETs) through my PhD, I'm keen to understand how they're being used in this exciting area. LLMs are chewing their way through public data (the internet) but to harness the power of machine learning on private data sets we have to step carefully where privacy must me preserved. If we can do things right, huge possibilities are available to us in areas such as medical research, personalised education, and more. One of OpenMined's primary goals is to make sure PETs are embedded as standard practise when allowing training on private data and to make these tools open and accessible to all. #30DaysOfFLCode #PETs #MPC #FHE #ZKP

🚀 Introducing the #30DaysOfFLCode Challenge! Ready to dive into #FederatedLearning and other Privacy-Enhancing Technologies? This is your chance to learn, build, and share over 30 days of community-based coding fun. Whether you’re a beginner or a seasoned pro, this challenge is designed to: ✅ Expand your skills ✅ Connect with like-minded innovators ✅ Showcase your work in a supportive community How It Works: • Commit Publicly: Announce your participation to stay accountable. • Study Daily: Utilize free resources and tutorials provided by OpenMined and other challenge participants. • Share Progress: Post daily updates using the hashtag #30DaysOfFLCode. Join the movement and start your journey here → 30DaysOfFLCode.com Let’s code together and shape the future of AI and privacy-enhancing technologies. 💻✨

Excellent new research by Department for Science, Innovation and Technology on the AI assurance market. We are proud to see that our work with AI Safety Institute and the Christchurch Call to enable external scrutiny of AI systems is featured ↓