🎯 AI Gold Nugget 👑 Community Knowledge Sharing
Hello everyone!
We're excited to introduce a new format to foster knowledge sharing and collective growth in our community! It is a 100% virtual event, making it accessible from anywhere 🌐
_______________________________________________________________
⏩️ Speculative Decoding
Raphaël Vienne, Head of AI @ Datacraft
What if we could speed up LLM inference at barely no cost?
Google Deepmind researchers introduced Speculative Decoding, an innovative method allowing faster generation at inference time. Furthermore, we will break the no-free-lunch rule by introducing self-speculative decoding, which enhances this trick even further.
Paper links :
Speculative Decoding : https://arxiv.org/pdf/2211.17192
Self-Speculative Decoding : https://arxiv.org/pdf/2309.08168
_______________________________________________________________
🫦 Lip Syncing
Olivier Risser-Maroix, CSO @ Equos
Lip synchronization, talking head avatars, video dubbing — a lot of terms for a tremendous research topic!
At first glance, it may seem like a rather tough challenge to tackle. In this presentation, we’ll demystify the general framework for "how to lip sync" through a few key research paper architectures and we'll see that's not so hard and even… fun ? :D
Paper links :
Wav2Lip : https://arxiv.org/pdf/2008.10010
DINet : https://arxiv.org/pdf/2303.03988
_______________________________________________________________
Can't wait to learn together during our event!
💼 Are you looking to recruit or be recruited ? Take a look at our Job board : https://tinyurl.com/2vz6ph9c !
All the best,
Paris AI Society