PG-OT builds prompt-specific Pareto frontiers and applies distribution-aware optimal transport to improve multi-reward alignment while introducing JDR and JCR metrics to measure synergy and hacking.
Title resolution pending
23 Pith papers cite this work. Polarity classification is still indexing.
23
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 23roles
background 1polarities
background 1representative citing papers
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.
Discrete Tilt Matching recasts dLLM fine-tuning as state-level matching of tilted local unmasking posteriors, producing a stable weighted cross-entropy loss that improves Sudoku and Countdown performance when applied to LLaDA-8B-Instruct.
Dialectal robustness and generation are dissociated in LLMs: benchmarks are driven by pretraining and SFT while alignment reshapes generation invisibly to benchmarks, and the method maximizing dialectal reward is least preferred by human evaluators.
LLM embeddings condition a generative transformer to enable faster convergence, better performance, and generalization to unseen LHC processes using a single model.
Contrastive Logit Steering isolates a linear refusal direction in safety-aligned LLMs, achieving higher jailbreak success than activation steering and enabling bidirectional control without retraining.
AutoVecCoder combines VecPrompt for automated intrinsic knowledge synthesis and VecRL for efficiency-aligned RL to train an 8B LLM that achieves SOTA on SimdBench SSE/AVX subsets and sometimes exceeds -O3 compiler results.
GAGPO computes step-aligned temporal advantages from grouped rollout samples without a learned critic, enabling stable policy optimization in multi-turn agent environments.
ODRPO decomposes discrete rewards into ordinal binary indicators to create robust, variance-aware advantage estimators for noisy RLAIF in LLM alignment.
LLMs contain identifiable COCO neurons that enable implicit self-correction against stereotypes; targeted editing of these neurons improves fairness and robustness to jailbreaks while preserving generation quality.
MARLaaS enables concurrent RL fine-tuning across up to 32 tasks using LoRA adapters and a disaggregated asynchronous architecture, matching single-task performance while improving accelerator utilization by 4.3x and cutting end-to-end time by 85%.
SHRED performs retain-set-free unlearning by selecting lowest-probability tokens as forget positions and applying a single KL self-distillation objective that demotes logits only at those positions.
Listwise Policy Optimization explicitly performs target-projection on the LLM response simplex, unifying and improving group-based RLVR methods with monotonic improvement and flexible divergences.
ZAYA1-8B is a reasoning MoE model with 700M active parameters that matches larger models on math and coding benchmarks and reaches 91.9% on AIME'25 via Markovian RSA test-time compute.
MAGNET multi-agent generation with persona grounding and ATLAS graph verification yields 34-50% fewer hallucinations and annotations than single-model or IBSEN baselines at 100-page scale.
Large reasoning models show a negative correlation between token entropy and logit-gradient norms, and the paper uses this 'Entropy-Gradient Inversion' as a reward regularizer (CorR-PO) in RL training.
CoMole combines motif-aware graph diffusion with RL policy optimization to deliver controllable molecular generation that outperforms baselines on nine targets across materials and drug benchmarks while keeping high validity.
On-policy distillation gains efficiency from early foresight in module allocation and update directions, which the proposed EffOPD method exploits for 3x faster training with comparable performance.
300 high-quality Stoic examples align small LLMs with inward virtues via preference optimization but leave outward cosmopolitan duties unlearned.
English-only safety alignment fails to transfer cross-lingually, while multilingual DPO training on the new RefusEU dataset improves safety across 12 European languages without degrading Global MMLU performance.
TabGRAA applies group-relative advantage alignment in an iterative reward-guided post-training loop to improve tabular language model generators on fidelity, utility, and privacy trade-offs across five benchmarks.
GradShield is a data filtering technique using FIHS scores and adaptive thresholding to prevent safety misalignment in finetuned LLMs while preserving utility.
LuckyStar 111B adapts Cohere's Command A model with four scaling techniques to improve tool-use, math reasoning, and NL2SQL in Korean-English while preserving general instruction following.
citing papers explorer
-
Pareto-Guided Optimal Transport for Multi-Reward Alignment
PG-OT builds prompt-specific Pareto frontiers and applies distribution-aware optimal transport to improve multi-reward alignment while introducing JDR and JCR metrics to measure synergy and hacking.
-
Post Reasoning: Improving the Performance of Non-Thinking Models at No Cost
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.
-
Discrete Tilt Matching
Discrete Tilt Matching recasts dLLM fine-tuning as state-level matching of tilted local unmasking posteriors, producing a stable weighted cross-entropy loss that improves Sudoku and Countdown performance when applied to LLaDA-8B-Instruct.
-
DiaLLM: An Investigation into the Robustness-Generation Gap in English Dialect Adaptation
Dialectal robustness and generation are dissociated in LLMs: benchmarks are driven by pretraining and SFT while alignment reshapes generation invisibly to benchmarks, and the method maximizing dialectal reward is least preferred by human evaluators.
-
One Generator, Any Process: LLM-Conditioning for the LHC
LLM embeddings condition a generative transformer to enable faster convergence, better performance, and generalization to unseen LHC processes using a single model.
-
The Geometry of Refusal: Linear Instability in Safety-Aligned LLMs
Contrastive Logit Steering isolates a linear refusal direction in safety-aligned LLMs, achieving higher jailbreak success than activation steering and enabling bidirectional control without retraining.
-
AutoVecCoder: Teaching LLMs to Generate Explicitly Vectorized Code
AutoVecCoder combines VecPrompt for automated intrinsic knowledge synthesis and VecRL for efficiency-aligned RL to train an 8B LLM that achieves SOTA on SimdBench SSE/AVX subsets and sometimes exceeds -O3 compiler results.
-
GAGPO: Generalized Advantage Grouped Policy Optimization
GAGPO computes step-aligned temporal advantages from grouped rollout samples without a learned critic, enabling stable policy optimization in multi-turn agent environments.
-
ODRPO: Ordinal Decompositions of Discrete Rewards for Robust Policy Optimization
ODRPO decomposes discrete rewards into ordinal binary indicators to create robust, variance-aware advantage estimators for noisy RLAIF in LLM alignment.
-
Modeling Implicit Conflict Monitoring Mechanisms against Stereotypes in LLMs
LLMs contain identifiable COCO neurons that enable implicit self-correction against stereotypes; targeted editing of these neurons improves fairness and robustness to jailbreaks while preserving generation quality.
-
MARLaaS: Multi-Tenant Asynchronous Reinforcement Learning as a Service
MARLaaS enables concurrent RL fine-tuning across up to 32 tasks using LoRA adapters and a disaggregated asynchronous architecture, matching single-task performance while improving accelerator utilization by 4.3x and cutting end-to-end time by 85%.
-
SHRED: Retain-Set-Free Unlearning via Self-Distillation with Logit Demotion
SHRED performs retain-set-free unlearning by selecting lowest-probability tokens as forget positions and applying a single KL self-distillation objective that demotes logits only at those positions.
-
Listwise Policy Optimization: Group-based RLVR as Target-Projection on the LLM Response Simplex
Listwise Policy Optimization explicitly performs target-projection on the LLM response simplex, unifying and improving group-based RLVR methods with monotonic improvement and flexible divergences.
-
ZAYA1-8B Technical Report
ZAYA1-8B is a reasoning MoE model with 700M active parameters that matches larger models on math and coding benchmarks and reaches 91.9% on AIME'25 via Markovian RSA test-time compute.
-
From Personas to Plot: Character-Grounded Multi-Agent Story Generation for Long-Form Narratives
MAGNET multi-agent generation with persona grounding and ATLAS graph verification yields 34-50% fewer hallucinations and annotations than single-model or IBSEN baselines at 100-page scale.
-
Entropy-Gradient Inversion: Moving Toward Internal Mechanism of Large Reasoning Models
Large reasoning models show a negative correlation between token entropy and logit-gradient norms, and the paper uses this 'Entropy-Gradient Inversion' as a reward regularizer (CorR-PO) in RL training.
-
Controllable Molecular Generative Foundation Models
CoMole combines motif-aware graph diffusion with RL policy optimization to deliver controllable molecular generation that outperforms baselines on nine targets across materials and drug benchmarks while keeping high validity.
-
Learning to Foresee: Unveiling the Unlocking Efficiency of On-Policy Distillation
On-policy distillation gains efficiency from early foresight in module allocation and update directions, which the proposed EffOPD method exploits for 3x faster training with comparable performance.
-
StoicLLM: Preference Optimization for Philosophical Alignment in Small Language Models
300 high-quality Stoic examples align small LLMs with inward virtues via preference optimization but leave outward cosmopolitan duties unlearned.
-
Multilingual Refusal Alignment for Safer Large Language Models
English-only safety alignment fails to transfer cross-lingually, while multilingual DPO training on the new RefusEU dataset improves safety across 12 European languages without degrading Global MMLU performance.
-
Self-Improving Tabular Language Models via Iterative Reward-Guided Post-Training
TabGRAA applies group-relative advantage alignment in an iterative reward-guided post-training loop to improve tabular language model generators on fidelity, utility, and privacy trade-offs across five benchmarks.
-
GradShield: Alignment Preserving Finetuning
GradShield is a data filtering technique using FIHS scores and adaptive thresholding to prevent safety misalignment in finetuned LLMs while preserving utility.
-
Think in English, Answer in Korean: Efficient Adaptation of Multilingual Tool-Using Agents
LuckyStar 111B adapts Cohere's Command A model with four scaling techniques to improve tool-use, math reasoning, and NL2SQL in Korean-English while preserving general instruction following.