Image de couverture de livre

Imbalanced Gradients in RL Post-Training of Multi-Task LLMs

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Wu, Runzhe, Samanta, Ankur, Jain, Ayush, Fujimoto, Scott, Kwon, Jeongyeol, Kretzu, Ben, Yu, Youliang, Hassani, Kaveh, Vidolov, Boris, Efroni, Yonathan
Format:	Preprint
Publié:	2025
Sujets:	Machine Learning Artificial Intelligence
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Self-Improvement of Language Models by Post-Training on Multi-Agent Debate
par: Samanta, Ankur, et autres
Publié: (2025)

Structure Enables Effective Self-Localization of Errors in LLMs
par: Samanta, Ankur, et autres
Publié: (2026)

Simple Optimizers for Convex Aligned Multi-Objective Optimization
par: Kretzu, Ben, et autres
Publié: (2025)

RL in Latent MDPs is Tractable: Online Guarantees via Off-Policy Evaluation
par: Kwon, Jeongyeol, et autres
Publié: (2024)

Credit Assignment with Resets in Language Model Reasoning
par: Samanta, Ankur, et autres
Publié: (2026)

Aligned Multi Objective Optimization
par: Efroni, Yonathan, et autres
Publié: (2025)

Hack-Verifiable Environments: Towards Evaluating Reward Hacking at Scale
par: Roth, Amit, et autres
Publié: (2026)

Exploiting Structure in Offline Multi-Agent RL: The Benefits of Low Interaction Rank
par: Zhan, Wenhao, et autres
Publié: (2024)

Gradient Free Deep Reinforcement Learning With TabPFN
par: Schiff, David, et autres
Publié: (2025)

Projection-Free Online Convex Optimization with Time-Varying Constraints
par: Garber, Dan, et autres
Publié: (2024)

Computationally Efficient RL under Linear Bellman Completeness for Deterministic Dynamics
par: Wu, Runzhe, et autres
Publié: (2024)

The Bias of Harmful Label Associations in Vision-Language Models
par: Hazirbas, Caner, et autres
Publié: (2024)

On the Complexity of First-Order Methods in Stochastic Bilevel Optimization
par: Kwon, Jeongyeol, et autres
Publié: (2024)

Generalizing Multi-Step Inverse Models for Representation Learning to Finite-Memory POMDPs
par: Wu, Lili, et autres
Publié: (2024)

Optimal Zeroth-Order Bilevel Optimization
par: Aghasi, Alireza, et autres
Publié: (2025)

On Penalty Methods for Nonconvex Bilevel Optimization and First-Order Stochastic Approximation
par: Kwon, Jeongyeol, et autres
Publié: (2023)

Making RL with Preference-based Feedback Efficient via Randomization
par: Wu, Runzhe, et autres
Publié: (2023)

Difficult Task Yes but Simple Task No: Unveiling the Laziness in Multimodal LLMs
par: Zhao, Sihang, et autres
Publié: (2024)

Two-Timescale Linear Stochastic Approximation: Constant Stepsizes Go a Long Way
par: Kwon, Jeongyeol, et autres
Publié: (2024)

A Classification View on Meta Learning Bandits
par: Mutti, Mirco, et autres
Publié: (2025)

An Empirical Study on the Power of Future Prediction in Partially Observable Environments
par: Kwon, Jeongyeol, et autres
Publié: (2024)

Time After Time: Deep-Q Effect Estimation for Interventions on When and What to do
par: Wald, Yoav, et autres
Publié: (2025)

A Comedy of Estimators: On KL Regularization in RL Training of LLMs
par: Shah, Vedant, et autres
Publié: (2025)

GAC: Stabilizing Asynchronous RL Training for LLMs via Gradient Alignment Control
par: Xu, Haofeng, et autres
Publié: (2026)

Partial Policy Gradients for RL in LLMs
par: Mathur, Puneet, et autres
Publié: (2026)

Exact analysis of potential flow past bodies of irregular shapes
par: Jain, Ankur
Publié: (2025)

ART: Adaptive Resampling-based Training for Imbalanced Classification
par: Basandrai, Arjun, et autres
Publié: (2025)

Detecting Causality with Symplectic Quandles
par: Jain, Ayush
Publié: (2023)

Improved Offline Contextual Bandits with Second-Order Bounds: Betting and Freezing
par: Ryu, J. Jon, et autres
Publié: (2025)

Exploration Hacking: Can LLMs Learn to Resist RL Training?
par: Jang, Eyon, et autres
Publié: (2026)

JigsawRL: Assembling RL Pipelines for Efficient LLM Post-Training
par: Hu, Zhengding, et autres
Publié: (2026)

Train on Validation (ToV): Fast data selection with applications to fine-tuning
par: Jain, Ayush, et autres
Publié: (2025)

Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation
par: Yang, Zhuolin, et autres
Publié: (2026)

SCOPE-RL: Stable and Quantitative Control of Policy Entropy in RL Post-Training
par: Wang, Chen, et autres
Publié: (2025)

Multimodal Emotion Recognition via Bi-directional Cross-Attention and Temporal Modeling
par: Byeon, Junhyeong, et autres
Publié: (2026)

Learning Dynamics in RL Post-Training for Language Models
par: Tomihari, Akiyoshi
Publié: (2026)

AI based Content Creation and Product Recommendation Applications in E-commerce: An Ethical overview
par: Jain, Aditi Madhusudan, et autres
Publié: (2025)

Efficient RL Training for LLMs with Experience Replay
par: Arnal, Charles, et autres
Publié: (2026)

Feed Two Birds with One Scone: Exploiting Wild Data for Both Out-of-Distribution Generalization and Detection
par: Bai, Haoyue, et autres
Publié: (2023)

Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing
par: Amico, Jeffrey, et autres
Publié: (2025)