Ce que vous devez savoir avant
Vous commencez
Débute 6 June 2026 09:15
Se termine 6 June 2026
Systèmes d'IA Pareto-efficaces : Élargir la frontière de qualité et d'efficacité de l'IA
Paul G. Allen School
6076 Cours
1 hour 3 minutes
Amélioration optionnelle disponible
Not Specified
Progressez à votre rythme
Free Video
Amélioration optionnelle disponible
Aperçu
Join us as we delve into the intriguing world of Pareto-efficient AI systems with pioneering research led by Simran Arora. This event focuses on expanding the quality and efficiency frontier of AI by exploring the Pareto frontier that balances AI capabilities with efficient throughput.
Attendees will gain insights into cutting-edge developments, such as the introduction of BASED architecture and the innovative ThunderKittens programming library, both designed to optimize language models' performance.
Conveniently hosted on YouTube, this session is part of a series dedicated to advancing knowledge within the realms of Artificial Intelligence and Computer Science. Whether you are a seasoned expert or a curious novice, this event promises to enrich your understanding of quality-throughput tradeoffs and equip you with new tools and strategies in the rapidly evolving AI landscape.
Programme
- Introduction à l'efficacité de Pareto dans l'IA
- Le compromis qualité-débit dans les modèles de langage
- L'architecture BASED
- Bibliothèque de programmation ThunderKittens
- Expansion de la frontière de Pareto
- Applications pratiques et études de cas
- Outils et techniques pour un développement efficace de l'IA
- Orientations futures et opportunités de recherche
- Conclusion
Matières
Computer Science