{"total":32,"items":[{"citing_arxiv_id":"2606.22716","ref_index":38,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Beyond Penalizing Mistakes: Stabilizing Efficiency Training in Large Reasoning Models via Adaptive Correct-Only Rewards","primary_cat":"cs.AI","submitted_at":"2026-06-21T23:27:12+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"ACOER applies adaptive correct-only efficiency rewards in GRPO to avoid reward collapse, yielding higher accuracy and over 60% fewer tokens on math reasoning benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.17890","ref_index":16,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Dynamic Rollout Editing for Reducing Overthinking in RL-Trained Reasoning Models","primary_cat":"cs.CL","submitted_at":"2026-06-16T13:10:30+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Dynamic Rollout Editing reduces overthinking in RL-trained LLMs by editing post-answer continuations in successful rollouts and preferring the edited versions within GRPO groups.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.08684","ref_index":5,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"BLUE: Toward Better Language Use in Efficient Vision-Language-Action Models for Autonomous Driving","primary_cat":"cs.CV","submitted_at":"2026-06-07T15:37:16+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"BLUE trains a lightweight gate on frozen VLA hidden states to selectively activate language generation only when beneficial, achieving SOTA results with 2.54x inference speedup on driving benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.05988","ref_index":23,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Compress-Distill: Reasoning Trace Compression for Efficient Knowledge Distillation","primary_cat":"cs.LG","submitted_at":"2026-06-04T10:30:58+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Post-hoc model-based compression of reasoning traces cuts training tokens to 12-30% and speeds training 2-7.6x while retaining up to 96% of raw-trace accuracy, though raw traces remain superior at every scale.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.04560","ref_index":41,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Rollout-Level Advantage-Prioritized Experience Replay for GRPO","primary_cat":"cs.LG","submitted_at":"2026-06-03T07:47:47+00:00","verdict":"CONDITIONAL","verdict_confidence":"MODERATE","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Rollout-level advantage-prioritized experience replay for GRPO recycles high-advantage individual rollouts with age eviction and fresh-anchored batches to outperform standard GRPO on math benchmarks, with gains increasing with model size.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.03503","ref_index":17,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"ThoughtFold: Folding Reasoning Chains via Introspective Preference Learning","primary_cat":"cs.AI","submitted_at":"2026-06-02T11:21:27+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"ThoughtFold applies introspective redundancy detection within correct CoT trajectories to create sub-trajectory spectra, then uses masked preference optimization to penalize redundant explorations, yielding 56% token reduction on DeepSeek-R1-Distill-Qwen-7B while preserving accuracy.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.02871","ref_index":11,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Adaptive Latent Agentic Reasoning","primary_cat":"cs.CL","submitted_at":"2026-06-01T20:36:06+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"ALAR trains LLM agents to perform most reasoning in a latent space supervised by actions and escalates to explicit CoT only when needed, cutting tokens by up to 84.6% while preserving accuracy on search and tool-use benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.01934","ref_index":3,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"HMPO: Hybrid Median-length Policy Optimization for Chain-of-Thought Compression","primary_cat":"cs.LG","submitted_at":"2026-06-01T09:01:55+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"HMPO is a single-stage RL framework for CoT compression that reports 19-46% token reduction with negligible accuracy loss on models from 9B to 122B parameters across math, code, science, and instruction tasks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.01532","ref_index":86,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Rethinking the Role of Positional Encoding: Sliding-Window Transformers without PE Remain Turing Complete","primary_cat":"cs.LG","submitted_at":"2026-06-01T01:28:42+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":8.0,"formal_verification":"none","one_line_summary":"Sliding-window transformers without positional encodings are Turing complete because the sliding window breaks permutation symmetry and suffices to simulate Post machines via a constant-size histogram state.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.01249","ref_index":246,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Trust Region On-Policy Distillation","primary_cat":"cs.LG","submitted_at":"2026-05-31T14:04:51+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"TrOPD stabilizes on-policy distillation for LLMs with trust-region learning, outlier estimation, and off-policy guidance, outperforming prior OPD methods on reasoning and code benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.01168","ref_index":10,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Thinking Economically: A Hierarchical Framework for Adaptive-Complexity Reasoning in LLMs","primary_cat":"cs.CL","submitted_at":"2026-05-31T11:20:00+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"HAB applies coarse-to-fine budgeting to LLM reasoning, predicting per-problem depth and learning intra-step token budgets via PPL comparisons and adaptive Pareto optimization, yielding higher accuracy and lower token use than standard CoT on GSM8K and MATH500.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.26781","ref_index":39,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"LiveK12Bench: Have Large Multimodal Models Truly Conquered High School-level Examinations?","primary_cat":"cs.AI","submitted_at":"2026-05-26T09:50:35+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"LiveK12Bench is a growing multi-disciplinary benchmark showing LMMs like GPT-5 drop from 79 to 53 under realistic exam constraints including process rigor and efficiency.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.25745","ref_index":11,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Selective Latent Thinking: Adaptive Compression of LLM Reasoning Chains","primary_cat":"cs.CL","submitted_at":"2026-05-25T11:57:09+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"SLT selectively compresses reasoning spans via anticipation and gating, trained in three stages including RL, yielding 22.7% higher accuracy than uniform latent baselines at similar compression and 58.4% shorter chains with 2.8% accuracy drop vs explicit CoT on math benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.25604","ref_index":16,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"DVAO: Dynamic Variance-adaptive Advantage Optimization for Multi-reward Reinforcement Learning","primary_cat":"cs.CL","submitted_at":"2026-05-25T08:55:16+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"DVAO dynamically weights multi-objective advantages by rollout-group reward variance to bound magnitudes, add cross-objective regularization, and outperform static baselines on math and tool-use tasks with Qwen models.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.22211","ref_index":33,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"CLORE: Content-Level Optimization for Reasoning Efficiency","primary_cat":"cs.AI","submitted_at":"2026-05-21T09:16:27+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"CLORE augments correct on-policy rollouts by deleting repetitive and irrelevant segments then optimizes with auxiliary DPO to improve accuracy-efficiency trade-off on math benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.13165","ref_index":64,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"STOP: Structured On-Policy Pruning of Long-Form Reasoning in Low-Data Regimes","primary_cat":"cs.CL","submitted_at":"2026-05-13T08:28:05+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"STOP uses structured on-policy analysis to prune long reasoning traces to their earliest correct node, cutting token usage 19-42% with little accuracy loss on math benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.11625","ref_index":28,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Nice Fold or Hero Call: Learning Budget-Efficient Thinking for Adaptive Reasoning","primary_cat":"cs.AI","submitted_at":"2026-05-12T06:51:31+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"BET reduces reasoning tokens by about 55% on average while improving performance across benchmarks by learning to short-solve easy queries, fold early on unsolvable ones, and preserve budget for hard solvable queries.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"CoT-Valve: Length-Compressible Chain-of-Thought Tuning. InAnnual Meeting of the Association for Computational Linguistics, 2025. URL https://api.semanticscholar.org/CorpusID: 276317564. [27] math-ai. AMC-23. https://huggingface.co/datasets/math-ai/amc23, 2025. Hugging Face dataset. 40 problems from the 2023 AMC 12A/12B benchmark; accessed 2026-04-16. [28] Xiaoye Qu, Yafu Li, Zhaoyu Su, Weigao Sun, Jianhao Yan, Dongrui Liu, Ganqu Cui, Daizong Liu, Shuxian Liang, Junxian He, Peng Li, Wei Wei, Jing Shao, Chaochao Lu, Yue Zhang, Xian-Sheng Hua, Bowen Zhou, and Yu Cheng. A Survey of Efficient Reasoning for Large Reasoning Models: Language, Multimodality, and Beyond.ArXiv, abs/2503.21614, 2025. URL https://api."},{"citing_arxiv_id":"2605.09806","ref_index":15,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"LEAD: Length-Efficient Adaptive and Dynamic Reasoning for Large Language Models","primary_cat":"cs.LG","submitted_at":"2026-05-10T23:05:02+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"LEAD uses online adaptive mechanisms including Potential-Scaled Instability and symmetric efficiency rewards based on correct rollouts to achieve higher accuracy-efficiency scores with substantially shorter reasoning outputs than base models on math benchmarks.","context_count":1,"top_context_role":"baseline","top_context_polarity":"baseline","context_text":"A growing body of work addresses the verbosity problem in reasoning mod- els, namely the tendency to generate unnecessarily long solutions when optimized primarily for correctness. Several methods introduce length penalties, pruning objectives, or budget constraints during training. L1 [ 14] trains reasoning models to follow user-specified length constraints, O1- Pruner [15] uses length-harmonizing fine-tuning to reduce redundant long-thought reasoning, and DRPO [18] decouples the learning signals for correct and incorrect rollouts to avoid penalizing valid long reasoning. LASER [19] formulates efficient reasoning through adaptive length-based reward shaping, while GFPO [ 21] encourages concise reasoning by filtering sampled rollouts according"},{"citing_arxiv_id":"2605.08665","ref_index":15,"ref_count":2,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Hint Tuning: Less Data Makes Better Reasoners","primary_cat":"cs.CL","submitted_at":"2026-05-09T04:07:16+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Hint Tuning reduces token usage 24-66% (31.5% avg) in reasoning models via 1K self-annotated samples aligned to an instruct model's capabilities while keeping benchmark accuracy.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"(c-d) With partial hints, instruct models need only 27.5% of episodes and 11.9% of tokens to succeed. problem and redundant in another.During-reasoningmethods [ 12, 13, 14] generate concise reasoning via token budgets or step-skipping, though uniform strategies fail on problems of varying complexity. Both families avoid the cost ofRL with length penalties[ 15, 16, 17, 18], which requires careful tuning of penalty coefficients and reward scaling, incurring significantly higher compute than SFT [8, 19]; moreover, fixed penalties cause over-compression on easy problems and under-compression on hard ones. The instruct model as a capability probe.Our key insight is that the corresponding instruct model serves as an ideal capability probe."},{"citing_arxiv_id":"2605.06165","ref_index":70,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Post Reasoning: Improving the Performance of Non-Thinking Models at No Cost","primary_cat":"cs.AI","submitted_at":"2026-05-07T12:51:49+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Post-Reasoning boosts LLM accuracy by reversing the usual answer-after-reasoning order, delivering mean relative gains of 17.37% across 117 model-benchmark pairs with zero extra cost.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.18839","ref_index":50,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"One Step Forward and K Steps Back: Better Reasoning with Denoising Recursion Models","primary_cat":"cs.LG","submitted_at":"2026-04-20T21:06:12+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Denoising Recursion Models train multi-step noise reversal in looped transformers and outperform the prior Tiny Recursion Model on ARC-AGI.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.14847","ref_index":10,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"TrigReason: Trigger-Based Collaboration between Small and Large Reasoning Models","primary_cat":"cs.AI","submitted_at":"2026-04-16T10:33:00+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"TrigReason matches large reasoning model accuracy on math and science benchmarks by delegating most steps to small models and intervening selectively on three triggers, cutting latency by 43.9% and cost by 73.3%.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.05355","ref_index":3,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"ETR: Entropy Trend Reward for Efficient Chain-of-Thought Reasoning","primary_cat":"cs.AI","submitted_at":"2026-04-07T02:53:36+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"ETR is a trajectory-aware reward that promotes progressive entropy reduction during CoT reasoning, integrated into GRPO to deliver higher accuracy and 67% shorter traces on tested models and benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.04120","ref_index":10,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Shorter, but Still Trustworthy? An Empirical Study of Chain-of-Thought Compression","primary_cat":"cs.CL","submitted_at":"2026-04-05T13:43:12+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"CoT compression frequently introduces trustworthiness regressions with method-specific degradation profiles; a proposed normalized efficiency score and alignment-aware DPO variant reduce length by 19.3% with smaller trustworthiness loss.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.02967","ref_index":8,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"FoE: Forest of Errors Makes the First Solution the Best in Large Reasoning Models","primary_cat":"cs.AI","submitted_at":"2026-04-03T11:03:02+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Errors in large reasoning models form a forest structure that grows with more steps, making the first solution best; RED refines the first and prunes the rest for higher performance with less compute.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"comes the parent. Accordingly, the set of valid candidate parents Pj for nodee j is defined as: Pj ≜{i∈[1, j−1] :s ij ≥τ}.(7) To ensure a forest structure where each node has at most one parent, we select the temporally closest candidate (i.e., the largest index) from Pj. The parent indexπ(j)is formally determined by: π(j)≜ ( maxP j,P j ̸=∅, 0,P j =∅. (8) Here, π(j) = 0 indicates that ej has no antecedent satisfying the threshold condition. In such cases, ej is treated as a root node, instantiating a newTree of Errors(ToE) within theForest of Errors( FoE). Otherwise, a directed edge eπ(j) →e j is added to the ToE containing eπ(j). By iterating j from 1 to n, this procedure constructs the complete forest, as"},{"citing_arxiv_id":"2601.05242","ref_index":41,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization","primary_cat":"cs.CL","submitted_at":"2026-01-08T18:59:24+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"GDPO decouples per-reward normalization in multi-reward RL to avoid advantage collapse and improve convergence over GRPO on tool-calling, math, and coding tasks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2601.03066","ref_index":3,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Do LLMs Encode Functional Importance of Reasoning Tokens?","primary_cat":"cs.CL","submitted_at":"2026-01-06T14:50:02+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"LLMs encode functional importance over reasoning tokens, demonstrated by greedy pruning that yields shorter effective chains and attention scores that predict pruning order.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2512.13564","ref_index":152,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Memory in the Age of AI Agents","primary_cat":"cs.CL","submitted_at":"2025-12-15T17:22:34+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"The paper maps agent memory research via three forms (token-level, parametric, latent), three functions (factual, experiential, working), and dynamics of formation/evolution/retrieval, plus benchmarks and future directions.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2509.14004","ref_index":9,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Early Stopping Chain-of-thoughts in Large Language Models","primary_cat":"cs.CL","submitted_at":"2025-09-17T14:14:05+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"ES-CoT shortens LLM chain-of-thought generation by tracking runs of identical step answers after linguistic markers, cutting tokens 16% on average while keeping accuracy comparable to full CoT across six datasets and three models.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2509.05489","ref_index":28,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Self-Aligned Reward: Towards Effective and Efficient Reasoners","primary_cat":"cs.LG","submitted_at":"2025-09-05T20:39:43+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Self-aligned reward uses relative perplexity differences to encourage concise, query-specific reasoning in LLMs, yielding 4% accuracy gains and 30% lower inference cost when added to PPO or GRPO.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2506.12876","ref_index":17,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"MaskPro: Linear-Space Probabilistic Learning for Strict (N:M)-Sparsity on LLMs","primary_cat":"cs.LG","submitted_at":"2025-06-15T15:02:59+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"MaskPro learns categorical distributions over groups of M weights to generate exact (N:M) sparsity via N-way sampling without replacement and stabilizes training with a moving average tracker of loss residuals.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2503.16419","ref_index":127,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models","primary_cat":"cs.CL","submitted_at":"2025-03-20T17:59:38+00:00","verdict":"ACCEPT","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"A survey organizing techniques to achieve efficient reasoning in LLMs by shortening chain-of-thought outputs.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"prompting [9] extends this concept by structuring thoughts into a graph, allowing iterative refinement of individual reasoning steps. While many CoT variants exist, they generally involve different prompting techniques to guide the behavior of models, sometimes incorporating controller-like mechanisms to manage thought progression and usage. 3 Taxonomy Model-basedEfficient Reasoning RL Optimizationvia Length Reward e.g. Kimi k1.5 [171]; O1-Pruner [127]; L1 [2]; Training [5]; Demystifying [217]; DAST [154];MRT [146]; Self-adaptive [212]; HA WKEYE [152]; ThinkPrune [66]; LongShort [134];ConciseRL [40]; Bingo [110]; Concise Reasoning [47]; Elastic Reasoning [207]; S-GRPO [33];TLDR [239]; SelfBudgeter [95]; Short-RL [225]; BRPO [144]; LASER [116]; ACPO [21];LIMOPro [198]; L-GRPO [160]; GRPO-λ[32]; AutoThink [175]; AdaptThink [230];DeGRPO [45]; HGPO [74]; DTO [3]; REO-RL [52]; ALP [196]; PLP [107]; LC-R1 [22];AdapThink [179]; AALC [90]; DuP-PO [36]; SCPO [63]; FCS [65]; CurriculumGRPO [57];GFPO [157]; SABER [242]; VSRM [228]; DR."}],"limit":50,"offset":0}