Imagen de Portada

Approximation of Log-Partition Function in Policy Mirror Descent Induces Implicit Regularization for LLM Post-Training

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Xu, Zhenghao, Lu, Qin, Yu, Changlong, Zhao, Tuo
Formato:	Preprint
Publicado:	2026
Materias:	Machine Learning
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

Sample Complexity of Neural Policy Mirror Descent for Policy Optimization on Low-Dimensional Manifolds
por: Xu, Zhenghao, et al.
Publicado: (2023)

On the Effect of Regularization in Policy Mirror Descent
por: Kleuker, Jan Felix, et al.
Publicado: (2025)

Convergence of Policy Mirror Descent Beyond Compatible Function Approximation
por: Sherman, Uri, et al.
Publicado: (2025)

A Unified Approach to Controlling Implicit Regularization via Mirror Descent
por: Sun, Haoyuan, et al.
Publicado: (2023)

Functional Acceleration for Policy Mirror Descent
por: Chelu, Veronica, et al.
Publicado: (2024)

Activation Sensitivity as a Unifying Principle for Post-Training Quantization
por: Xu, Bruce Changlong
Publicado: (2026)

Ask a Strong LLM Judge when Your Reward Model is Uncertain
por: Xu, Zhenghao, et al.
Publicado: (2025)

The Hidden Cost of Approximation in Online Mirror Descent
por: Schlisselberg, Ofir, et al.
Publicado: (2025)

Policy Mirror Descent with Lookahead
por: Protopapas, Kimon, et al.
Publicado: (2024)

Mirror, Mirror of the Flow: How Does Regularization Shape Implicit Bias?
por: Jacobs, Tom, et al.
Publicado: (2025)

Implicit Bias and Convergence of Matrix Stochastic Mirror Descent
por: Akhtiamov, Danil, et al.
Publicado: (2026)

On the Convergence of Policy in Unregularized Policy Mirror Descent
por: Lin, Dachao, et al.
Publicado: (2022)

One-Step Flow Policy Mirror Descent
por: Chen, Tianyi, et al.
Publicado: (2025)

Stability and Robustness via Regularization: Bandit Inference via Regularized Stochastic Mirror Descent
por: Halder, Budhaditya, et al.
Publicado: (2026)

Think-RM: Enabling Long-Horizon Reasoning in Generative Reward Models
por: Hong, Ilgee, et al.
Publicado: (2025)

Flow Matching Policy Optimization with Mirror Descent and Entropy Constraints
por: Gao, Ting, et al.
Publicado: (2026)

Efficient Training of Boltzmann Generators Using Off-Policy Log-Dispersion Regularization
por: Schopmans, Henrik, et al.
Publicado: (2026)

On the Convergence of Policy Mirror Descent with Temporal Difference Evaluation
por: Liu, Jiacai, et al.
Publicado: (2025)

StaQ it! Growing neural networks for Policy Mirror Descent
por: Shilova, Alena, et al.
Publicado: (2025)

Beyond State-Wise Mirror Descent: Offline Policy Optimization with Parametric Policies
por: Li, Xiang, et al.
Publicado: (2026)

Dr. Post-Training: A Data Regularization Perspective on LLM Post-Training
por: Hu, Pingbang, et al.
Publicado: (2026)

Understanding the Implicit Regularization of Gradient Descent in Over-parameterized Models
por: Ma, Jianhao, et al.
Publicado: (2025)

Path-Space Mirror Descent for On-Policy Reinforcement Learning under the Generalized Schrödinger Bridge
por: Gong, Yuehu, et al.
Publicado: (2026)

Stochastic MeanFlow Policies: One-Step Generative Control with Entropic Mirror Descent
por: Wang, Zeyuan, et al.
Publicado: (2026)

Policy Regularized Distributionally Robust Markov Decision Processes with Linear Function Approximation
por: Gu, Jingwen, et al.
Publicado: (2025)

Improving LLM General Preference Alignment via Optimistic Online Mirror Descent
por: Zhang, Yuheng, et al.
Publicado: (2025)

Mirror Descent on Riemannian Manifolds
por: Jiang, Jiaxin, et al.
Publicado: (2026)

Parameter-free Mirror Descent
por: Jacobsen, Andrew, et al.
Publicado: (2022)

Optimistic Online Mirror Descent for Bridging Stochastic and Adversarial Online Convex Optimization
por: Chen, Sijia, et al.
Publicado: (2023)

A Mirror Descent Perspective of Smoothed Sign Descent
por: Wang, Shuyang, et al.
Publicado: (2024)

A Universal Banach--Bregman Framework for Stochastic Iterations: Unifying Stochastic Mirror Descent, Learning and LLM Training
por: Zhang, Johnny R., et al.
Publicado: (2025)

Implicit Regularization for Tubal Tensor Factorizations via Gradient Descent
por: Karnik, Santhosh, et al.
Publicado: (2024)

Mirror Descent Policy Optimisation for Robust Constrained Markov Decision Processes
por: Bossens, David M., et al.
Publicado: (2025)

Mask in the Mirror: Implicit Sparsification
por: Jacobs, Tom, et al.
Publicado: (2024)

Mirror Descent-Type Algorithms for the Variational Inequality Problem with Functional Constraints
por: Alkousa, Mohammad S., et al.
Publicado: (2026)

COSMOS: A Hybrid Adaptive Optimizer for Memory-Efficient Training of LLMs
por: Liu, Liming, et al.
Publicado: (2025)

Asymptotic Behavior of Multi--Task Learning: Implicit Regularization and Double Descent Effects
por: Alrashdi, Ayed M., et al.
Publicado: (2026)

Reference-Sampled Boltzmann Projection for KL-Regularized RLVR: Target-Matched Weighted SFT, Finite One-Shot Gaps, and Policy Mirror Descent
por: Shu, Yao, et al.
Publicado: (2026)

Implicit Regularization of the Deep Inverse Prior Trained with Inertia
por: Buskulic, Nathan, et al.
Publicado: (2025)

Rank-Induced PL Mirror Descent: A Rank-Faithful Second-Order Algorithm for Sleeping Experts
por: Zhang, Tiantian
Publicado: (2025)