Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Sanyal, Aniket, Sidahmed, Baraah A. M., Burkholz, Rebekka, Chavdarova, Tatjana
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2601.18409
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910000985669632
author Sanyal, Aniket
Sidahmed, Baraah A. M.
Burkholz, Rebekka
Chavdarova, Tatjana
author_facet Sanyal, Aniket
Sidahmed, Baraah A. M.
Burkholz, Rebekka
Chavdarova, Tatjana
contents Learning in smooth games fundamentally differs from standard minimization due to rotational dynamics, which invalidate classical hyperparameter tuning strategies. Despite their practical importance, effective methods for tuning in games remain underexplored. A notable example is LookAhead (LA), which achieves strong empirical performance but introduces additional parameters that critically influence performance. We propose a principled approach to hyperparameter selection in games by leveraging frequency estimation of oscillatory dynamics. Specifically, we analyze oscillations both in continuous-time trajectories and through the spectrum of the discrete dynamics in the associated frequency-based space. Building on this analysis, we introduce \emph{Modal LookAhead (MoLA)}, an extension of LA that selects the hyperparameters adaptively to a given problem. We provide convergence guarantees and demonstrate in experiments that MoLA accelerates training in both purely rotational games and mixed regimes, all with minimal computational overhead.
format Preprint
id arxiv_https___arxiv_org_abs_2601_18409
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Frequency-Based Hyperparameter Selection in Games
Sanyal, Aniket
Sidahmed, Baraah A. M.
Burkholz, Rebekka
Chavdarova, Tatjana
Machine Learning
Learning in smooth games fundamentally differs from standard minimization due to rotational dynamics, which invalidate classical hyperparameter tuning strategies. Despite their practical importance, effective methods for tuning in games remain underexplored. A notable example is LookAhead (LA), which achieves strong empirical performance but introduces additional parameters that critically influence performance. We propose a principled approach to hyperparameter selection in games by leveraging frequency estimation of oscillatory dynamics. Specifically, we analyze oscillations both in continuous-time trajectories and through the spectrum of the discrete dynamics in the associated frequency-based space. Building on this analysis, we introduce \emph{Modal LookAhead (MoLA)}, an extension of LA that selects the hyperparameters adaptively to a given problem. We provide convergence guarantees and demonstrate in experiments that MoLA accelerates training in both purely rotational games and mixed regimes, all with minimal computational overhead.
title Frequency-Based Hyperparameter Selection in Games
topic Machine Learning
url https://arxiv.org/abs/2601.18409