analiza aiqbench
This research extends the concept of "edge coupling" to momentum-based optimizers like heavy-ball and Nesterov, formally linking their oscillatory dynamics at the edge of stability to discrete mechanics via symmetric Verlet action. By characterizing fixed points and two-point orbits through this functional's Hessian, the study provides a rigorous mathematical bridge between gradient descent behavior and conservative physical systems. This theoretical advancement offers deeper insights into the stability landscapes of modern optimization algorithms used in large-scale machine learning training.
Variational Structure at the Edge of Stability
arXiv:2608.21660v1 Announce Type: new Abstract: When discrete-time optimizers operate at the edge of stability, they exhibit near-two-periodic behavior. These oscillatory dynamics are reminiscent of conservative systems, such as the dynamics generated by symplectic integrators. However, a precise formulation of the connection between discrete-time optimizers at the edge of stability and discrete mechanics remains underexplored. Recently, Litman introduced the "edge coupling": a functional on consecutive gradient descent iterates whose critical points encode the fixed points and two-point orbits of the gradient descent dynamics. Here we extend the edge coupling to heavy-ball and Nesterov momentum. We show that its critical points characterize the fixed points and two-point orbits, with its Hessian characterizing their stability. We also show that the edge coupling can be identified with the symmetric Verlet action, formalizing the connection between the edge of stability and discrete mechanics.
Abstract:When discrete-time optimizers operate at the edge of stability, they exhibit near-two-periodic behavior. These oscillatory dynamics are reminiscent of conservative systems, such as the dynamics generated by symplectic integrators. However, a precise formulation of the connection between discrete-time optimizers at the edge of stability and discrete mechanics remains underexplored. Recently, Litman introduced the "edge coupling": a functional on consecutive gradient descent iterates whose critical points encode the fixed points and two-point orbits of the gradient descent dynamics. Here we extend the edge coupling to heavy-ball and Nesterov momentum. We show that its critical points characterize the fixed points and two-point orbits, with its Hessian characterizing their stability. We also show that the edge coupling can be identified with the symmetric Verlet action, formalizing the connection between the edge of stability and discrete mechanics.
Machine Learning (cs.LG)
arXiv:2608.21660 [cs.LG]
(or arXiv:2608.21660v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.21660 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 227 cuvinte
Subiecte conexe

Symmetry-enriched topological order in tensor networks: Defects, gauging and anyon condensation
Quantum Journal · 1 sept. 2026

News
Quantum Computing Report · 25 iul. 2026

Trapped Ions Demonstrate First Digital Fermionic Laughlin State
Quantum Zeitgeist · 10 iun. 2026
High-Dimensional Latents Should Be Diagnosed Through Phase Structure
arXiv · 3 iun. 2026