Bagel 🥯

Recently, our founder, Bidhan R., joined NEAR Protocol’s AI Office Hours to share his vision for the future of machine learning, AI and blockchain. In this insightful conversation, Bidhan discusses how Bagel Network is leading the charge in reshaping AI with privacy-first solutions and decentralized technologies. Here are some key takeaways from the conversation: 🥯 Machine Unlearning - Future AI models will need to forget unnecessary data, streamlining processes while enhancing privacy and efficiency. 🥯 Decentralized AI and Blockchain Tech - Bagel is leading the way for the future of decentralized AI where users will own and monetize their data—changing the landscape of data ownership and privacy. 🥯 Data Privacy Leadership - As the demand for global AI regulation grow, Bidhan explains why privacy-preserving AI is essential and how Bagel is leading innovation in this space .… And so much more! We believe AI’s future is decentralized, and it’s built on privacy. Full episode link in the comments.

Our first-ever community event is officially in the books! We had the privilege of connecting with over a dozen teams, offering them exclusive early access to the Bakery, where they’re now building a host of new AI solutions. Missed the event? No problem—this is just the beginning. Watch the recap video to catch a glimpse of the incredible work being done, and stay tuned for exciting updates on the Bagel roadmap.   🥯 Join the waitlist: waitlist.bagel.net 👨🍳 Access the Bakery: bakery.bagel.net

What an incredible turnout at our Hackathon! We had over 200 community members join us to push the boundaries of AI development. Big thanks to betaworks and CoinFund for their partnership on this event. Here's a look at what you've missed. P.S. There's still time to participate. Register now at waitlist.bagel.net and get connected to a team!

There's nothing better than connecting with a community who use what you've built. Last night at the Bagel 🥯 Hack-AI-thon, I truly felt that. Thanks for having some NY style pizza with us. Few moments from the evening:

Machine Unlearning: AI Forgetting on Command Dive into our research on machine unlearning, where AI models can now selectively erase data on demand. Without the need for complete retraining. Join us on a journey to the forefront of AI model safety and data privacy. Read more: https://lnkd.in/dfDYmQYD

Looking forward to hosting Bagel 🥯 next Friday, 9/13 for their inaugural hybrid Hack-AI-thon. This is an exclusive opportunity for the brightest minds in AI and ML to come together and build custom solutions for critical global challenges. Register at https://lnkd.in/g_y_MrHS Hope to see you there!

What Could You Build with $5,000 and the Power of AI? In partnership with betaworks and CoinFund, we’re pleased to announce our inaugural hybrid Hack-AI-Thon — an exclusive opportunity for the brightest minds in AI and ML! Event Details: Join us in-person at 6 PM on 9/13 at the Betaworks office or participate virtually. Virtual Participation: Open until 12 PM ET on 09/26 (register at hackathon.bagel.net) Submissions will be accepted until 12 PM ET on 09/26. This is your chance to collaborate with leading ML/AI engineers, researchers, and data analysts to develop custom solutions for critical global challenges. Gain access to a thriving ecosystem of builders, showcase your innovations to top investors, and compete for $5,000 in cash prizes and Bagel credits. Why Participate? • Be among the first to gain access to Bagel’s unreleased product beta. • Experiment with Monetizing Your AI Developments: Fine-tune LLAMA3 models, utilize privacy-preserving datasets and earn rewards. • Win up to $5,000 in cash and 5,000 in Bagel credits. • Have your work showcased to Bagel’s growing community of builders, researchers and investors Who Should Join? AI Developers, Data Scientists, Python/JavaScript Devs, Data Providers, and Researchers/Technical Writers. Prize Categories: ⭐ Best Bagel Integration: 2,500 Bagel credits + Featured marketplace listing ⭐ Most Innovative Solution: 1,500 Bagel credits + Bagel blog feature ⭐ Best Platform Tutorial: 1,000 Bagel credits + Bagel developer badge 🏆 GRAND PRIZE: $5,000 USD + Bagel Credits to share with your team! Spots are limited to the first 25 teams —secure your place now: https://lu.ma/63lak3tl

Dangers of building AI in closed systems? They can easily go rogue, unchecked due to the mistakes of a small group of people. In 'A Beginner's Guide to AI' podcast episode, our founder Bidhan R. shared insights into why keeping AI in open systems is crucial for safety and transparency. Listen in! Episode link: https://lnkd.in/dirtJhHR

Understanding Decentralized Natural Policy Gradient for Multi-Agent Reinforcement Learning In decentralized multi-agent reinforcement learning (MARL), coordinating multiple agents is challenging, especially as the number of agents grows. Traditional centralized methods struggle with scalability, so a new approach called Momentum-based Decentralized Natural Policy Gradient (MDNPG) has been introduced. MDNPG optimizes how agents work together by using natural gradient descent, momentum-based variance reduction, and gradient tracking, all within a decentralized network where agents only communicate with their immediate neighbors. MDNPG is designed to handle the complexities of decentralized MARL, allowing each agent to update its policies independently while reducing the communication burden typical of centralized systems. The method’s use of momentum-based variance reduction helps to smooth out the randomness in gradient estimation, leading to faster and more efficient convergence compared to other decentralized methods. A key strength of MDNPG is its ability to scale effectively, with its sample complexity showing that it can handle more agents without losing accuracy or speed. The method has been shown to outperform other leading algorithms in various tests, making it a promising tool for developers working on reinforcement learning in decentralized settings. By incorporating advanced techniques such as variance reduction and natural gradient optimization, MDNPG offers a path forward for creating more efficient and effective multi-agent systems.

Accessing a fundamental resource like knowledge, for both human-driven and autonomous AI, should not entail a compromise on privacy. Read more 👇 https://lnkd.in/dTp67EBV