Challenge: Large language models (LLMs) are used as action proposers in reinforcement learning (RL) but they are expensive and require prohibitive computational costs.
Approach: They propose a cache-efficient framework for Bayesian RL that leverages large language models as action proposers and optimizes meta-learning based on policy performance to enable efficient inference across text-based games and robotic control tasks.
Outcome: The proposed framework achieves 3.8–4.7 reduction in LLM queries and 4.0–12.0 lower median latencies (85–93ms on consumer hardware) while retaining 96–98% of the uncached policy’s performance.

Similar Papers

Enhancing Efficiency and Exploration in Reinforcement Learning for LLMs (2025.emnlp-main)

Copied to clipboard

Challenge: Existing approaches allocate an equal number of rollouts to all questions during the RL process, which is inefficient.
Approach: They propose a mechanism for dynamically allocating rollout budgets based on the difficulty of the problems, enabling more efficient RL training.
Outcome: The proposed model improves response precision while preserving exploratory ability to uncover potential correct pathways.
Cache Saver: A Modular Framework for Efficient, Affordable, and Reproducible LLM Inference (2025.findings-emnlp)

Copied to clipboard

Challenge: Inference is the major cost throughout the lifecycle of a large language model (LLM).
Approach: They propose a plug-and-play, asynchronous framework that facilitates high-level inference optimizations.
Outcome: The proposed framework reduces cost and CO2 by 35% across methods, tasks, and LLMs.
A Survey of Reinforcement Learning for Large Language Models under Data Scarcity: Challenges and Solutions (2026.acl-long)

Copied to clipboard

Challenge: Existing research on reinforcement learning for LLMs under data scarcity has not been unified.
Approach: They propose a top-up hierarchical framework built around three complementary perspectives: data-centric, training-centric and framework-centric.
Outcome: The proposed framework provides a clear conceptual foundation for understanding the design space of data-efficient RL for large language models and to guide researchers working in this emerging area.
Adaptive Constraint Propagation: Scaling Structured Inference for Large Language Models via Meta-Reinforcement Learning (2026.acl-long)

Copied to clipboard

Challenge: Large language models increasingly require structured inference, says aaron sagar . meta-learning learns universal constraint propagation policies without task-specific training . standard schedulers are inexpensive but myopic, because they optimize local effects .
Approach: MetaJuLS learns universal constraint propagation policies applicable across languages and tasks without task-specific retraining.
Outcome: MetaJuLS achieves 1.5-2.0 speedups over GPU-optimized baselines while maintaining accuracy within 0.2% of state-of-the-art parsers.
Memory-R1: Enhancing Large Language Model Agents to Manage and Utilize Memories via Reinforcement Learning (2026.acl-long)

Copied to clipboard

Challenge: Large Language Models (LLMs) are stateless and limited by a finite context window, preventing them from maintaining knowledge across long conversations or evolving tasks.
Approach: They propose a reinforcement learning framework that empowers LLMs to actively manage external memory through two specialized agents.
Outcome: The proposed framework outperforms baselines and benchmarks across diverse question types, three benchmarks, and multiple model scales.
Cache & Distil: Optimising API Calls to Large Language Models (2024.findings-acl)

Copied to clipboard

Challenge: Large Language Models are expensive to run and expose the entire request stream to external providers.
Approach: They propose to locally train a small private language model on the LLM's predictions to minimise the costs and data exposure associated with calling the API.
Outcome: The proposed model can handle an increasing number of user requests independently and is able to perform better than other policies and baselines across tasks and budgets.
Semantic-Aware Action Space Compression via LLM-DRL Synergy for Efficient Task-oriented Dialogue Policy Exploration (2025.findings-emnlp)

Copied to clipboard

Challenge: Pre-trained large language models (LLMs) with world knowledge and semantic understanding are promising for task-oriented dialogue systems.
Approach: a framework that synergizes pre-trained large language models with DRL is proposed . a lightweight action pruning mechanism is employed to eliminate implausible actions .
Outcome: a new framework synergizes pre-trained large language models with DRL to guide decision-making . the proposed framework eliminates semantically implausible or low-potential actions from multi-turn dialogue context .
Sparse-RL: Breaking the Memory Wall in LLM Reinforcement Learning via Stable Sparse Rollouts (2026.acl-long)

Copied to clipboard

Challenge: Existing methods for storing key-value caches during long-horizon rollouts cause performance collapses.
Approach: They propose a new training paradigm that empowers stable RL training under sparse rollouts.
Outcome: The proposed model reduces rollout overhead while maintaining the performance.
Boundary-Guided Policy Optimization for Memory-efficient RL of Diffusion Large Language Models (2026.acl-long)

Copied to clipboard

Challenge: Existing methods for reinforcement learning (RL) require a large sample size to be implemented.
Approach: They propose a memory-efficient RL algorithm that maximizes a lower bound of the ELBO-based objective.
Outcome: Experiments show that BGPO outperforms previous RL algorithms for diffusion large language models in math problem solving, code generation, and planning tasks.
Learning from Cognition: Enhancing RL Efficiency for LLM Reasoning via Hierarchical Metacognitive Decomposition and Refinement (2026.acl-long)

Copied to clipboard

Challenge: Recent advances in Large Language Models have demonstrated notable inferential capacities via reinforcement learning (RL) however, “zero-RL” approaches relying on fixed prompt templates introduce substantial sampling inefficiencies for weak LLMs.
Approach: They propose a hierarchical metacognitive RL framework that decomposes zero-accuracy problems into subproblems and prompts the policy to refine answers by referencing previous wrong solutions.
Outcome: The proposed framework improves sample utilization and sample efficiency and accelerates convergence compared to baselines.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations