Through the River: Understanding the Benefit of Schedule-Free Methods for Language Model Training
Minhak Song, Beomhan Baek, Kwangjun Ahn, Chulhee Yun
NeurIPS 2025 | July 2025
Kwangjun Ahn
Senior Researcher
Publications
-
-
-
Dion: Distributed Orthonormalized Updates
Kwangjun Ahn, Byron Xu, Natalie Abreu, John Langford
May 2025
-
Does SGD really happen in tiny subspaces?
Minhak Song, Kwangjun Ahn, Chulhee Yun
ICLR 2025 | May 2025
-
-
-
Adam with model exponential moving average is effective for nonconvex optimization
Kwangjun Ahn, Ashok Cutkosky
NeurIPS 2024 | December 2024
-
General framework for online-to-nonconvex conversion: Schedule-free SGD is also effective for nonconvex optimization
Kwangjun Ahn, Gagik Magakyan, Ashok Cutkosky
ICML 2025 | November 2024
-
The Belief State Transformer
Edward S. Hu, Kwangjun Ahn, Qinghua Liu, Haoran Xu, Manan Tomar, Ada Langford, Dinesh Jayaraman, Alex Lamb, John Langford
ICLR 2025 | October 2024
-
-
-
Through the River: Understanding the Benefit of Schedule-Free Methods for Language Model Training
Minhak Song, Beomhan Baek, Kwangjun Ahn, Chulhee Yun
NeurIPS 2025 | July 2025
-
Dion: Distributed Orthonormalized Updates
Kwangjun Ahn, Byron Xu, Natalie Abreu, John Langford
May 2025
-
Does SGD really happen in tiny subspaces?
Minhak Song, Kwangjun Ahn, Chulhee Yun
ICLR 2025 | May 2025
-
Adam with model exponential moving average is effective for nonconvex optimization
Kwangjun Ahn, Ashok Cutkosky
NeurIPS 2024 | December 2024
-
General framework for online-to-nonconvex conversion: Schedule-free SGD is also effective for nonconvex optimization
Kwangjun Ahn, Gagik Magakyan, Ashok Cutkosky
ICML 2025 | November 2024
-
The Belief State Transformer
Edward S. Hu, Kwangjun Ahn, Qinghua Liu, Haoran Xu, Manan Tomar, Ada Langford, Dinesh Jayaraman, Alex Lamb, John Langford
ICLR 2025 | October 2024
-
-
-
Through the River: Understanding the Benefit of Schedule-Free Methods for Language Model Training
Minhak Song, Beomhan Baek, Kwangjun Ahn, Chulhee Yun
NeurIPS 2025 | July 2025
-
Does SGD really happen in tiny subspaces?
Minhak Song, Kwangjun Ahn, Chulhee Yun
ICLR 2025 | May 2025
-
Adam with model exponential moving average is effective for nonconvex optimization
Kwangjun Ahn, Ashok Cutkosky
NeurIPS 2024 | December 2024
-
General framework for online-to-nonconvex conversion: Schedule-free SGD is also effective for nonconvex optimization
Kwangjun Ahn, Gagik Magakyan, Ashok Cutkosky
ICML 2025 | November 2024
-
The Belief State Transformer
Edward S. Hu, Kwangjun Ahn, Qinghua Liu, Haoran Xu, Manan Tomar, Ada Langford, Dinesh Jayaraman, Alex Lamb, John Langford
ICLR 2025 | October 2024
-
-
-
Dion: Distributed Orthonormalized Updates
Kwangjun Ahn, Byron Xu, Natalie Abreu, John Langford
May 2025
-