{"total":14,"items":[{"citing_arxiv_id":"2606.18089","ref_index":138,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoning","primary_cat":"cs.LG","submitted_at":"2026-06-16T15:55:28+00:00","verdict":null,"verdict_confidence":null,"novelty_score":null,"formal_verification":null,"one_line_summary":null,"context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.11726","ref_index":10,"ref_count":2,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Block-R1: Rethinking the Role of Block Size in Multi-domain Reinforcement Learning for Diffusion Large Language Models","primary_cat":"cs.LG","submitted_at":"2026-05-12T08:09:42+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Introduces Block-R1 benchmark, Block-R1-41K dataset, and a conflict score to handle domain-specific optimal block sizes in RL post-training of diffusion LLMs.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"perform worse than the base model. The underlying cause lies in the heterogeneity of dLLM parallel generation structures across domains, since distinct tasks inherently reside on different best-improved parallel decoding granularities. For example, mathematical reasoning tasks such as GSM8K may benefit from smaller blocks for fine-grained intermediate verification [10, 14], whereas puzzle-solving tasks, such as Sudoku, may favour larger blocks to preserve global format consistency across rows, columns, and subgrids [1]. When a one-for-all block size is imposed during multi-domain RL, all domains generate rollouts under the same block-based decoding structure for policy optimisation, even though their best-improved decoding granularities may differ."},{"citing_arxiv_id":"2605.02263","ref_index":1,"ref_count":2,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Break the Block: Dynamic-size Reasoning Blocks for Diffusion Large Language Models via Monotonic Entropy Descent with Reinforcement Learning","primary_cat":"cs.LG","submitted_at":"2026-05-04T06:17:49+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"b1 is a plug-and-play post-training framework that trains diffusion LLMs to produce dynamic-size reasoning blocks by optimizing a monotonic entropy descent objective via reinforcement learning.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.00610","ref_index":26,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Decouple before Integration: Test-time Synthesis of SFT and RLVR Task Vectors","primary_cat":"cs.LG","submitted_at":"2026-05-01T12:20:44+00:00","verdict":"CONDITIONAL","verdict_confidence":"MODERATE","novelty_score":6.0,"formal_verification":"none","one_line_summary":"DoTS decouples SFT and RLVR training then synthesizes their task vectors at inference time to match integrated training results at ~3% compute cost.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.11206","ref_index":13,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Compatibility-Aware Dynamic Fine-Tuning for Large Language Models","primary_cat":"cs.CL","submitted_at":"2026-04-22T14:47:30+00:00","verdict":"CONDITIONAL","verdict_confidence":"MODERATE","novelty_score":4.0,"formal_verification":"none","one_line_summary":"CADFT improves supervised fine-tuning of large language models by dynamically down-weighting training samples whose low model-likelihood indicates high gradient variance, yielding better stability and generalization.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.04066","ref_index":37,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Adapt to Thrive! Adaptive Power-Mean Policy Optimization for Improved LLM Reasoning","primary_cat":"cs.CL","submitted_at":"2026-04-11T07:34:59+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"APMPO boosts average Pass@1 scores on math reasoning benchmarks by 3 points over GRPO by using an adaptive power-mean policy objective and feedback-driven clipping bounds in RLVR training.","context_count":1,"top_context_role":"method","top_context_polarity":"use_method","context_text":"(II) Estimation Error (36) Part I: Bounding the Policy Interpolation Er- ror.The difference between the true objective and the theoretical surrogate is bounded by the quadratic variation of the policy (Schulman et al., 2015). Utilizing Pinsker's inequality to relate Total Variation divergence to KL divergence, we have: |J(π)−L true(π)| ≤C·max s DKL(πold(·|s)∥π(·|s)) (37) Setting β=C , this yields the first term β· DKL(π∥π old). (A detailed derivation of this bound is provided in the subsequent subsection). Part II: Bounding the Estimation Error.The empirical surrogate ˆL deviates from Ltrue due to advantage estimation noise. Let ˆA(s, a) = Aπold(s, a)+ξ(s, a), where ξ is noise with variance σ2 R. |Ltrue(π)− ˆL(π)|="},{"citing_arxiv_id":"2605.04065","ref_index":37,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Free Energy-Driven Reinforcement Learning with Adaptive Advantage Shaping for Unsupervised Reasoning in LLMs","primary_cat":"cs.CL","submitted_at":"2026-04-11T07:26:04+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"FREIA applies free energy principles and adaptive advantage shaping to unsupervised RL, outperforming baselines by 0.5-3.5 Pass@1 points on math reasoning with a 1.5B model.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.06916","ref_index":45,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"FP4 Explore, BF16 Train: Diffusion Reinforcement Learning via Efficient Rollout Scaling","primary_cat":"cs.LG","submitted_at":"2026-04-08T10:14:47+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Sol-RL decouples FP4-based candidate exploration from BF16 policy optimization in diffusion RL, delivering up to 4.64x faster convergence with maintained or superior alignment performance on models like FLUX.1 and SD3.5.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"line, adapting DPO-style learning to diffusion model post-training without explicit rollouts. FMPG [44] and especially AWM [9] place this line on firmer policy-optimization footing by using the ELBO as a proxy for policy likelihood. This connection makes forward-process optimization a particularly compelling direction. DiffusionNFT [ 8] can be interpreted as an NFT-style [45] forward-process version of GRPO. Other works also explore forward-process variants for diffusion post-training [46, 47], while Choi et al. [48] provide a discussion of forward-based diffusion RL. 9 FP4 Explore, BF16 Train: Diffusion Reinforcement Learning via Efficient Rollout Scaling 5.2. Efficient Inference with Low-bit Quantization Model quantization has become a mainstream technique for deploying large foundation models."},{"citing_arxiv_id":"2604.03688","ref_index":38,"ref_count":1,"confidence":0.92,"is_internal_anchor":false,"paper_title":"Fusion and Alignment Enhancement with Large Language Models for Tail-item Sequential Recommendation","primary_cat":"cs.IR","submitted_at":"2026-04-04T11:19:08+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"FAERec fuses collaborative ID embeddings with LLM semantic embeddings using adaptive gating and dual-level alignment to enhance tail-item sequential recommendations.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Figure 1: Visualization of the long-tail item problem (a) and comparison between the existing LLM-based sequential rec- ommendation methods and our FAERec (b, c). initialization[ 18, 46, 53] utilizes language embeddings to pro- vide effective initialization for the ID embedding layer, gradually evolving from semantic space to collaborative space. 2)Knowledge transfer[ 38, 47] maximizes mutual information between ID and LLM embeddings through cross-view contrastive learning. Despite their effectiveness, these methods face two critical chal- lenges: (i)Insufficient Semantic Leveraging.As shown in Figure 1 (b), semantic initialization and knowledge transfer strategies use LLM embeddings solely as auxiliary signals for learning ID embed-"},{"citing_arxiv_id":"2509.21882","ref_index":4,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Position: The Hidden Costs and Measurement Gaps of Reinforcement Learning with Verifiable Rewards","primary_cat":"cs.LG","submitted_at":"2025-09-26T05:06:25+00:00","verdict":null,"verdict_confidence":null,"novelty_score":null,"formal_verification":null,"one_line_summary":null,"context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2509.16117","ref_index":3,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"DiffusionNFT: Online Diffusion Reinforcement with Forward Process","primary_cat":"cs.LG","submitted_at":"2025-09-19T16:09:33+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"DiffusionNFT performs online RL for diffusion models on the forward process via flow matching and positive-negative contrasts, delivering up to 25x efficiency gains and rapid benchmark improvements over prior reverse-process methods.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Focusing solely on the reverse sampling process breaks adherence to the forward diffusion process, risking the model degenerating into cascaded Gaussians. (2) Solver restriction. The data collection process relies on first-order SDE samplers, precluding the full utilization of ODE or high-order solvers that are default to flow models and ad- vantageous for generation efficiency. (3) Complicated CFG integration. Diffusion models heavily rely on Classifier-Free Guidance (CFG) (Ho & Salimans, 2022), which requires training both con- ditional and unconditional models. Current RL practices typically incorporate CFG in post-training, leading to a complicated and inefficient two-model optimization scheme. We aim to disentangle data collection, remove solver restriction, and maintain consistency with stan-"},{"citing_arxiv_id":"2509.08827","ref_index":56,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"A Survey of Reinforcement Learning for Large Reasoning Models","primary_cat":"cs.CL","submitted_at":"2025-09-10T17:59:43+00:00","verdict":"ACCEPT","verdict_confidence":"LOW","novelty_score":3.0,"formal_verification":"none","one_line_summary":"A survey compiling RL methods, challenges, data resources, and applications for enhancing reasoning in large language models and large reasoning models since DeepSeek-R1.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2509.03403","ref_index":3,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Beyond Correctness: Harmonizing Process and Outcome Rewards through RL Training","primary_cat":"cs.LG","submitted_at":"2025-09-03T15:28:51+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"PROF curates RL training data via PRM-ORM consistency to improve both final-answer accuracy and intermediate reasoning quality while reducing reliance on strong process reward models.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2504.14945","ref_index":54,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Learning to Reason under Off-Policy Guidance","primary_cat":"cs.LG","submitted_at":"2025-04-21T08:09:13+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"LUFFY mixes off-policy reasoning traces into RLVR training via Mixed-Policy GRPO and regularized importance sampling, delivering over 6-point gains on math benchmarks and enabling training of weak models where on-policy RLVR fails.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null}],"limit":50,"offset":0}