Albert Gu

Timeline

1. 2022

   Modeling long sequences with S4

   With Karan Goel and Christopher Ré, Gu proposes S4, a structured state-space sequence model that offers a more efficient way to model long sequences. ICLR gives the paper an Honorable Mention.

2. 2023

   From Stanford to CMU

   After earning B.S. degrees in computer science and mathematics at Carnegie Mellon and a computer science Ph.D. at Stanford, he joins CMU's Machine Learning Department as an assistant professor.

3. 2023

   A selective memory for sequences

   With Tri Dao, he introduces Mamba, a selective state-space model for linear-time sequence modeling. The paper reports state-of-the-art performance across language, audio, and genomics.

4. 2024

   Connecting state spaces and attention

   The state-space duality framework connects state-space models with variants of attention. With Tri Dao, he uses it to design Mamba-2, whose core layer runs 2–8 times faster than Mamba's selective SSM while remaining competitive with Transformers on language modeling.

5. 2025

   Distilling language models into recurrence

   With four co-authors, he introduces Llamba, recurrent language models distilled from Llama-3.x into Mamba. The models achieve higher inference throughput and support larger batch sizes than Transformer-based models, with comparable benchmark performance.

6. 2026

   Designing Mamba-3 for inference

   Gu coauthors Mamba-3, which combines a more expressive recurrence, complex-valued state tracking, and a multi-input, multi-output option for better model performance without increasing decode latency.

7. As of October 2026

   Research at CMU and Cartesia

   As of October 2026, he is an assistant professor in Carnegie Mellon's Machine Learning Department and Cartesia's chief scientist and co-founder.