Group Relative Policy Optimization

Despite the success of foundation models in language and vision, molecular graph generation still lacks a unified framework for heterogeneous design tasks with reliable controllability. While reinforcement learning (RL) offers a natural post-training mechanism for task-specific optimization, applying it to graph generative models is hindered by the vast atom-wise action spaces and chemically invalid intermediate states. We propose \textbf{Co}ntrollable \textbf{Mole}cular Generative Foundation Models (CoMole), built with a unified motif-aware graph diffusion pipeline. By learning a motif-aware graph space, CoMole transfers pretrained structural priors into controllable generation, where RL optimizes conditional reverse policies over chemically meaningful decisions. We theoretically characterize the bottleneck of atom-level RL and justify motif-aware policy optimization. Across three heterogeneous benchmarks spanning materials and drug discovery, CoMole ranks first in controllability on all nine targets, reduces MAE by up to 48.2% relative to the strongest baselines, and maintains validity above 0.94 without rule-based correction or post-hoc filtering. We further show that CoMole transfers controllability to unseen properties by optimizing only task embeddings with the generator frozen, achieving performance competitive with strong task-specific baselines.

Visual reasoning, often interleaved with intermediate visual states, has emerged as a promising direction in the field. A straightforward approach is to directly generate images via unified models during reasoning, but this is computationally expensive and architecturally non-trivial. Recent alternatives include agentic reasoning through code or tool calls, and latent reasoning with learnable hidden embeddings. However, agentic methods incur context-switching latency from external execution, while latent methods lack task generalization and are difficult to train with autoregressive parallelization. To combine their strengths while mitigating their limitations, we propose ATLAS, a framework in which a single discrete 'word', termed as a functional token, serves both as an agentic operation and a latent visual reasoning unit. Each functional token is associated with an internalized visual operation, yet requires no visual supervision and remains a standard token in the tokenizer vocabulary, which can be generated via next-token prediction. This design avoids verbose intermediate visual content generation, while preserving compatibility with the vanilla scalable SFT and RL training, without architectural or methodological modifications. To further address the sparsity of functional tokens during RL, we introduce Latent-Anchored GRPO (LA-GRPO), which stabilizes the training by anchoring functional tokens with a statically weighted auxiliary objective, providing stronger gradient updates. Extensive experiments and analyses demonstrate that ATLAS achieves superior performance on challenging benchmarks while maintaining clear interpretability. We hope ATLAS offers a new paradigm inspiring future visual reasoning research.

Causal autoregressive video diffusion models support real-time streaming generation by extrapolating future chunks from previously generated content. Distilling such generators from high-fidelity bidirectional teachers yields competitive few-step models, yet a persistent gap between the history distributions encountered during training and those arising at inference constrains generation quality over long horizons. We introduce the Real-time Autoregressive Video Extrapolation Network (RAVEN), a training-time test framework that repacks each self rollout into an interleaved sequence of clean historical endpoints and noisy denoising states. This formulation aligns training attention with inference-time extrapolation and allows downstream chunk losses to supervise the history representations on which future predictions depend. We further propose Consistency-model Group Relative Policy Optimization (CM-GRPO), which reformulates a consistency sampling step as a conditional Gaussian transition and applies online Reinforcement Learning (RL) directly to this kernel, avoiding the Euler-Maruyama auxiliary process adopted in prior flow-model RL formulations. Experiments demonstrate that RAVEN surpasses recent causal video distillation baselines across quality, semantic, and dynamic degree evaluations, and that CM-GRPO provides further gains when combined with RAVEN.

Interactive real-time autoregressive video generation is essential for applications such as content creation and world modeling, where visual content must adapt to dynamically evolving event conditions. A fundamental challenge lies in balancing reactivity and stability: models must respond promptly to new events while maintaining temporal coherence over long horizons. Existing approaches distill bidirectional models into autoregressive generators and further adapt them via streaming long tuning, yet often exhibit persistent drift after condition changes. We identify the cause as conditional bias, where the teacher may provide condition-aligned but trajectory-agnostic guidance, biasing generation toward locally valid yet globally inconsistent modes. Inspired by Trust Region Policy Optimization, we propose Delta Forcing, a simple yet effective framework that constrains unreliable teacher supervision within an adaptive trust region. Specifically, Delta Forcing estimates transition consistency from the latent delta between teacher and generator trajectories, and uses it to balance teacher supervision with a monotonic continuity objective. This suppress unreliable teacher-induced shifts while preserving responsiveness to new events. Extensive experiments demonstrate that Delta Forcing significantly improves consistency while maintaining event reactivity.

Extending large language models (LLMs) to low-resource languages often incurs an "alignment tax": improvements in the target language come at the cost of catastrophic forgetting in general capabilities. We argue that this trade-off arises from the rigidity of supervised fine-tuning (SFT), which enforces token-level surface imitation on narrow and biased data distributions. To address this limitation, we propose a semantic-space alignment paradigm powered by Group Relative Policy Optimization (GRPO), where the model is optimized using embedding-level semantic rewards rather than likelihood maximization. This objective encourages meaning preservation through flexible realizations, enabling controlled updates that reduce destructive interference with pretrained knowledge. We evaluate our approach on Tibetan-Chinese machine translation and Tibetan headline generation. Experiments show that our method acquires low-resource capabilities while markedly mitigating alignment tax, preserving general competence more effectively than SFT. Despite producing less rigid surface overlap, semantic RL yields higher semantic quality and preference in open-ended generation, and few-shot transfer results indicate that it learns more transferable and robust representations under limited supervision. Overall, our study demonstrates that reinforcement learning with semantic rewards provides a safer and more reliable pathway for inclusive low-resource language expansion.

Aligning streaming autoregressive (AR) video generators with human preferences is challenging. Existing reinforcement learning methods predominantly rely on noise-based exploration and SDE-based surrogate policies that are mismatched to the deterministic ODE dynamics of distilled AR models, and tend to perturb low-level appearance rather than the high-level semantic storyline progression critical for long-horizon coherence. To address these limitations, we present KVPO, an ODE-native online Group Relative Policy Optimization (GRPO) framework for aligning streaming video generators. For diversity exploration, KVPO introduces a causal-semantic exploration paradigm that relocates the source of variation from stochastic noise to the historical KV cache. By stochastically routing historical KV entries, it constructs semantically diverse generation branches that remain strictly on the data manifold. For policy modeling, KVPO introduces a velocity-field surrogate policy based on Trajectory Velocity Energy (TVE), which quantifies branch likelihood in flow-matching velocity space and yields a reward-weighted contrastive objective fully consistent with the native ODE formulation. Experiments on multiple distilled AR video generators demonstrate consistent gains in visual quality, motion quality, and text-video alignment across both single-prompt short-video and multi-prompt long-video settings.

Contrastive reinforcement learning (CRL) learns goal-conditioned Q-values through a contrastive objective over state-action and goal representations, removing the need for hand-crafted reward functions. Despite impressive success in achieving viable self-supervised learning in RL, all existing CRL algorithms rely on off-policy optimisation and are mostly constrained to continuous action spaces, with little research invested in discrete environments. This leaves CRL disconnected from widely used and effective, modern on-policy training pipelines adopted across both single-agent and multi-agent RL in continuous and discrete environments. To establish a first connection, we introduce Contrastive Proximal Policy Optimisation (CPPO). CPPO is an on-policy contrastive RL algorithm that derives policy advantages directly from contrastive Q-values and optimises them via the standard PPO objective, without requiring a reward function or a replay buffer. We evaluate CPPO across continuous and discrete, single-agent and cooperative multi-agent tasks. Whilst the existence of an on-policy approach is inherently useful, we observe that \textbf{CPPO not only significantly outperforms the previous CRL baselines in 14 out of 18 tasks, but also matches or exceeds PPO's performance, which uses hand-crafted dense rewards, in 12 out of the 18 tasks tested.}

There is growing interest in utilizing flow-based models as decision-making policies in reinforcement learning due to their high expressive capacity. However, effectively leveraging this expressivity for value maximization remains challenging, as naive gradient-based optimization requires backpropagating through numerical solvers and often leads to instability. Existing approaches typically address this issue by restricting the expressive capacity of flow-based policies, resulting in a trade-off between optimization stability and representational flexibility. To resolve this, we introduce Q-Flow, a framework that leverages the deterministic nature of flow dynamics to explicitly propagate terminal trajectory value to intermediate latent states along the policy-induced flow. This formulation enables stable policy optimization using intermediate value gradients without unrolling the numerical solver, effectively bridging the gap between stability and expressivity. We evaluate Q-Flow in the offline learning setting on the challenging OGBench suite, where it consistently outperforms state-of-the-art baselines by an average of 10.6 percentage points, while also enabling stable online adaptation within the same framework.

In Vision-Language Models (VLMs), processing a massive number of visual tokens incurs prohibitive computational overhead. While recent training-aware pruning methods attempt to selectively discard redundant tokens, they largely rely on continuous-gradient relaxations. However, visual token pruning is inherently a discrete, non-convex combinatorial problem; consequently, these continuous approximations frequently trap the optimization in sub-optimal local minima, especially under aggressive compression budgets. To overcome this fundamental bottleneck, we propose GRIP-VLM, a Group-Relative Importance Pruning framework driven by Reinforcement Learning. Rather than relying on smooth-gradient assumptions, GRIP-VLM formulates pruning as a Markov Decision Process, employing a Group Relative Policy Optimization (GRPO) paradigm anchored by supervised warm-up to directly explore the discrete selection space. Integrated with a budget-aware scorer, our lightweight agent dynamically evaluates per-token importance and adapts to arbitrary compression ratios without retraining. Extensive experiments across diverse multimodal benchmarks demonstrate that GRIP-VLM consistently outperforms heuristic and supervised-learning baselines, achieving a superior Pareto frontier and delivering up to a 15% inference speedup at equal accuracy.

On-policy distillation (OPD) has become a promising paradigm for reasoning-oriented post-training of large language models (LLMs), especially when combined with reinforcement learning from verifiable rewards (RLVR). Existing OPD methods rely on reverse KL (RKL)-based teacher supervision over trajectories sampled from the student policy. However, we identify a critical limitation: under large teacher--student policy divergence, RL-driven exploration often produces trajectories outside the teacher distribution, resulting in uninformative negative feedback. To address this, we propose Teacher-Guided Policy Optimization (TGPO), an on-policy reasoning distillation method that remains effective under large policy divergence settings. Rather than relying solely on evaluative supervision, TGPO uses teacher to directly guide token level generation conditioning on student-generated contexts; together with RLVR-style trajectory level rewards, TGPO steers exploration toward improved continuations. Experiments on reasoning benchmarks show that TGPO consistently outperforms existing RKL-based OPD methods and remains robust across different teacher models.

Reinforcement learning has become a powerful paradigm for post-training large language model agents, yet credit assignment in multi-turn environments remains a challenge. Agents often receive sparse, trajectory-level rewards only at the end of an episode, making it difficult to determine which intermediate actions contributed to success or failure. As a result, propagating delayed outcomes back to individual decision steps without relying on costly auxiliary value models remains an open problem. We propose Generalized Advantage Grouped Policy Optimization (GAGPO), a critic-free reinforcement learning method for precise, step-aligned temporal credit assignment. GAGPO constructs a non-parametric grouped value proxy from sampled rollouts and uses it to compute TD/GAE-style temporal advantages, recursively propagating outcome supervision backward through time. Combined with group-wise advantage normalization and an action-level importance ratio, GAGPO extracts stable, localized optimization signals directly from multi-turn trajectories. Experiments on ALFWorld and WebShop show that GAGPO outperforms strong reinforcement learning baselines. Further analyses demonstrate faster early-stage learning, improved interaction efficiency, and smoother optimization dynamics, suggesting that GAGPO offers a simple yet effective framework for multi-turn agentic reinforcement learning.

Multi-Agent Proximal Policy Optimization (MAPPO) is a variant of the Proximal Policy Optimization (PPO) algorithm, specifically tailored for multi-agent reinforcement learning (MARL). MAPPO optimizes cooperative multi-agent settings by employing a centralized critic with decentralized actors. However, in case of multi-dimensional environment, MAPPO can not extract optimal policy due to non-stationary agent observation. To overcome this problem, we introduce a novel approach, Entropy Regularization-based Proximal Policy Optimization (ERPPO). For the policy optimization, we first define the object detection ambiguity under multi-dimensional observation environment. Distributional Spatiotemporal Ambiguity (DSA) learner is trained to estimate object detection uncertainty in non-stationary constraints. Then, we enhance PPO with a novel Entropy Regularization term. This regularization dynamically adjusts the policy update by applying a stronger (L1) regularization in high-ambiguity observation to encourage significant exploratory actions and a weaker (L2) regularization in low-ambiguity observation to stabilize the proximal policy optimization. This approach is designed to enhance the probability of successful object localization in time-critical operations by reducing detection failures and optimizing search policy. Experiments on a testbed with AirSim-based maritime searching scenarios show that the proposed ERPPO improves accuracy performance. Our proposed method improves higher gradient than MAPPO. Qualitative results confirm that ERPPO effectiveness in terms of suppressing false detection in visually uncertain conditions.

Backtesting large language models on historical events requires reasoning exclusively from information available before a specified cutoff date. Yet models routinely leak post-cutoff knowledge from pre-training into their reasoning, inflating apparent accuracy and undermining evaluation validity. Prompt-based constraints fail when suppressed content is causally related to the prediction, and knowledge unlearning cannot address this problem because temporal compliance is instance-specific: the same fact may be legitimate evidence for one cutoff date and a violation for another. Rather than erasing knowledge, the model must learn temporal discipline: selecting evidence conditioned on each instance's cutoff date. We propose TEMPO (Temporal Enforcement via Mode-separated Policy Optimization), which trains this discipline via two contributions: (1) a two-mode reward where a leakage mode drives post-cutoff claims to zero as a hard prerequisite before a performance mode optimizes task performance; and (2) a GRPO-based training pipeline that enables the model to discover temporally valid reasoning strategies. We prove that training monotonically decreases leakage, converges to the leak-free optimum, and improves task performance once compliance is achieved. On three prediction tasks and two models, TEMPO reduces leakage from 213% to 0.63.7% across all conditions, with task performance improving 6~13% where strong pre-cutoff signals exist and maintained where the prediction task is inherently difficult from valid information alone.

Traditional retrieval pipelines optimize utility through stages of candidate retrieval and reranking, where ranking operates over a predefined candidate set. Large Language Models (LLMs) broaden this into a generative process: given a candidate pool, an LLM can generate a subset and order it within a single autoregressive pass. However, this flexibility introduces a new optimization challenge: the model must search a combinatorial output space while receiving utility feedback only after the full ranked list is generated. Because this feedback is defined over the completed sequence, it cannot distinguish whether a poor result arises from failing to generate a relevant subset or from failing to rank that subset correctly. This credit assignment gap makes end-to-end optimization unstable and sample-inefficient. Existing systems often address this by separating candidate generation from ranking. However, such decoupling remains misaligned with downstream utility because ranking is limited by the candidate set it receives. To bridge this gap, we propose a unified framework that performs both within a single autoregressive rollout and optimizes them end-to-end via factorized group-relative policy optimization (F-GRPO). Our framework factorizes the policy into candidate generation and ranking while sharing a single LLM backbone, and jointly trains them with an order-invariant coverage reward and a position-aware utility reward. To address the resulting phase-specific credit assignment problem, we use separate group-relative advantages for generation and ranking within a two-phase sequence-level objective. Across sequential recommendation and multi-hop question answering benchmarks, F-GRPO improves top-ranked performance over GRPO and decoupled baselines, outperforms supervised alternatives, and remains competitive with strong zero-shot rerankers, with no architectural changes at inference time.

Group Relative Policy Optimization (GRPO) is one of the most widely adopted RLVR algorithms for post-training large language models on reasoning tasks. We first show that GRPO admits an equivalent discriminative reformulation, in which policy optimization maximizes the expected score gap between verified positive and negative rollouts. This reformulation reveals two objective-level limitations: likelihood-misaligned surrogate scores, in which clipped ratio-based scores are optimized rather than the sequence likelihoods that govern generation, and score-insensitive credit assignment, in which rollout-level credit does not reflect the current score gaps between positive and negative rollouts. To address these limitations, we propose ConSPO, a Contrastive Sequence-level Policy Optimization method that uses length-normalized sequence log-probabilities as rollout scores and contrasts verified positive rollouts against negative distractors within the same group. ConSPO optimizes a group-wise InfoNCE-style objective to adaptively strengthen updates for poorly separated positives and high-scoring negatives, together with a curriculum-scheduled margin that preserves separation pressure as training progresses. Experiments across diverse settings show that ConSPO outperforms strong baselines on challenging reasoning benchmarks. Code will be released upon paper acceptance.

Reinforcement learning (RL) for large language models (LLMs) is dominated by the cost of rollout generation, which has motivated the use of low-precision rollouts (e.g., FP8) paired with a BF16 trainer to improve throughput and reduce memory pressure. This introduces a rollout-training mismatch that biases the policy gradient and can cause training to collapse outright on reasoning benchmarks. We show that the mismatch is non-stationary and acts as a double-edged sword: early in training it provides a stochastic exploration bonus, exposing the gradient to trajectories the trainer would otherwise under-sample, but the same perturbation transitions into a destabilizing source of bias as the policy concentrates. To solve this, we propose Adaptive Importance Sampling (AIS), a correction framework that adjusts the strength of its intervention on a per-batch basis. AIS combines three real-time diagnostics, namely weight reliability, divergence severity, and variance amplification, into a single mixing coefficient that interpolates between the uncorrected and fully importance-weighted gradients, suppressing the destabilizing component of the mismatch while preserving its exploratory benefit. We integrate AIS into GRPO and evaluate it on the diffusion-based LLaDA-8B-Instruct and the autoregressive Qwen3-8B and Qwen3.5-9B across mathematical reasoning and planning benchmarks. AIS matches the BF16 baseline on most tasks while retaining the 1.5 to 2.76x rollout speedup of FP8.

Multi-objective reinforcement learning in robotic domains requires balancing complex, non-convex trade-offs between conflicting objectives. While linear scalarization methods provide stability, they are theoretically incapable of recovering solutions within non-convex regions of the Pareto front. Conversely, static non-linear scalarizations (e.g., Tchebycheff) can theoretically access these regions but often suffer from severe gradient variance and optimization instability in deep RL. In this work, we propose an Adaptive Smooth Tchebycheff framework that resolves this tension by dynamically modulating the curvature of the optimization landscape. We introduce a novel conflict-driven controller that regulates the optimization smoothness based on real-time gradient interference. This allows the agent to anneal toward precise, non-convex scalarization when objectives align, while elastically reverting to stable, smooth approximations when destructive gradient conflicts emerge. We validate our approach on a challenging robotic stealth visual search task -- a proxy for monitoring of protected/fragile ecosystems -- where an agent must balance search, exposure/interference minimization and exploration speed. Extensive ablations confirm that our conflict-aware adaptation enables the robust discovery of Pareto-optimal policies in non-convex regions inaccessible to linear baselines and unstable for static non-linear methods. Website: https://alejandromllo.github.io/research/pasta/

The alignment of Large Language Models (LLMs) utilizes Reinforcement Learning from AI Feedback (RLAIF) for non-verifiable domains such as long-form question answering and open-ended instruction following. These domains often rely on LLM based auto-raters to provide granular, multi-tier discrete rewards (e.g., 1-10 rubrics) that are inherently stochastic due to prompt sensitivity and sampling randomness. We empirically verify the stochasticity of auto-raters that can propagate and corrupt standard advantage estimators like GRPO and MaxRL, as a noisy reward samples can skew normalization statistics and degrade the global learning signal. Empirically, sampling more rewards and taking majority voting may reduce the noise and improve performance, but this approach is computationally expensive. To address this bottleneck, we introduce $\textbf{O}$rdinal $\textbf{D}$ecomposition for $\textbf{R}$obust $\textbf{P}$olicy $\textbf{O}$ptimization ($\textbf{ODRPO}$), a framework that structurally isolates evaluation noise by decomposing discrete rewards into a sequence of ordinal binary indicators. By independently computing and accumulating advantages across these progressively challenging success thresholds, ODRPO prevents outlier evaluations from corrupting the global update while establishing an implicit, variance-aware learning curriculum. Empirically, ODRPO achieves robust performance on Qwen2.5-7B and Qwen3-4B models, outperforming baselines with relative improvements of upto 14.8% on FACTS-grounding-v2 and 7.5% on Alpaca-Evals. Critically, these gains are achieved with negligible training-time overhead, as ODRPO requires no additional compute per step compared to standard estimators. Supported by theoretical analysis confirming its optimization stability, ODRPO provides a scalable and robust framework for aligning models within the noisy, discrete evaluation landscape of modern RLAIF.

Continuous-action policies trained on a single demonstrated trajectory per scene suffer from mode collapse: samples cluster around the demonstrated maneuver and the policy cannot represent semantically distinct alternatives. Under preference-based evaluation, this caps best-of-N performance -- even oracle selection cannot recover what the sampling distribution does not contain. We introduce DIAL, a two-stage Driving-Intent-Amplified reinforcement Learning framework for preference-aligned continuous-action driving policies. In the first stage, DIAL conditions the flow-matching action head on a discrete intent label with classifier-free guidance (CFG), which expands the sampling distribution along distinct maneuver modes and breaks single-demonstration mode collapse. In the second stage, DIAL carries this expanded distribution into preference RL through multi-intent GRPO, which spans all intent classes within every preference group and prevents fine-tuning from re-collapsing around the currently preferred mode. Instantiated for end-to-end driving with eight rule-derived intents and evaluated on WOD-E2E: competitive Vision-to-Action (VA) and Vision-Language-Action (VLA) Supervised Finetuning (SFT) baselines plateau below the human-driven demonstration at best-of-128, with the strongest prior (RAP) capping at Rater Feedback Score (RFS) 8.5 even with best-of-64; intent-CFG sampling lifts this ceiling to RFS 9.14 at best-of-128, surpassing both the prior best (RAP 8.5) and the human-driven demonstration (8.13) for the first time; and multi-intent GRPO improves held-out RFS from 7.681 to 8.211, while every single-intent baseline peaks lower and degrades by training end. These results suggest that the bottleneck of preference RL on continuous-action policies trained from demonstrations is not only how to update the policy, but to expand and preserve the sampling distribution being optimized.

In this paper, we propose AlphaGRPO, a novel framework that applies Group Relative Policy Optimization (GRPO) to AR-Diffusion Unified Multimodal Models (UMMs) to enhance multimodal generation capabilities without an additional cold-start stage. Our approach unlocks the model's intrinsic potential to perform advanced reasoning tasks: Reasoning Text-to-Image Generation, where the model actively infers implicit user intents, and Self-Reflective Refinement, where it autonomously diagnoses and corrects misalignments in generated outputs. To address the challenge of providing stable supervision for real-world multimodal generation, we introduce the Decompositional Verifiable Reward (DVReward). Unlike holistic scalar rewards, DVReward utilizes an LLM to decompose complex user requests into atomic, verifiable semantic and quality questions, which are then evaluated by a general MLLM to provide reliable and interpretable feedback. Extensive experiments demonstrate that AlphaGRPO yields robust improvements across multimodal generation benchmarks, including GenEval, TIIF-Bench, DPG-Bench and WISE, while also achieving significant gains in editing tasks on GEdit without training on editing tasks. These results validate that our self-reflective reinforcement approach effectively leverages inherent understanding to guide high-fidelity generation. Project page: https://huangrh99.github.io/AlphaGRPO/

In settings where labeled verifiable training data is the binding constraint, each checked example should be allocated to the model and reward density where it is most informative. We identify a reward-density principle that governs this allocation: sparse sequence-level reward is most useful on models that can explore and discover better behavior, while dense token-level teacher supervision is better suited for compressing that behavior into a smaller deployment model. The principle yields a simple allocation rule: use scarce labeled data upstream on the strongest available teacher, then transfer the reward-shaped behavior downstream as dense supervision. We evaluate this rule through a four-stage workflow -- teacher RL, forward-KL warmup, on-policy distillation, optional post-bridge student RL -- on verifiable math with Qwen3 and Llama models. At fixed Qwen3-1.7B deployment-student size, an RL-improved 8B teacher distilled through the dense bridge outperforms direct GRPO on the same student ($79.3\%$ vs.\ $75.9\%$ on MATH; $25.2\%$ vs.\ $19.8\%$ on AIME~2024, avg@16), while transfer from the same teacher \emph{before} RL underperforms. A component ablation confirms that each stage is load-bearing: replacing the RL-improved teacher with a raw teacher costs $7.8$ MATH points, removing the forward-KL warmup costs $1.7$, and removing on-policy distillation costs $3.3$. The teacher-quality ordering -- raw-teacher transfer $<$ direct GRPO $<$ RL-teacher transfer -- replicates on Llama-3.1-8B-Instruct with a Llama-3.3-70B-Instruct teacher. The operational lesson is to avoid spending scarce labeled data on the least prepared policy: use sparse reward for teacher-side discovery, dense transfer for student compression, and student-side sparse reward only after the bridge.

We study the \textit{parameter placement problem}: given a fixed budget of $k$ trainable entries within the B matrix of a LoRA adapter (A frozen), does the choice of which $k$ matter? Under supervised fine-tuning, random and informed subsets achieve comparable performance. Under GRPO on base models, random placement fails to improve over the base model, while gradient-informed placement recovers standard LoRA accuracy. This regime dependence traces to gradient structure: SFT gradients are low-rank and directionally stable, so any subset accumulates coherent updates; GRPO gradients are high-rank and near-orthogonal across steps, so only elements with consistently signed gradients retain the learning signal. Our scoring procedure identifies these critical parameters in under 10 seconds at less than 0.5% of training cost. Selected parameters concentrate on residual-stream-writing projections (V, O, Down), stable across model families and scales (1.5B - 8B).

Group Relative Policy Optimisation (GRPO) enhances large language models by estimating advantages across a group of sampled trajectories. However, mapping these trajectory-level advantages to policy updates requires aggregating token-level probabilities within each sequence. Relying on a fixed aggregation mechanism for this step fundamentally limits the algorithm's adaptability. Empirically, we observe a critical trade-off: certain fixed aggregations frequently suffer from training collapse, while others fail to yield satisfactory performance. To resolve this, we propose \textbf{HölderPO}, a generalised policy optimisation framework unifying token-level probability aggregation via the Hölder mean. By explicitly modulating the parameter $p$, our framework provides continuous control over the trade-off between gradient concentration and variance bounds. Theoretically, we prove that a larger $p$ concentrates the gradient to amplify sparse learning signals, whereas a smaller $p$ strictly bounds gradient variance. Because no static configuration can universally resolve this concentration-stability trade-off, we instantiate the framework with a dynamic annealing algorithm that progressively schedules $p$ across the training lifecycle. Extensive evaluations demonstrate superior stability and convergence over existing baselines. Specifically, our approach achieves a state-of-the-art average accuracy of $54.9\%$ across multiple mathematical benchmarks, yielding a substantial $7.2\%$ relative gain over standard GRPO and secures an exceptional $93.8\%$ success rate on ALFWorld.

Large Language Models (LLMs) suffer from order bias, where their performance is affected by the arrangement order of input elements. This unfairness limits the model's applications in scenarios such as in-context learning and Retrieval-Augmented Generation (RAG). Recent studies attempt to obtain optimal or suboptimal arrangements based on statistical results or using dataset-based search, but these methods increase inference overhead while leaving the model's inherent order bias unresolved. Other studies mitigate order sensitivity through supervised fine-tuning using augmented training sets with multiple order variants, but often at the cost of accuracy, trapping the model in consistent yet incorrect hallucinations. In this paper, we propose \textbf{D}ual \textbf{G}roup \textbf{A}dvantage \textbf{O}ptimization (\textbf{DGAO}), which aims to improve model accuracy and order stability simultaneously. DGAO calculates and balances intra-group relative accuracy advantage and inter-group relative stability advantage, rewarding the policy model for generating order-stable and correct outputs while penalizing order-sensitive or incorrect responses. This marks the first time reinforcement learning has been used to mitigate LLMs' order sensitivity. We also propose two new metrics, Consistency Rate and Overconfidence Rate, to reveal the pseudo-stability of previous methods and guide more comprehensive evaluation. Extensive experiments demonstrate that DGAO achieves superior order fairness while improving performance on RAG, mathematical reasoning, and classification tasks. Our code is available at: https://github.com/Hyalinesky/DGAO.

Softmax policy gradient converges at $O(1/t)$, but its transient behavior near sub-optimal corners of the simplex can be exponentially slow. The bottleneck is self-trapping: negative-advantage actions reinforce the corner policy and can initially push the optimal action backward. We study \emph{Delightful Policy Gradient} (DG), which gates each policy-gradient term by the product of advantage and action surprisal. For $K$-armed bandits, we prove that the zero-temperature limit of DG removes this corner-trapping mechanism on a quantitative sector near any sub-optimal corner, yielding a first-exit escape bound logarithmic in the initial probability ratio. At every fixed temperature, the same local mechanism persists because harmful actions are polynomially suppressed as they become rare. A key structural insight is that every action better than the corner action is an \emph{ally}: its contribution to escape is non-negative. Combining corner instability with a monotonic value improvement identity, we prove that DG converges globally to the optimal policy in both bandits and tabular MDPs at an asymptotic $O(1/t)$ rate. We also show, via an exact counterexample, that this tabular mechanism can fail under shared function approximation. In MNIST contextual bandits with a shared-parameter neural network, DG nevertheless recovers from bad initializations faster than standard policy gradient, suggesting that the counterexample marks a boundary of the theory rather than a practical prohibition.

Reinforcement learning has become a widely used post-training approach for LLM agents, where training commonly relies on outcome-level rewards that provide only coarse supervision. While finer-grained credit assignment is promising for effective policy updates, obtaining reliable local credit and assigning it to the right parts of the long-horizon trajectory remains an open challenge. In this paper, we propose Granularity-adaptivE Advantage Reweighting (GEAR), an adaptive-granularity credit assignment framework that reshapes the trajectory-level GRPO advantage using token- and segment-level signals derived from self-distillation. GEAR compares an on-policy student with a ground-truth-conditioned teacher to obtain a reference-guided divergence signal for identifying adaptive segment boundaries and modulating local advantage weights. This divergence often spikes at the onset of a semantic deviation, while later tokens in the same autoregressive continuation may return to low divergence. GEAR therefore treats such spikes as anchors for adaptive credit regions: where the student remains aligned with the teacher, token-level resolution is preserved; where it departs, GEAR groups the corresponding continuation into an adaptive segment and uses the divergence at the departure point to modulate the segment' s advantage. Experiments across eight mathematical reasoning and agentic tool-use benchmarks with Qwen3 4B and 8B models show that GEAR consistently outperforms standard GRPO, self-distillation-only baselines, and token- or turn-level credit-assignment methods. The gains are especially strong on benchmarks with lower GRPO baseline accuracy, reaching up to around 20% over GRPO, suggesting that the proposed adaptive reweighting scheme is especially useful in more challenging long-horizon settings.

Group Relative Policy Optimization (GRPO) has emerged as a promising approach for improving the reasoning capabilities of large language models. However, it struggles to effectively balance the tradeoff between exploration and exploitation during training, often resulting in suboptimal performance. Motivated by the theoretical insight that changes in entropy are governed by the covariance between token probabilities and their corresponding advantages, we propose a hyperparameter-free, covariance-weighted optimization method that dynamically down-weights extreme token-level updates via a Gaussian kernel. This approach automatically reduces the instability caused by exploration-exploitation trade-off while preserving informative learning signals. Extensive empirical evaluations show that our approach improves downstream performance across reasoning benchmarks compared with GRPO, and effectively stablizes entropy as training progresses.

Soft Actor-Critic (SAC) and its variants dominate Multi-Task Reinforcement Learning (MTRL) due to their off-policy sample efficiency, while on-policy methods such as Proximal Policy Optimization (PPO) remain underexplored. We diagnose that PPO in MTRL suffers from a previously overlooked issue: critic-side gradient ill-conditioning, which may cause tail tasks to stall while easy tasks dominate the value function's updates. To address this, we propose TOPPO (Tail-Optimized PPO), a reformulation of PPO via Critic Balancing -- a set of modules that improve gradient conditioning and balance learning dynamics across tasks. Unlike prior approaches that rely on modular architectures or large models, TOPPO targets the optimization bottleneck within PPO itself. Empirically, TOPPO achieves stronger mean and tail-task performance than published SAC-family and ARS-family baselines while using substantially fewer parameters and environment steps on Meta-World+ benchmark. Notably, TOPPO matches or surpasses strong SAC baselines early in training and maintains superior performance at full budget. Ablations confirm the effectiveness of each module in TOPPO and provide insights into their interactions. Our results demonstrate that, with proper optimization, on-policy methods can rival or exceed off-policy approaches in MTRL, challenging the prevailing reliance on SAC and highlighting critic-side gradient conditioning as the central bottleneck.

Decentralized multi-agent path finding (MAPF) routes a team of agents on a shared grid, each acting from its own local view. The standard solution trains one shared neural policy with Proximal Policy Optimization (PPO), a popular on-policy reinforcement learning algorithm. Such a policy works well on clean observations, but a small input perturbation on one agent often changes its action, which then blocks a neighbour, and the team jams. In this paper we present two training recipes that keep the same network and the same deployment loop, yet make the policy hold up under perturbed observations. The first recipe, Adv-PPO, trains the shared policy against worst-case perturbations of its own input and selects the checkpoint by performance under adversarial perturbation. The second recipe, Adv-PPO+MACER, fine-tunes that checkpoint with a small on-policy smoothness term whose gradient follows the certified radius of randomized smoothing. On POGEMA with 8x8 maps and four agents, the unprotected PPO policy reaches 95.8% clean success but only 2.5% under the strongest attack. Adv-PPO recovers worst-case success to 59.2% at one percentage point of clean cost. Adv-PPO+MACER recovers it to 77.5% +/- 6.0% across three independent seeds at less than one percentage point of clean cost. We support these numbers with per-attack curves, a certified action-stability sanity check (which measures the smoothed-policy wrapper, not the deployed argmax policy), and side-by-side rollout storyboards that show the failure mode and the fix inside one environment instance.

Reasoning models post-hoc rationalize answers they have already committed to internally, producing chains of reasoning theater: deliberative-looking steps that contribute nothing to correctness. This wastes inference tokens, pollutes interpretability, and obscures what the model actually computed. We introduce ProFIL (Probe-Filtered Reinforcement Learning) to reduce theater, increase chain-of-thought faithfulness, and shrink chain length in a single, drop-in extension to Group Relative Policy Optimization (GRPO). A multi-head attention probe is trained once on the frozen base model to detect post-commitment steps from internal activations alone; during GRPO, rollouts whose probe score exceeds a threshold have their advantage zeroed. Our central finding is that a probe trained on a frozen base, with verifier-derived labels and no human annotation, provides a stable signal that suppresses theater while resisting the RL-obfuscation failure mode predicted by prior work. Across four reasoning domains (GSM8K, LiveCodeBench, ToolUse, MMLU-Redux) and two model architectures (Llama-8B, Qwen-7B), ProFIL reduces post-commitment theater by 11--100%, raises faithful-fraction (e.g., +24pp on LiveCodeBench under an independent Claude 3.7 Sonnet judge), and shortens chains by 4--19%, all while preserving or improving task accuracy. ProFIL also beats a matched length-penalty GRPO baseline, isolating the gain as semantic commitment-detection rather than chain compression. Probe weights, training configurations, and rollouts are released across all four domains.

Reinforcement learning with verifiers (RLVR) has become a central paradigm for improving LLM reasoning, yet popular group-based optimization algorithms like GRPO often suffer from exploration collapse, where the models prematurely converge on a narrow set of high-scoring patterns, lacking the ability to explore new solutions. Recent efforts attempt to alleviate this by adding entropy regularization or diversity bonus. However, these approaches do not change the \textit{winner-takes-all} nature, where rollouts still compete for individual advantage rather than cooperating for maximizing global diversity. In this work, we propose Group Cooperative Policy Optimization (GCPO), which shifts the training paradigm from rollout competition to team cooperation. Specifically, GCPO replaces independent rollout scoring with team-level credit assignment: a rollout is rewarded by how much it contributes to the team's valid solution coverage, rather than its individual accuracy. This coverage is described as a determinant volume over reward-weighted semantic embeddings, where only correct and non-redundant rollouts contribute to this volume. During advantage estimation, GCPO redistributes the collective team reward to each single rollout according to its average marginal contribution to the team. This cooperative training paradigm routes optimization toward non-redundant correct reasoning paths. Experiments across multiple reasoning benchmarks demonstrate that GCPO significantly improves both reasoning accuracy and solution diversity over existing approaches. Code will be released at https://github.com/bradybuddiemarch/gcpo.

Reinforcement Learning with Verifiable Rewards (RLVR) has become the standard paradigm for LLM mathematical reasoning, with Group Relative Policy Optimization (GRPO) serving as the dominant algorithm. We identify two overlooked inefficiencies inherent in GRPO. First, a fixed KL coefficient overly restricts policy exploration at moments when the model needs to diverge significantly from the reference policy. Second, uniform question sampling overlooks that moderately difficult problems produce the most informative gradient signals. We propose FG-ExPO, short for Frontier-Guided Exploration-Prioritized Policy Optimization, which integrates two lightweight components. Accuracy-Conditioned KL Scaling (AKL) adjusts the KL penalty strength through a smooth nonlinear function of batch average accuracy, loosening the constraint when the model performs poorly and strengthening it when the model achieves satisfactory results. Gaussian Curriculum Sampling (GCS) assigns sampling weights to questions following a Gaussian distribution centered at a moderate accuracy level around 0.5, focusing model training on its learning frontier. We conduct evaluations on DeepSeek-R1-Distill-Qwen-1.5B and Qwen3-8B-Base across six mainstream mathematical reasoning benchmarks. Experimental results demonstrate that FG-ExPO consistently outperforms vanilla GRPO. It delivers an absolute improvement of 13.34 on the AIME 2025 pass@32 metric, rising from 63.33 percent to 76.67 percent, and obtains an average pass@32 gain of 2.66 on the 8B model. The substantially larger performance gains observed on pass@32 compared to pass@1 verify that FG-ExPO enlarges the model's effective exploration space under a fixed inference budget.

Automated scientific discovery using large language models relies on identifying genuinely novel solutions. Standard reinforcement learning penalizes high-variance mutations, which leads the policy to prioritize familiar patterns. As a result, the maximum reward plateaus even as the average reward increases. Overcoming this limitation requires a signal that distinguishes unexplored regions from intrinsically difficult problems. This necessitates measuring disagreement across independently adapted weight hypotheses rather than relying on a single network's confidence. UG-TTT addresses this challenge by maintaining a small ensemble of low-rank adapters over a frozen base model. The per-token disagreement, quantified as the mutual information between ensemble predictions and weight hypotheses, isolates epistemic uncertainty and identifies positions where insufficient coverage leads to adapter divergence rather than intrinsic problem difficulty. This measure is incorporated as an exploration bonus into the policy gradient, directing the policy toward positions where persistent adapter disagreement signals low training coverage, the same frontier where genuine discovery is possible. A nuclear norm regularizer ensures the adapters remain distinct from one another, thereby preserving the exploration signal throughout training. Across four scientific discovery benchmarks, UG-TTT increases the maximum reward on three tasks, maintains substantially higher solution diversity, and an ablation study confirms that the regularizer is essential for sustaining this behavior.

Neural warm starts can sharply reduce the number of Newton-Raphson iterations required to solve the AC power flow problem, but existing supervised approaches generalize poorly on heavily loaded instances near voltage collapse. We prove a lower bound on the Newton-Raphson iteration count that depends on the direction of the warm start error rather than on its magnitude, and show as a corollary that the bound becomes vacuous as the smallest singular value of the power-flow Jacobian shrinks, identifying the failure mode of supervised regression near the saddle-node bifurcation. Motivated by this analysis, we introduce Newton's Lantern, a finetuning pipeline that combines group relative policy optimization with a learned reward model trained on perturbations of the base model's predictions, using the iteration count itself as the supervisory signal. Across IEEE 118-bus, GOC 500-bus, and GOC 2000-bus benchmarks, Newton's Lantern is the only method that converges on every test snapshot while attaining the smallest mean iteration count.

Recently, post-training methods based on reinforcement learning, with a particular focus on Group Relative Policy Optimization (GRPO), have emerged as the robust paradigm for further advancement of text-to-image (T2I) models. However, these methods are often prone to reward hacking, wherein models exploit biases in imperfect reward functions rather than yielding genuine performance gains. In this work, we identify that normalization could lead to miscalibration and directly removing the prompt-level standard deviation term yields an optimal policy ascent direction that is linear in the advantage but still limits the separation of genuine signals from noise. To mitigate the above issues, we propose Super-Linear Advantage Shaping (SLAS) by revisiting the functional update from an information geometry perspective. By extending the Fisher-Rao information metric with advantage-dependent weighting, SLAS introduces a non-linear geometric structure that reshapes the local policy space. This design relaxes constraints along high-advantage directions to amplify informative updates, while tightening those in low-advantage regions to suppress illusory gradients. In addition, batch-level normalization is applied to stabilize training under varying reward scales. Extensive evaluations demonstrate that SLAS consistently surpasses the DanceGRPO baseline across multiple backbones and benchmarks. In particular, it yields faster training dynamics, improved out-of-domain performance on GenEval and UniGenBench++, and enhanced robustness to model scaling, while mitigating reward hacking and preserving semantic and compositional fidelity in generations.

This work revisits standard policy gradient methods used on restricted policy classes, which are known to get stuck in suboptimal critical points. We identify an important cause for this phenomenon to be that the policy gradient is itself fundamentally myopic, i.e. it only improves the policy based on the one-step $Q$-function. In this work, we propose a generalized $k$-step policy gradient method that couples the randomness within a $k$-step time window and can escape the myopic local optima in MDPs with restricted policy classes. We show this new method is theoretically guaranteed to converge to a solution that is exponentially close in performance to the optimal deterministic policy with respect to $k$. Further, we show projected gradient descent and mirror descent with this $k$-step policy gradient can achieve this exponential guarantee in $O(\frac{1}{T})$ iterations, despite only assuming smoothness and differentiability of the value function. This will provide near optimal solutions to previously elusive applications like state aggregation and partially observable cooperative multi-agent settings. Moreover, our bounds avoid the ubiquitous distribution mismatch factors $||d_μ^{π^*} / d_μ^π||_\infty$ and $||d_μ^{π^*} / μ||_\infty$ enabling the $k$-step policy gradient method to escape suboptimal critical points that emerge from poor exploration in fully observable settings.

Training deep research agents, namely systems that plan, search, evaluate evidence, and synthesize long-form reports, pushes reinforcement learning beyond the regime of verifiable rewards. Their outputs lack ground-truth answers, their trajectories span many tool-augmented decisions, and standard post-training offers little mechanism for turning past attempts into reusable experience. In this work, we argue that rubrics should serve not merely as final-answer evaluators, but as the shared interface that structures policy execution, judge feedback, and agent memory. Based on this view, we introduce RubricEM, a rubric-guided reinforcement learning framework that combines stagewise policy decomposition with reflection-based meta-policy evolution. RubricEM first makes research trajectories stage-aware by conditioning planning, evidence gathering, review, and synthesis on self-generated rubrics. It then assigns credit with Stage-Structured GRPO, which uses stagewise rubric judgments to provide denser semantic feedback for long-horizon optimization. In parallel, RubricEM trains a shared-backbone reflection meta-policy that distills judged trajectories into reusable rubric-grounded guidance for future attempts. The resulting RubricEM-8B achieves strong performance across four long-form research benchmarks, outperforming comparable open models and approaching proprietary deep-research systems. Beyond final performance, we perform thorough analyses to understand the key ingredients of RubricEM.

Although Large Language Models (LLMs) have made remarkable progress, current preference optimization methods still struggle to align directional consistency while preserving reasoning diversity. To address this limitation, we propose Directional-Groupwise Preference Optimization (DGPO), a lightweight framework that aggregates supervision signals at the group level and explicitly models direction-aware alignment through multi-candidate comparisons. DGPO organizes forward and reverse question-answer instances into structured sets and optimizes a margin-based likelihood objective that separates coherent reasoning paths from inconsistent alternatives. This group-wise formulation captures richer relative information than pairwise objectives and reinforces consistency across diverse reasoning pathways. Empirical results show that our constructed reverse data yields a 3.2% average improvement across five benchmarks, while DGPO further delivers consistent gains across multiple datasets and model families, achieving average accuracy improvements of up to 3.6%.

We study policy gradient methods for reinforcement learning in non-Markovian decision processes (NMDPs), where observations and rewards depend on the entire interaction history. To handle this dependence, the agent maintains an internal state that is recursively updated to provide a compact summary of past observations and actions. In contrast to approaches that treat the agent state dynamics as fixed or learn it via predictive objectives, we propose a reward-centric formulation that jointly optimizes the agent state dynamics and the control policy to maximize the expected cumulative reward. To this end, we consider a class of Agent State-Markov (ASM) policies, comprising an agent state dynamics and a control policy that maps the agent state to actions. We establish a novel policy gradient theorem for ASM policies, extending the classical policy gradient results from the Markovian setting to episodic and infinite-horizon discounted NMDPs. Building on this gradient expression, we propose the Agent State-Markov Policy Gradient (ASMPG) algorithm, which leverages the recursive structure of the agent state dynamics for efficient optimization. We establish finite-time and almost sure convergence guarantees, and empirically demonstrate that, on a range of non-Markovian tasks, ASMPG outperforms baselines that learn state representations via predictive objectives.

Multi-negative preference optimization under the Plackett--Luce (PL) model extends Direct Preference Optimization (DPO) by leveraging comparative signals across one preferred and multiple rejected responses. However, optimizing over large negative pools is costly, and many candidates contribute redundant gradients due to their similar effects on policy updates. We introduce MASS-DPO, a multi-negative active sample selection method that derives a PL-specific Fisher-information objective for selecting compact, informative negative subsets within each prompt. The resulting log-determinant objective selects negatives that contribute complementary information for policy updates, yielding compact subsets that retain the full pool's information while reducing redundancy. In practice, this favors negatives whose gradients cover different update directions, reducing redundant signal from near-duplicate candidates while preserving the most useful training information. Across four benchmarks spanning recommendation and multiple-choice QA and three model families, MASS-DPO consistently exceeds or matches existing methods in accuracy, improves Recall/NDCG and margin-based optimization dynamics, and delivers stronger alignment with substantially fewer negatives.

In this work, we show that natural policy gradient, a core algorithm in reinforcement learning, admits an exact formulation as a smoothed and averaged form of policy iteration. Specifically, we introduce doubly smoothed policy iteration (DSPI), a Bellman-operator framework in which each policy is obtained by applying a regularized greedy step to a weighted average of past $Q$-functions. DSPI includes policy iteration, dual-averaged policy iteration, natural policy gradient, and more general policy dual averaging methods as special cases. Using only monotonicity and contraction of smoothed Bellman operators, we prove distribution-free global geometric convergence of DSPI. Consequently, standard natural policy gradient and policy dual averaging achieve an iteration complexity of $\mathcal{O}((1-γ)^{-1}\log((1-γ)^{-1}ε^{-1}))$ for computing an $ε$-optimal policy, without modifying the MDP, adding regularization beyond the mirror map inherent in the update, or using adaptive, trajectory-dependent stepsizes. For the unregularized greedy case, corresponding to dual-averaged policy iteration, we also prove finite termination. The same Bellman-operator framework further extends to discounted MDPs with linear function approximation and stochastic shortest path problems.

LLM-based automatic heuristic design has shown promise for generating executable heuristics for combinatorial optimization, but existing methods mainly rely on delayed endpoint performance. We propose a \emph{teacher-aware evolutionary framework} that uses independently trained learned optimization policies as behavioral teachers. Instead of deploying or imitating the teacher, our method queries it on states visited by candidate heuristic programs and uses its action preferences as local feedback for evolution. The resulting search discovers static executable heuristics guided by both task performance and teacher-derived behavioral signals. Experiments on scheduling, routing, and graph optimization benchmarks show that our method improves over performance-driven LLM heuristic evolution baselines while requiring no neural inference at deployment. These results suggest that learned optimization policies can be repurposed as behavioral feedback sources for automatic heuristic discovery.

Efficient branching policies are essential for accelerating Mixed Integer Linear Programming (MILP) solvers. Their design has long relied on hand-crafted heuristics, and now machine learning has emerged as a promising paradigm to automate this process. However, existing learning-based methods are often hindered by their dependence on expensive expert demonstrations and the gap between training objectives and the solver's end-to-end performance. In this work, we propose LLM4Branch, a novel framework that leverages Large Language Models (LLMs) to automate the discovery of efficient branching policies. Specifically, the discovered policy is an executable program with a program skeleton generated by the LLM and a parameter vector, which is optimized via a zeroth-order method over a few instances with their end-to-end performance feedback. Extensive experiments on standard MILP benchmarks demonstrate that LLM4Branch establishes a new state-of-the-art among CPU-based methods and achieves performance competitive with advanced GPU-based models. Codes are available at https://github.com/hzn18/LLM4Branch.

Diffusion large language models (dLLMs) offer a promising route to parallel and efficient text generation, but improving their reasoning ability requires effective post-training. Reinforcement learning with verifiable rewards (RLVR) is a natural choice for this purpose, yet its application to dLLMs is hindered by the absence of tractable sequence-level log-ratios, which are central to standard policy optimization. The lack of tractable sequence-level log-ratios forces existing methods to rely on high-variance ELBO-based approximations, where high verifier rewards can amplify inaccurate score estimates and destabilize RL training. To overcome this issue, we propose \textbf{R}elative \textbf{S}core \textbf{P}olicy \textbf{O}ptimization (RSPO), a simple RLVR method that uses verifiable rewards to calibrate noisy likelihood estimates in dLLMs. The core of our algorithm relies on a key observation: a reward advantage can be interpreted not only as an update direction, but also as a target for the relative log-ratio between the current and reference policies. Accordingly, RSPO calibrates this noisy relative log-ratio estimate by comparing its reward advantage with the reward-implied target relative log-ratio, updating the policy according to the gap between the current estimate and the target rather than the raw advantage alone. Experiments on mathematical reasoning and planning benchmarks show that RSPO yields especially strong gains on planning tasks and competitive mathematical-reasoning performance.

Proactive streaming video understanding requires models to continuously process video streams and decide when to respond, rather than merely what to respond. This naturally introduces a decision-making problem under partial observations, where models must balance early prediction against sufficient evidence. However, existing benchmarks largely follow a "see-then-answer" paradigm, where responses are triggered only after explicit evidence appears, effectively reducing proactive reasoning to delayed perception. As a result, they fail to evaluate a model's ability to make timely and reliable decisions under incomplete observations. Moreover, training proactive models is inherently challenging due to the extreme imbalance between silence and response signals in streaming trajectories, as well as the need to jointly optimize response correctness and timing. To address these challenges, we introduce StreamPro-Bench, a new benchmark that evaluates streaming models from three complementary perspectives: Perception Understanding, Temporal Reasoning, and Proactive Agency, where the last measures a model's ability to make early yet reliable decisions under partial observations. We further propose StreamPro, a two-stage training framework for proactive learning. First, we introduce CB-Stream Loss to mitigate the severe supervision imbalance during supervised fine-tuning (SFT). Then, we apply Group Relative Policy Optimization (GRPO) with a multi-grained reward design that involves both turn-level and trajectory-level rewards. Experiments show that StreamPro significantly improves proactive performance. On StreamPro-Bench, it achieves 41.5, substantially outperforming the previous best (10.4), while also maintaining strong performance on real-time streaming benchmarks, achieving 78.9 on StreamingBench-RTVU.

Reinforcement Learning with Verifiable Rewards (RLVR) has become the standard paradigm for LLM mathematical reasoning, where Group Relative Policy Optimization (GRPO) serves as the mainstream algorithm. We point out two understudied inefficiencies existing in GRPO. First, the fixed KL penalty coefficient overly restricts policy exploration at stages where the model requires significant deviation from the reference policy. Second, uniform sampling of training questions ignores that moderately difficult problems provide the most informative gradient signals for optimization. We propose Exploration-Prioritized Policy Optimization (EXPO) with two lightweight plug-in modules. The Accuracy-Conditioned KL Scaling (AKL) dynamically adjusts KL regularization strength through a smooth nonlinear function of batch average accuracy, relaxing the penalty when the model underperforms and strengthening it when the model achieves good results. The Gaussian Curriculum Sampling (GCS) assigns sampling weights to questions following a Gaussian distribution centered at moderate accuracy around 0.5, focusing training on the model's learning frontier. We conduct extensive experiments on DeepSeek-R1-Distill-Qwen-1.5B and Qwen3-8B-Base over six mathematical reasoning benchmarks. The results show EXPO steadily surpasses vanilla GRPO. It obtains an absolute gain of 13.34 on AIME 2025 pass@32, rising from 63.33 percent to 76.67 percent, and achieves an average pass@32 improvement of 2.66 on the 8B model. The much larger performance gains on pass@32 compared with pass@1 demonstrate that EXPO effectively enlarges the model's exploration boundary under a fixed inference cost budget.

Recent studies suggest that Reinforcement Fine-Tuning (RFT) is inherently more resilient to catastrophic forgetting than Supervised Fine-Tuning (SFT). However, whether RFT (e.g., GRPO) can effectively overcome forgetting in challenging visual continual learning settings, such as class-incremental learning (CIL) and domain-incremental learning (DIL), remains an open problem. Through a pilot study, we confirm that while RFT consistently outperforms SFT, it still suffers from non-negligible forgetting. We empirically trace this bottleneck to Trajectory-level Drift Agnosticism: among candidate rollouts achieving identical task rewards, the KL divergence from the preceding-task policy varies substantially, which strongly correlates with catastrophic forgetting across sequential tasks. Motivated by this insight, we propose Retention-aware Policy Optimization (RaPO), a simple yet effective RFT method that explicitly mitigates forgetting through trajectory-level reward shaping. Specifically, RaPO comprises two core components: (1) Retention Reward that converts trajectory-level distribution drift into a continuous reward signal, preferentially reinforcing knowledge-preserving rollouts within each group; (2) Cross-Task Advantage Normalization (CTAN), which maintains a persistent exponential moving average of reward statistics across task boundaries to stabilize the optimization progress during continual learning. Leveraging the free-form textual generalization of MLLMs, we comprehensively evaluate RaPO across five visual continual learning settings. Extensive experiments demonstrate that RaPO achieves leading performance, substantially reducing catastrophic forgetting while preserving strong plasticity. To the best of our knowledge, this work represents the first systematic exploration of RFT in visual continual learning, offering insights that we hope will inspire future research.

Vision-Language-Action (VLA) models remain brittle in long-horizon, contact-rich manipulation because success-only imitation provides little supervision for execution drift, while failed rollouts are often discarded. We introduce RePO-VLA, a recovery-driven policy optimization framework that assigns distinct roles to success, recovery, and failure trajectories. RePO-VLA first applies Recovery-Aware Initialization (RAI), slicing recovery segments and resetting history so corrective actions depend on the current adverse state rather than the preceding failure. It then learns a Progress-Aware Semantic Value Function (PAS-VF), aligning spatiotemporal trajectory features with instructions and successful references. The resulting labels salvage useful failure prefixes via reliability decay, while low-value labels mark drift and terminal breakdowns, teaching differences among nominal, failed, and corrective actions. The data engine turns adverse states into planner-generated or human-collected corrective rollouts, teaching recovery to the success manifold. Value-Conditioned Refinement (VCR) trains the policy to prefer high-progress actions. At deployment, a fixed high value ($v=1.0$) biases actions toward the learned success manifold without online failure detectors or heuristic retries. We introduce FRBench, with standardized error injection and recovery-focused evaluation. Across simulated and real-world bimanual tasks, RePO-VLA improves robustness, raising adversarial success from 20% to 75% on average and up to 80% in scaled real-world trials.

Discrete flow models (DFMs) are a class of flexible generative models for generating discrete data, and diffusion large language models (dLLMs) can be viewed as a special case with a specific choice of mixture path and a masked source distribution. While several recent works have explored reinforcement learning into dLLMs, its application to more general discrete flow models remains underexplored. In this work, we present discrete Flow-GRPO (dFlowGRPO), a unified reinforcement learning framework for discrete flow models that supports a broad family of probability paths and non-masked source distributions. We derive the full trajectory probability for DFMs and formulate denoising as a Markov decision process, enabling dFlowGRPO to incorporate information from both the associated conditional transition rates and the posterior model during reinforcement learning. We apply dFlowGRPO to FUDOKI, a recent multimodal discrete flow model, and evaluate it on both image generation and multimodal understanding tasks. Empirical results show that dFlowGRPO outperforms existing GRPO-type methods for dLLMs on text-to-image generation tasks and achieves performance competitive with continuous flow-based models trained using FlowGRPO, while also demonstrating strong capabilities on understanding tasks.

Group Relative Policy Optimization (GRPO) has emerged as a powerful algorithm for improving the reasoning capabilities of language models, but often fails to improve small models due to sparse rewards on difficult tasks. Existing works mitigate this issue by leveraging a larger model, either to provide hints for rollouts or to provide dense reward signals through knowledge distillation (KD). However, this assumes the existence of such an oracle, and training one can significantly increase total training time. In this work, we propose CoDistill-GRPO, a co-distillation algorithm that simultaneously trains a large and a small model by maximizing carefully designed GRPO objectives. The two models learn from each other: the small model uses an on-policy KD reward to learn from the large model's distribution, while the large model is updated using rollouts generated by the small model with importance reweighting, reducing the computational overhead of rollout generation. We show that CoDistill-GRPO substantially improves small model performance over standard GRPO on mathematical benchmarks across both Qwen and Llama models. Specifically, with Qwen2.5-Math-1.5B, we observe an accuracy increase of over 11.6 percentage points over the base model and an additional 6.0 percentage points over GRPO on the Minerva dataset. Interestingly, the larger model (Qwen2.5-Math-7B) trained with CoDistill-GRPO nearly matches standard GRPO performance despite training on small-model rollouts. This highlights CoDistill-GRPO as a cost-effective alternative to GRPO for larger models, yielding an approximate 18% speedup, which may be of independent interest.

Multi-agent large language model (LLM) systems often rely on a controller to coordinate a pool of heterogeneous models, yet existing controllers are typically limited to one-shot routing: they select a model once and return its output directly. Such routing-only designs provide no mechanism to critique intermediate drafts or support iterative refinement. To address this limitation, we propose a critique-and-routing controller that casts multi-agent coordination as a sequential decision problem. At each turn, the controller evaluates the current draft, decides whether to stop or continue, and, if needed, selects the next agent for further refinement. We formulate this process as a finite-horizon Markov Decision Process (MDP) with explicit agent-utilization constraints, design a composite reward for controller decisions across turns, and optimize the controller via policy gradients under a Lagrangian-relaxed objective. Extensive experiments across multiple heterogeneous multi-agent systems and seven reasoning benchmarks show that our method consistently outperforms state-of-the-art baselines and substantially narrows the gap to the strongest agent, while using it for fewer than 25% of total calls.

Existing Flow Matching (FM) text-to-image models suffer from two critical bottlenecks under multi-task alignment: the reward sparsity induced by scalar-valued rewards, and the gradient interference arising from jointly optimizing heterogeneous objectives, which together give rise to a 'seesaw effect' of competing metrics and pervasive reward hacking. Inspired by the success of On-Policy Distillation (OPD) in the large language model community, we propose Flow-OPD, the first unified post-training framework that integrates on-policy distillation into Flow Matching models. Flow-OPD adopts a two-stage alignment strategy: it first cultivates domain-specialized teacher models via single-reward GRPO fine-tuning, allowing each expert to reach its performance ceiling in isolation; it then establishes a robust initial policy through a Flow-based Cold-Start scheme and seamlessly consolidates heterogeneous expertise into a single student via a three-step orchestration of on-policy sampling, task-routing labeling, and dense trajectory-level supervision. We further introduce Manifold Anchor Regularization (MAR), which leverages a task-agnostic teacher to provide full-data supervision that anchors generation to a high-quality manifold, effectively mitigating the aesthetic degradation commonly observed in purely RL-driven alignment. Built upon Stable Diffusion 3.5 Medium, Flow-OPD raises the GenEval score from 63 to 92 and the OCR accuracy from 59 to 94, yielding an overall improvement of roughly 10 points over vanilla GRPO, while preserving image fidelity and human-preference alignment and exhibiting an emergent 'teacher-surpassing' effect. These results establish Flow-OPD as a scalable alignment paradigm for building generalist text-to-image models. The codes and weights will be released in: https://github.com/CostaliyA/Flow-OPD .

Direct Preference Optimization (DPO) aligns language models using pairwise preference comparisons, offering a simple and effective alternative to Reinforcement Learning (RL) from human feedback. However, in many practical settings, training data consists of multiple rollouts per prompt, inducing rich preference structure that pairwise DPO fails to exploit. Collapsing such data into independent pairs discards transitivity, introduces redundant or conflicting supervision, and can lead to unstable optimization. We propose Graph Direct Preference Optimization (GraphDPO), a principled generalization of DPO that operates over directed acyclic preference graphs induced by rollout rankings. GraphDPO encodes dominance relations as edges and optimizes a graph-structured Plackett--Luce-inspired objective that aggregates supervision over graph neighborhoods, enforcing transitivity while recovering standard DPO as a special case. To handle discrete or sparse signals, we introduce an equivalence-class construction where responses with identical preferences form graph layers, and intra-layer edges contribute zero loss, preventing spurious gradients. Despite leveraging full graph structure, GraphDPO maintains linear per-prompt complexity via efficient log-sum-exp aggregation. We further incorporate optional ground-truth anchoring by inserting verified solutions as dominant nodes and applying an annealed schedule that stabilizes early training while gradually relaxing oracle supervision. Experiments on reasoning and program synthesis tasks demonstrate superior performance, suggesting that graph-structured preference modeling is a scalable and robust alternative to pairwise and listwise alignment objectives.

In structured decision-making workflows such as form filling, compliance checking, and maintenance reporting, LLM outputs must be locally correct, globally consistent, and auditable against task-specific rules. Existing refinement methods often rely on heuristic debate, self-play, or LLM-generated supervision, creating a second-order assurance problem. We propose DPA-GRPO (Dual Paired-Action Group-Relative Policy Optimization), a paired-action training method for a two-player generator--verifier game with structured verifier interventions. The generator proposes outputs and may revise them when challenged; the verifier either remains silent or raises a safety assurance case (SAC) containing a claim, argument, and evidence. These SAC/no-SAC and KEEP/REVISE decisions induce paired counterfactual action groups, which DPA-GRPO uses for role-specific KL-regularized GRPO updates. We analyze the unregularized game and show that positive probability on strictly lower-reward intervention or revision actions creates a profitable unilateral deviation. Under standard stochastic-approximation assumptions, DPA-GRPO tracks the corresponding game ODE, whose isolated asymptotically stable limit points are stationary and candidate local equilibria under role-wise local optimality. Experiments on TaxCalcBench TY24 show that DPA-GRPO improves structured decision accuracy over zero-shot generation and generator-only RL baselines across Qwen3-4B and Qwen3-8B. Training increases correct silent acceptance, reduces missed errors, and improves calibrated revision behavior, indicating gains for both generator and verifier.

Communication is fundamental to sustaining reciprocity and cooperation in strategic interactions. We identify and formulate the influence attribution problem as the central optimization difficulty inherent in such dynamics for a learning agent: any action or signal the agent emits reshapes the reputations of many third parties along combinatorially branching paths before feeding back into its own future rewards, forcing the agent to account for all of these indirect channels at once when choosing every action. To address this, we introduce the reciprocity gradient, which explicitly backpropagates reward gradients through private estimators of opponents' policies trained from public observations. The gradient flows through the reputation chain itself analytically, rather than being estimated from sampled returns. It jointly optimizes actions and evaluative signals without intrinsic rewards or reward shaping. Empirically, the method recovers near-optimal context-sensitive policies, while sample-based baselines collapse into constant-output policies.

Existing LLM-based policy optimizers see only scalar rewards: that a policy scored 0.45, but not whether the agent got stuck in a loop, fell into a hole on the third step, or performed well on 19 out of 20 rollouts and failed catastrophically on one. We propose Reflective Prompted Policy Optimization (R2PO), a two-stage LLM framework for policy search over compact policy classes that augments scalar reward feedback with trajectory-level behavioral evidence. A Search-LLM proposes candidate policy parameters; the environment executes them; a Critic-LLM inspects the resulting rollouts and proposes targeted revisions grounded in observed states, actions, and rewards. Across ten environments, ablations show R2PO's gains require separating global search from behavior-grounded revision and using selection to filter high-variance edits. We further identify a dominant failure mode, salience bias: when presented with multiple rollouts, the Critic-LLM fixates on improving a single failure even when most trajectories succeed. In a three-trajectory variant where the Critic-LLM sees the best, worst, and median rollout, this behavior explains 76.6% of regressions on CartPole. R2PO mitigates this by reasoning over aggregate rollout statistics, median-trajectory selection, and a revision rule. Using a 20B open-weight model, R2PO achieves the highest mean best reward across all ten environments, reaches near-optimal performance substantially earlier (e.g., near-maximum CartPole reward within ~500 episodes), and trains far more stably than both deep RL and prior LLM-based methods. These results show that treating trajectories as first-class in-context evidence, rather than artifacts reduced to scalar returns, changes how even comparatively small LLMs search over policy spaces, enabling them to learn faster, diagnose more precisely, and reliably improve external controllers.

Balancing exploration and exploitation is a core challenge in sequential decision-making and black-box optimization. We introduce POETS ($\textbf{Po}$licy $\textbf{E}$nsembles for $\textbf{T}$hompson $\textbf{S}$ampling), a novel framework that bridges uncertainty quantification and policy optimization. Our approach is grounded in the insight that policies trained with Kullback-Leibler (KL) regularization implicitly encode an underlying reward function. Building on this, POETS bypasses the complex, nested process of training an uncertainty-aware reward model and separately fitting a policy to this model. Instead, we directly train a policy ensemble to capture epistemic uncertainty by matching implicitly encoded reward functions to online, bootstrapped data. To overcome the prohibitive compute and memory constraints of ensembling Large Language Models (LLMs), POETS utilizes an efficient architecture: the ensemble shares a pre-trained backbone while maintaining diversity through independent Low-Rank Adaptation (LoRA) branches. Theoretically, we prove that POETS implicitly conducts KL-regularized Thompson sampling and thus inherits strong cumulative regret bounds of ${\mathcal O}(\sqrt{T γ_T})$. Empirically, we demonstrate that POETS achieves state-of-the-art sample efficiency across diverse scientific discovery domains, including protein search and quantum circuit design. Furthermore, it improves the optimization trajectories of reinforcement learning, proving particularly robust in off-policy settings with experience replay or in small dataset regimes.

Group Relative Policy Optimization (GRPO) is a standard algorithm for reinforcement learning from verifiable rewards, but its group-mean-centered advantage can fail under binary rewards. The failure mode is gradient starvation: when every response in a group is correct or every response is wrong, the centered advantage is exactly zero and the policy receives no learning signal. We prove that the true degeneracy rate always exceeds the i.i.d. Bernoulli prediction by Jensen's inequality, and observe a 0.69 degeneracy rate at group size four in logged Qwen3.5-9B GSM8K training. We then show that the fixed-reference Sign advantage, $A=2r-1$, performs pass@$G$ failure descent by increasing the probability that at least one sample in the group succeeds. On the full GSM8K test set across seven seeds, Sign reaches 73.8% accuracy versus 28.4% for standard normalized group-mean DrGRPO at group size four, a 45.4 point gain with $p<0.0001$. The effect is directionally consistent on Llama-3.1-8B and positive but underpowered on a MATH-500 transfer check. Pass@$k$ analysis indicates that the main benefit is search compression rather than large capacity expansion, aligning the empirical gains with recent RLVR ceiling observations.

Differentiable planning enables gradient-based optimization of decision-making problems by leveraging differentiable models of system dynamics. However, in highly nonlinear and hybrid discrete-continuous domains, the resulting optimization landscapes are often ill-conditioned, with flat regions and sharp transitions that hinder effective optimization. We propose Model-Driven Policy Optimization (MDPO), a framework that introduces stochastic exploration into differentiable planning by injecting noise into the action space during optimization. Leveraging access to the model, MDPO further adapts the noise magnitude based on gradient-derived sensitivity of the trajectory objective, yielding a time-dependent exploration profile. This enables improved exploration of the objective landscape and helps escape poor local optima via dynamic allocation of exploration across timesteps and iterations. Experiments on benchmark domains demonstrate that MDPO consistently outperforms deterministic differentiable planning, including both the noise-free variant of our method and available state-of-the-art implementations, as well as model-free baselines such as PPO, significantly improving solution quality across challenging nonlinear and hybrid settings. We further analyze the evolution of the adaptive noise magnitude across both time steps and optimization iterations, providing insight into how exploration is allocated during learning.

Adaptive rank allocation for LoRA, allocating more parameters to important layers and fewer to unimportant ones, consistently improves efficiency under supervised fine-tuning (SFT). We investigate whether this success transfers to reinforcement learning, specifically Group Relative Policy Optimization (GRPO). Using gradient-magnitude profiling on Qwen 2.5 1.5B with GSM8K, we find that it does not: proportional rank allocation degrades accuracy by 4.5 points compared to uniform allocation (70.0% vs. 74.5%), despite using identical parameter budgets. We identify two mechanisms behind this failure. First, the gradient landscape under GRPO is fundamentally flatter than under SFT, the max-to-min layer importance ratio is only 2.17x, compared to >10x reported in SFT literature. All layers carry meaningful gradient signal; none are truly idle. Second, we discover a gradient amplification effect: non-uniform allocation widens the importance spread from 2.17x to 3.00x, creating a positive feedback loop where high-rank layers absorb more gradient while low-rank layers are progressively silenced. Our results suggest that gradient importance does not predict capacity requirements under RL, and that naive transfer of SFT-era rank allocation to alignment training should be avoided.