sample-efficiency

This paper introduces Product-Unit Residual Networks (PURe), which integrate multiplicative product units with residual connections to explicitly model nonlinear feature interactions, showing improved accuracy, robustness, and interpretability over standard MLPs.

This paper presents CERO, a cross-epoch adaptive rollout optimization method for RL post-training of LLMs, which allocates a fixed rollout budget across prompts and epochs using Bayesian posterior variance to maximize sample efficiency, achieving theoretical regret bounds and outperforming GRPO on mathematical reasoning tasks.

CLaaS is a system for continual learning of LLM agents in deployment, using experience replay for sample-efficient online adaptation.

Deliberate Evolution (DE) is an agentic framework that improves LLM-based symbolic regression by decoupling candidate generation from search control, using adaptive operators, structural diagnosis tools, and reflective memory to achieve better results with only 40% of the standard sample budget.

Fair Reinforcement Learning introduces Democratic Alignment to incorporate multiple competing value sets from different agents, overcoming traditional RLHF limitations, and achieves orders of magnitude faster optimization via a black-box policy wrapper.

GPLD introduces a gradient-penalized latent dynamics regularizer for DreamerV3 to enforce local smoothness in transition learning, improving sample efficiency on continuous control tasks, especially complex locomotion.

MIT researchers propose Pedagogical RL, a new reinforcement learning method that uses a teacher model with privileged information and a spike-aware learnability reward to significantly improve sample efficiency and convergence speed over existing methods like GRPO and OPSD.

Proposes R2R2, a regularization method for self-predictive learning in reinforcement learning to mitigate overfitting under high update-to-data ratios, achieving significant improvements on continuous control tasks.

This paper argues that monolithic scaling of a single model is insufficient for achieving AGI and proposes Agentic AI with multi-agent collaboration as a necessary paradigm, demonstrating theoretically that agentic systems achieve exponentially superior generalization and sample efficiency.

FEST is a few-shot demonstration-guided reinforcement learning algorithm that achieves strong performance with minimal supervised fine-tuning data by combining supervised signals, on-policy learning, and weighted training to prevent overfitting.

This paper proposes a sample-efficient framework using the cross-entropy method to estimate extreme reliability ('five-nines') in LLMs, addressing the limitations of standard benchmarks in detecting rare failures.

This paper introduces Pinductor, a method that uses language model priors to efficiently learn POMDP world models from limited observation-action data, achieving performance comparable to methods with privileged hidden state access while surpassing traditional tabular approaches.

FeatCal is a calibration method that reduces performance gaps in post-merging models by layer-wise weight updates without gradient descent, achieving superior results on CLIP and GLUE benchmarks with high sample efficiency.

A fast-slow learning framework for LLMs combines fixed slow weights with optimized fast context weights, achieving up to 3x better sample efficiency and reduced catastrophic forgetting in continual learning scenarios.

This paper introduces the DR.Q algorithm, which improves model-based representations for Q-learning by maximizing mutual information and using faded prioritized experience replay to reduce bias and overfitting in continuous control tasks.

This paper analyzes the 'training in imagination' paradigm in model-based reinforcement learning, deriving optimal sample allocation strategies and characterizing how dynamics and reward model errors affect policy returns.

This paper introduces PNAPO, an offline preference optimization framework for rectified flow models that augments preference data with noise samples and uses dynamic regularization to improve training efficiency and sample efficiency.

This paper introduces ROPD, a rubric-based on-policy distillation framework that achieves superior sample efficiency compared to traditional logit-based methods. It enables model alignment in black-box scenarios by using structured semantic rubrics instead of teacher logits.

StraTA proposes strategic trajectory abstraction for long-horizon LLM agents, using hierarchical GRPO-style rollout with diverse strategy sampling and critical self-judgment to improve sample efficiency and final performance over frontier models and prior RL baselines.

FreshPER introduces a freshness-aware prioritized experience replay method for LLM/VLM reinforcement learning that addresses the 'priority staleness' problem by applying exponential age decay to stored priorities, enabling off-policy reuse of trajectories. Evaluated on eight agentic, reasoning, and math tasks, FreshPER significantly outperforms on-policy baselines with gains up to +367% on Sokoban.