Replay Failures as Successes: Sample-Efficient Reinforcement Learning for Instruction Following

Reinforcement Learning (RL) has shown promise for aligning Large Language Models (LLMs) to follow instructions with various constraints. Despite the encouraging results, RL improvement inevitably relies on sampling successful, high-quality responses; however, the initial model often struggles to generate responses that satisfy all constraints due to its limited capabilities, yielding sparse or indistinguishable rewards that impede learning. In this work, we propose Hindsight instruction Replay (HiR), a novel sample-efficient RL framework for complex instruction following tasks, which employs a select-then-rewrite strategy to replay failed attempts as successes based on the constraints that have been satisfied in hindsight. We perform RL on these replayed samples as well as the original ones, theoretically framing the objective as dual-preference learning at both the instruction- and response-level to enable efficient optimization using only a binary reward signal. Extensive experiments demonstrate that the proposed HiR yields promising results across different instruction following tasks, while requiring less computational budget. Our code and dataset is available at https://github.com/sastpg/HIR.

Dual-branch Spatial-Temporal Self-supervised Representation for Enhanced Road Network Learning

Road network representation learning (RNRL) has attracted increasing attention from both researchers and practitioners as various spatiotemporal tasks are emerging. Recent advanced methods leverage Graph Neural Networks (GNNs) and contrastive learning to characterize the spatial structure of road segments in a self-supervised paradigm. However, spatial heterogeneity and temporal dynamics of road networks raise severe challenges to the neighborhood smoothing mechanism of self-supervised GNNs. To address these issues, we propose a $\textbf{D}$ual-branch $\textbf{S}$patial-$\textbf{T}$emporal self-supervised representation framework for enhanced road representations, termed as DST. On one hand, DST designs a mix-hop transition matrix for graph convolution to incorporate dynamic relations of roads from trajectories. Besides, DST contrasts road representations of the vanilla road network against that of the hypergraph in a spatial self-supervised way. The hypergraph is newly built based on three types of hyperedges to capture long-range relations. On the other hand, DST performs next token prediction as the temporal self-supervised task on the sequences of traffic dynamics based on a causal Transformer, which is further regularized by differentiating traffic modes of weekdays from those of weekends. Extensive experiments against state-of-the-art methods verify the superiority of our proposed framework. Moreover, the comprehensive spatiotemporal modeling facilitates DST to excel in zero-shot learning scenarios.

CoVeR: Conformal Calibration for Versatile and Reliable Autoregressive Next-Token Prediction

Autoregressive pre-trained models combined with decoding methods have achieved impressive performance on complex reasoning tasks. While mainstream decoding strategies such as beam search can generate plausible candidate sets, they often lack provable coverage guarantees, and struggle to effectively balance search efficiency with the need for versatile trajectories, particularly those involving long-tail sequences that are essential in certain real-world applications. To address these limitations, we propose \textsc{CoVeR}, a novel model-free decoding strategy wihtin the conformal prediction framework that simultaneously maintains a compact search space and ensures high coverage probability over desirable trajectories. Theoretically, we establish a PAC-style generalization bound, guaranteeing that \textsc{CoVeR} asymptotically achieves a coverage rate of at least $1 - \alpha$ for any target level $\alpha \in (0,1)$.

Breaking the Exploration Bottleneck: Rubric-Scaffolded Reinforcement Learning for General LLM Reasoning

Recent advances in Large Language Models (LLMs) have underscored the potential of Reinforcement Learning (RL) to facilitate the emergence of reasoning capabilities. Despite the encouraging results, a fundamental dilemma persists as RL improvement relies on learning from high-quality samples, yet the exploration for such samples remains bounded by the inherent limitations of LLMs. This, in effect, creates an undesirable cycle in which what cannot be explored cannot be learned. In this work, we propose Rubric-Scaffolded Reinforcement Learning (RuscaRL), a novel instructional scaffolding framework designed to break the exploration bottleneck for general LLM reasoning. Specifically, RuscaRL introduces checklist-style rubrics as (1) explicit scaffolding for exploration during rollout generation, where different rubrics are provided as external guidance within task instructions to steer diverse high-quality responses. This guidance is gradually decayed over time, encouraging the model to internalize the underlying reasoning patterns; (2) verifiable rewards for exploitation during model training, where we can obtain robust LLM-as-a-Judge scores using rubrics as references, enabling effective RL on general reasoning tasks. Extensive experiments demonstrate the superiority of the proposed RuscaRL across various benchmarks, effectively expanding reasoning boundaries under the best-of-N evaluation. Notably, RuscaRL significantly boosts Qwen-2.5-7B-Instruct from 23.6 to 50.3 on HealthBench-500, surpassing GPT-4.1. Furthermore, our fine-tuned variant on Qwen3-30B-A3B-Instruct achieves 61.1 on HealthBench-500, outperforming leading LLMs including OpenAI-o3.

Consistent Paths Lead to Truth: Self-Rewarding Reinforcement Learning for LLM Reasoning

Recent advances of Reinforcement Learning (RL) have highlighted its potential in complex reasoning tasks, yet effective training often relies on external supervision, which limits the broader applicability. In this work, we propose a novel self-rewarding reinforcement learning framework to enhance Large Language Model (LLM) reasoning by leveraging the consistency of intermediate reasoning states across different reasoning trajectories. Our key insight is that correct responses often exhibit consistent trajectory patterns in terms of model likelihood: their intermediate reasoning states tend to converge toward their own final answers (high consistency) with minimal deviation toward other candidates (low volatility). Inspired by this observation, we introduce CoVo, an intrinsic reward mechanism that integrates Consistency and Volatility via a robust vector-space aggregation strategy, complemented by a curiosity bonus to promote diverse exploration. CoVo enables LLMs to perform RL in a self-rewarding manner, offering a scalable pathway for learning to reason without external supervision. Extensive experiments on diverse reasoning benchmarks show that CoVo achieves performance comparable to or even surpassing supervised RL. Our code is available at https://github.com/sastpg/CoVo.

Intra-Trajectory Consistency for Reward Modeling

Reward models are critical for improving large language models (LLMs), particularly in reinforcement learning from human feedback (RLHF) or inference-time verification. Current reward modeling typically relies on scores of overall responses to learn the outcome rewards for the responses. However, since the response-level scores are coarse-grained supervision signals, the reward model struggles to identify the specific components within a response trajectory that truly correlate with the scores, leading to poor generalization on unseen responses. In this paper, we propose to leverage generation probabilities to establish reward consistency between processes in the response trajectory, which allows the response-level supervisory signal to propagate across processes, thereby providing additional fine-grained signals for reward learning. Building on analysis under the Bayesian framework, we develop an intra-trajectory consistency regularization to enforce that adjacent processes with higher next-token generation probability maintain more consistent rewards. We apply the proposed regularization to the advanced outcome reward model, improving its performance on RewardBench. Besides, we show that the reward model trained with the proposed regularization induces better DPO-aligned policies and achieves better best-of-N (BON) inference-time verification results. Our code is provided in https://github.com/chaoyang101/ICRM.

BiTrajDiff: Bidirectional Trajectory Generation with Diffusion Models for Offline Reinforcement Learning

Recent advances in offline Reinforcement Learning (RL) have proven that effective policy learning can benefit from imposing conservative constraints on pre-collected datasets. However, such static datasets often exhibit distribution bias, resulting in limited generalizability. To address this limitation, a straightforward solution is data augmentation (DA), which leverages generative models to enrich data distribution. Despite the promising results, current DA techniques focus solely on reconstructing future trajectories from given states, while ignoring the exploration of history transitions that reach them. This single-direction paradigm inevitably hinders the discovery of diverse behavior patterns, especially those leading to critical states that may have yielded high-reward outcomes. In this work, we introduce Bidirectional Trajectory Diffusion (BiTrajDiff), a novel DA framework for offline RL that models both future and history trajectories from any intermediate states. Specifically, we decompose the trajectory generation task into two independent yet complementary diffusion processes: one generating forward trajectories to predict future dynamics, and the other generating backward trajectories to trace essential history transitions.BiTrajDiff can efficiently leverage critical states as anchors to expand into potentially valuable yet underexplored regions of the state space, thereby facilitating dataset diversity. Extensive experiments on the D4RL benchmark suite demonstrate that BiTrajDiff achieves superior performance compared to other advanced DA methods across various offline RL backbones.

SeRL: Self-Play Reinforcement Learning for Large Language Models with Limited Data

Recent advances have demonstrated the effectiveness of Reinforcement Learning (RL) in improving the reasoning capabilities of Large Language Models (LLMs). However, existing works inevitably rely on high-quality instructions and verifiable rewards for effective training, both of which are often difficult to obtain in specialized domains. In this paper, we propose Self-play Reinforcement Learning(SeRL) to bootstrap LLM training with limited initial data. Specifically, SeRL comprises two complementary modules: self-instruction and self-rewarding. The former module generates additional instructions based on the available data at each training step, employing robust online filtering strategies to ensure instruction quality, diversity, and difficulty. The latter module introduces a simple yet effective majority-voting mechanism to estimate response rewards for additional instructions, eliminating the need for external annotations. Finally, SeRL performs conventional RL based on the generated data, facilitating iterative self-play learning. Extensive experiments on various reasoning benchmarks and across different LLM backbones demonstrate that the proposed SeRL yields results superior to its counterparts and achieves performance on par with those obtained by high-quality data with verifiable rewards. Our code is available at https://github.com/wantbook-book/SeRL.

VORTA: Efficient Video Diffusion via Routing Sparse Attention

Video Diffusion Transformers (VDiTs) have achieved remarkable progress in high-quality video generation, but remain computationally expensive due to the quadratic complexity of attention over high-dimensional video sequences. Recent attention acceleration methods leverage the sparsity of attention patterns to improve efficiency; however, they often overlook inefficiencies of redundant long-range interactions. To address this problem, we propose \textbf{VORTA}, an acceleration framework with two novel components: 1) a sparse attention mechanism that efficiently captures long-range dependencies, and 2) a routing strategy that adaptively replaces full 3D attention with specialized sparse attention variants throughout the sampling process. It achieves a $1.76\times$ end-to-end speedup without quality loss on VBench. Furthermore, VORTA can seamlessly integrate with various other acceleration methods, such as caching and step distillation, reaching up to $14.41\times$ speedup with negligible performance degradation. VORTA demonstrates its efficiency and enhances the practicality of VDiTs in real-world settings.

Bi-level Mean Field: Dynamic Grouping for Large-Scale MARL

Large-scale Multi-Agent Reinforcement Learning (MARL) often suffers from the curse of dimensionality, as the exponential growth in agent interactions significantly increases computational complexity and impedes learning efficiency. To mitigate this, existing efforts that rely on Mean Field (MF) simplify the interaction landscape by approximating neighboring agents as a single mean agent, thus reducing overall complexity to pairwise interactions. However, these MF methods inevitably fail to account for individual differences, leading to aggregation noise caused by inaccurate iterative updates during MF learning. In this paper, we propose a Bi-level Mean Field (BMF) method to capture agent diversity with dynamic grouping in large-scale MARL, which can alleviate aggregation noise via bi-level interaction. Specifically, BMF introduces a dynamic group assignment module, which employs a Variational AutoEncoder (VAE) to learn the representations of agents, facilitating their dynamic grouping over time. Furthermore, we propose a bi-level interaction module to model both inter- and intra-group interactions for effective neighboring aggregation. Experiments across various tasks demonstrate that the proposed BMF yields results superior to the state-of-the-art methods. Our code will be made publicly available.