Insertion Transformers
Pablo discusses innovative approaches in language modeling, highlighting the insertion transformer that allows for token insertion at various points, potentially enhancing lyric generation. He also touches on the music transformer from the Magenta team, which offers a more coherent structure for music creation. Additionally, Pablo reflects on the complexities of integrating reinforcement learning into these models, emphasizing the importance of simplicity for stability and reproducibility.In this clip
From this podcast
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Music & AI Plus a Geometric Perspective on Reinforcement Learning with Pablo Samuel Castro - #339
Related Questions