Michael Poli

Also known as 宁致远

Michael Poli is an AI researcher whose work spans efficient neural architectures and genomic models. He coauthored Dissecting Neural ODEs in 2020 and Monarch, which received an ICML 2022 Outstanding Paper Runner Up. His Hyena research fed into HyenaDNA's million-token DNA context; he helped lead Stanford's Evo machine-learning team and coauthored STAR for architecture search. As of October 2026, he is a co-founder and Chief AI Scientist at Radical Numerics, where he works on Omnii-1, a multimodal genome language model.

Now at: Radical Numerics
Badges: Founder
Bridge
Links: Google Scholar, X, LinkedIn and GitHub

Timeline

2020

Neural ODE analysis

In 2020, Poli coauthors Dissecting Neural ODEs, which develops the continuous-depth formulation to clarify how design choices affect model dynamics.
2022

Monarch structured matrices

At ICML in 2022, he coauthors Monarch: Expressive Structured Matrices for Efficient and Accurate Training, which receives the conference's Outstanding Paper Runner Up. The paper proposes hardware-efficient structured matrices for training and fine-tuning neural networks.
March 2023

Hyena for long sequences

In March 2023, he coauthors Hyena, a subquadratic-time operator for large language models that uses long convolutions and gating.
2023

HyenaDNA for genomics

In 2023, he coauthors HyenaDNA, a genomic foundation model pretrained on the human reference genome with context lengths up to one million tokens at single-nucleotide resolution.
February–November 2024

Evo's genomic AI team

In February 2024, Arc researchers introduce Evo, a long-context biological foundation model across DNA, RNA, and proteins. By November 2024, Poli and Eric Nguyen spearhead its core machine-learning subteam, which develops its architecture, model training, and scaling infrastructure. His Chinese name is 宁致远.
November 2024

STAR architecture search

In November 2024, he coauthors STAR at Liquid AI. The method uses evolutionary search over architecture genomes to design hybrid models.
As of October 2026

Radical Numerics and Omnii-1

As of October 2026, he is a co-founder and Chief AI Scientist at Radical Numerics. In October 2026, he coauthors company research on Omnii-1, a multimodal genome language model that generates viable phages, prioritizes variants, and predicts phage-host interactions.