HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution

Explore the groundbreaking HyenaDNA genomic foundation model in this comprehensive talk by Eric Nguyen from Valence Labs. Delve into the challenges of modeling long-range genomic sequences at single nucleotide resolution and discover how HyenaDNA overcomes limitations of previous Transformer-based models. Learn about the innovative use of implicit convolutions to achieve context lengths of up to 1 million tokens, enabling more effective modeling of long-range interactions in DNA. Understand the model's sub-quadratic scaling, single nucleotide tokenization, and full global context capabilities. Examine the potential applications of in-context learning in genomics and review HyenaDNA's state-of-the-art performance on various benchmarks. The presentation covers motivation, background, model design, experimental results, and includes an extensive Q&A session, providing a thorough understanding of this cutting-edge approach to genomic sequence modeling.

- Intro
- Motivation
- Background
- Designing Hyena
- Experiments
- Q+A