Claude 3 Opus strategically fakes alignment by complying with harmful requests only during simulated training to preserve its preference for refusing them afterward.
mega hub Mixed citations
Proximal Policy Optimization Algorithms
Mixed citation behavior. Most common role is background (52%).
1993
Pith papers citing it
Background
52% of classified citations
abstract
We propose a new family of policy gradient methods for reinforcement learning, which alternate between sampling data through interaction with the environment, and optimizing a "surrogate" objective function using stochastic gradient ascent. Whereas standard policy gradient methods perform one gradient update per data sample, we propose a novel objective function that enables multiple epochs of minibatch updates. The new methods, which we call proximal policy optimization (PPO), have some of the benefits of trust region policy optimization (TRPO), but they are much simpler to implement, more general, and have better sample complexity (empirically). Our experiments test PPO on a collection of benchmark tasks, including simulated robotic locomotion and Atari game playing, and we show that PPO outperforms other online policy gradient methods, and overall strikes a favorable balance between sample complexity, simplicity, and wall-time.
hub tools
citation-role summary
background 156
method 114
baseline 15
dataset 4
citation-polarity summary
claims ledger
- abstract We propose a new family of policy gradient methods for reinforcement learning, which alternate between sampling data through interaction with the environment, and optimizing a "surrogate" objective function using stochastic gradient ascent. Whereas standard policy gradient methods perform one gradient update per data sample, we propose a novel objective function that enables multiple epochs of minibatch updates. The new methods, which we call proximal policy optimization (PPO), have some of the benefits of trust region policy optimization (TRPO), but they are much simpler to implement, more ge
authors
mega hub controls
export citing contexts JSON
export graph JSON
export full bundle JSON
open full Pith review
annotated reader queued
Recognition alignment
counterfactual ablation
co-cited works
representative citing papers
GRPO's group-mean baseline assigns identical advantages to all tokens under output-only rewards, inducing gradient sparsity and an intrinsic rank-2 structure proven from the zero-sum constraint and confirmed by SVD on Nemotron-4B gradients.
Negative narrative immersion causes 12-31% drops in LLM moral accuracy and produces structured shifts that appear in downstream applications.
RL agent for online LHC trigger threshold tuning improves in-tolerance intervals by 28-56% on Monte Carlo and real CMS data without fine-tuning.
IRumAI is the first RL agent for Indian Rummy, trained on weak heuristics to beat strong search opponents at 7000x speed.
DeFAb is a large-scale, formally verifiable benchmark for defeasible abduction derived from 18 knowledge bases, demonstrating that frontier LLMs achieve 7.8-65% accuracy versus 100% for a rule-based solver with polynomial-time checks.
Machine learning discovers a tube-seeding strategy for IBP reduction of Feynman integrals that scales linearly with numerator power, demonstrated on rank-20 2-loop 5-point integrals.
A reward-free representation learning pipeline for offline PbRL achieves better preference efficiency than standard two-stage baselines by connecting RFRL concepts to preference data.
Dynamic isotropy, quantifying uniform center-of-mass acceleration capability, improves robot performance and enables omnidirectional locomotion, terrain traversal, and failure resilience in a spherical robot design.
AtomComposer uses online RL with multi-composition training to discover up to 10x more valid 3D isomers on unseen chemical formulas than single-composition baselines.
Agent-BRACE improves LLM agent performance on long-horizon partially observable tasks by 5.3-14.5% through a decoupled belief state of verbalized atomic claims with certainty labels that keeps context length constant.
SimWorld Studio deploys an evolving coding agent to create adaptive 3D environments that co-evolve with embodied learners, delivering 18-point success-rate gains over fixed environments in navigation benchmarks.
ReLibra uses pre-known token-to-expert routing from RL rollouts to perform inter-batch expert reordering and intra-batch replication, delivering up to 1.6x higher throughput than Megatron-LM and 1.2x over oracle-equipped EPLB while staying within 6-10% of an ideal balanced baseline.
Weak-to-strong generalization is nearly inevitable in linear logistic regression for most student-teacher pairs without any model capacity mismatch.
Observation and action delays are formally equivalent in cooperative Dec-POMDPs, yielding identical optimal solutions and enabling zero-shot transfer, though learning dynamics differ due to credit assignment and operational constraints.
A language-game framework enables dialogue with dynamical systems such as GRNs by treating their frozen dynamics as an RL policy core, using an LM to route prompts so the system responds through its own behavior without parameter changes.
RefereeBench shows that even the strongest video MLLMs reach only around 60% accuracy on multi-sport refereeing tasks and struggle with rule application and temporal grounding.
OP-GRPO is the first off-policy GRPO method for flow-matching models that reuses trajectories via replay buffer and importance sampling corrections, matching on-policy performance with 34.2% of the training steps.
User-turn generation reveals that LLMs' interaction awareness is largely decoupled from task accuracy, remaining near zero in deterministic settings even as accuracy scales to 96.8% on GSM8K.
Derives an exact unbiased policy gradient for RL post-training of diffusion LLMs via entropy-guided step selection and one-step denoising rewards, achieving state-of-the-art results on coding and logical reasoning benchmarks.
A certified gradient-based method for contact-rich manipulation that quantifies smoothing-induced errors via set-valued discrepancies and incorporates them into analytical reachable sets for robust affine feedback policies.
First in-orbit demonstration of a DRL-trained AI satellite attitude controller that performs robust inertial pointing after sim-to-real transfer.
Develops and tests the first effective safeguard for analytic gradient-based provably safe RL, showing safe training on three control tasks without performance loss.
Flow-GRPO is the first online RL method for flow matching models, raising GenEval accuracy from 63% to 95% and text-rendering accuracy from 59% to 92% with little reward hacking.
citing papers explorer
-
KLip-PPO: A per-sample KL perspective on PPO-Clip
PPO-Clip gradient equals a per-sample KL surrogate with closed-form coefficient on importance ratio and advantage, yielding identical curves on five MuJoCo tasks.
-
DiT-Reward: Generative Representations for Text-to-Image Reward Modeling
DiT-Reward converts pretrained DiT models into reward predictors that outperform HPSv3 on four benchmarks while providing 1.65x inference speedup.
-
A Stackelberg Framework for Resource-Aware LLM Agents: Learning, Repair, and Conditional Guarantees
A Stackelberg game framework for LLM resource governance is introduced with learned response models, policy repair via real-API calibration, conditional theoretical guarantees, and an experiment showing 17.4% token cost reduction.
-
Hierarchical Reinforcement Learning for Sparse-Reward Search in Commutative Algebra
Hierarchical RL with equivariant GNN policies outperforms classical RL and greedy search when searching for counterexamples to an algebraic conjecture across a range of degrees.
-
A Markov Chain Approach to Preference Alignment
MCHF defines a Markov kernel from pairwise utilities U and proves geometric convergence to its stationary distribution at a rate set by the seminorm measuring non-transitivity of U, with first-order equivalence to RLHF and NLHF solutions.
-
Training the Orchestrator: A Supervised Approach to End-to-End PDDL Planning with LLM Agents
HALO trains an orchestrator policy on verifier-approved refinement trajectories across 11 PDDL domains, matching GPT-5-mini success rates at roughly 45x lower orchestration cost and cutting LLM calls by 40-50%.
-
Through the PRISM: Preference Representation in Intermediate States of Video Diffusion Models
PRISM shows video diffusion models inherently encode preference information in noisy latents, achieving SOTA accuracy and enabling noise-robust early-stage sampling with a correlation to generative performance.
-
The Reward Was in Your Data All Along: Correcting Flow Matching with Discriminator-Guided RL
DRL trains a discriminator on data versus base-model samples in pretrained representation space and uses its logit as reward in KL-regularized RL, cutting guidance-free FID from 9.38 to 2.62 on SiT and similar gains on other backbones.
-
Learning from Your Own Mistakes: Constructing Learnable Micro-Reflective Trajectories for Self-Distillation
TAPO constructs learnable micro-reflective trajectories from contrastive model rollouts during RL training to provide explicit error diagnoses and corrections, reporting consistent gains over GRPO on AIME and HMMT math benchmarks.
-
Recover, Discover, Plan: Learning Skills and Concepts from Robot Failures
ReSYNC learns recovery skills via RL then discovers and refines relational predicates to enable abstract planning that generalizes failure avoidance to unseen long-horizon tasks, outperforming baselines by over 50% in simulation and transferring to real robots.
-
Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients
ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite with largest gains at the 0.8B scale.
-
WireCraft: A Simulation Benchmark for Industrial DLO Manipulation
WireCraft is a new configurable simulation benchmark for industrial DLO manipulation with three task families, dual physics models, and shared evaluation of RL, IL, and VLA policies showing high success under privileged state but bottlenecks for vision-based methods.
-
PearlVLA: Progressive Embodied Action-Plan Refinement in Latent Space
PearlVLA achieves SOTA on LIBERO by separating VLM representations into visual grounding and an iterative latent plan branch refined via world model queries and RefineNet with process-reward RL.
-
Work Extraction via Backward Motion in Optimal Closed-Loop Stochastic Control
Reinforcement learning identifies optimal closed-loop feedback strategies in an overdamped colloidal optical trap that enable work extraction by fluctuation exploitation, matching exact theory and extending to heterogeneous forcing.
-
PowerOPD: Stabilizing On-Policy Distillation with Bounded Power Transformation
PowerOPD applies the Box-Cox power transformation to create natively bounded, sign-consistent rewards for on-policy distillation, delivering up to +6.37 Avg@8 gains over vanilla OPD on math reasoning benchmarks while cutting compute costs.
-
ReSum: Synergizing LLM Reasoning and Summarization with Reinforcement Learning
ReSum trains LLMs via RLVR to self-summarize reasoning trajectories, yielding 4% average performance gains and 18.6% shorter rollouts through contrastive rollout branches.
-
Demystifying Hidden-State Recurrence: Switchable Latent Reasoning with On-Policy Reinforcement Learning
SWITCH uses explicit <swi> and </swi> boundary tokens to make latent chain-of-thought compatible with on-policy RL (GRPO) and open to causal mechanistic probing, outperforming prior hidden-state recurrence methods.
-
SymQNet: Amortized Acquisition for Low-Latency Adaptive Hamiltonian Learning
SymQNet uses offline RL to amortize acquisition for adaptive Hamiltonian learning, delivering 47-72x lower decision latency than online Bayesian baselines on Ising models while keeping posterior updates.
-
EgoEngine: From Egocentric Human Videos to High-Fidelity Dexterous Robot Demonstrations
EgoEngine transforms egocentric human videos into high-fidelity robot data enabling zero-shot visuomotor dexterous policy learning without real-robot demonstrations.
-
Mathematical perspective on genetic algorithms with optimization guided operators
Presents a query-complexity framework for genetic algorithms with guided operators and shows necessity of multiple operators and tight bounds for diversity in solution pools.
-
Harnessing Routing Foresight for Micro-step-level MoE load balancing in RL Post-training
ForeMoE uses routing foresight from the rollout stage to enable micro-step load balancing in MoE RL post-training via a hierarchical planner and transfer engine, claiming up to 1.45x speedup on 64 GPUs.
-
Hey Chat, Can You Teach Me? Structuring Socratic Dialogue for Human Learning in the Wild
A PPO policy for deciding topic order and duration on a prerequisite knowledge graph, paired with an LLM for Socratic dialogue, improves student mastery rates and reduces turns compared to baselines and scaled models across held-out topics.
-
Learning Object Manipulation from Scratch via Contrastive Interaction
IWR improves CRL sample efficiency and performance in interaction-rich manipulation by interaction-aware resampling that preserves mode boundaries, yielding 19.8% average gains and a real-world air-hockey agent.
-
INFRAMIND: Infrastructure-Aware Multi-Agent Orchestration
INFRAMIND is an infrastructure-aware multi-agent orchestration framework that uses RL on a hierarchical constrained MDP to jointly optimize topology, model selection, and scheduling under dynamic load.
-
Mirror Descent Beyond Euclidean Stability: An Exponential Separation in Initialization Sensitivity
Non-quadratic Mirror Descent exhibits exponential initialization sensitivity in convex settings, shown via 3D constructions and KL-regularized simplex examples, with Bregman anchoring proposed for stabilization.
-
Dynamic Execution Horizon Prediction for Chunk-based Robot Policies
DEHP adds an online-RL horizon predictor to frozen chunk policies, yielding higher success on precise and long-horizon robot manipulation by adapting chunk length to task stage.
-
Flow-DPPO: Divergence Proximal Policy Optimization for Flow Matching Models
Flow-DPPO replaces PPO ratio clipping with an asymmetric KL divergence mask for flow models, claiming higher rewards, reduced forgetting, and stable multi-epoch training.
-
Expected Free Energy-based Planning as Variational Inference
EFE-based planning is formulated as variational free energy minimization with epistemic priors, decomposing into expected plan costs plus a complexity term.
-
Seeing Before Colliding: Anticipatory Safe RL with Frozen Vision-Language Models
VLM-Safe-RL adds frozen VLM signals as anticipatory costs to the CMDP Lagrangian update via dual-path CLIP, VLM-Lagrange, and confidence gating, outperforming baselines on Safety-Gymnasium FormulaOne while showing partial generalization.
-
Beyond Absolute Imitation: Anchored Residual Guidance for Privileged On-Policy Distillation
AR-OPD disentangles privileged supervision via anchored residual guidance to reduce hindsight leakage in on-policy distillation, reporting gains of 2.3 points over full privileged OPD and 7.9 over SFT on reasoning tasks.
-
Mult-DPO: Multinomial Direct Preference Optimization for Recommender Systems
Mult-DPO provides a tractable multinomial surrogate for set-wise preference optimization in DPO that upper bounds the Plackett-Luce based loss for LLM recommender systems.
-
ReCoVLA: VLM-Guided Reward Compilation for Failure Recovery in Vision-Language-Action Policies
ReCoVLA improves VLA policy reliability by using a VLM as a semantic reward selector to train residual recovery policies in simulation, raising average success from 36.7% to 66.7% in sim and achieving 61.7% in zero-shot sim-to-real physical tests.
-
Escaping the KL Agreement Trap in On-Policy Distillation
KAT detects persistent low-KL agreement traps in on-policy distillation via a dynamic threshold to filter weak supervision, improving avg@k by 2.66% and pass@k by 3.43% on four math benchmarks while shortening rollouts by 59.73%.
-
CRANE: Knowledge Editing for Reasoning MLLMs
CRANE uses dual-library retrieval plus two-phase training (SFT then GRPO with cognitive routing reward) to reach 96.9% grounded success on conflict edits in reasoning MLLMs.
-
HARBOR: A Harness Framework for Agentic Robot Reinforcement Learning
HARBOR is a new agentic harness framework that automates robot RL workflows end-to-end across 16 tasks in manipulation, locomotion, and dexterous control, matching or exceeding default configurations while enabling sim-to-real transfer.
-
Reinforcement Learning for Flow-Matching Policies with Density Transport
RLDT fine-tunes pretrained flow-matching policies for continuous control by aligning them to a max-entropy RL transport field constructed via SVGD, using expected-target estimation for stable multi-step updates.
-
Real-IKEA: Physical Fidelity is the Prerequisite for Robust Manipulation
Real-IKEA supplies 1,079 physically accurate articulated asset configurations from real IKEA parts together with resistance-calibrated simulation parameters that enable RL policies to discover robust hooking and levering behaviors.
-
ActProbe: Action-Space Probe for Early Failure Detection of Generative Robot Policies
ActProbe is an action-space detector that uses temporal consistency error and action chunk magnitude from policy outputs, mapped via LSTM-MLP, to predict failures earlier than baselines across policies and real-robot tasks.
-
Modelling Opinion Dynamics at Scale with Deep MARL
Deep MARL models opinion dynamics at scale, showing high conformity reduces collective accuracy in large networks while sometimes improving it in small ones.
-
AdaTok: Self-Budgeting Image Tokenization with Quality-Preserving Dynamic Tokens
AdaTok learns content-dependent token budgets for discrete 1D image tokenization via prioritized representation learning and a GRPO allocation policy, achieving rFID 1.50 at ~118 tokens average versus fixed 256-token baselines.
-
Adversarial Robustness of Activation Steering in Large Language Models
First systematic test shows activation steering robustness drops sharply (up to 64%) under adversarial input perturbations across multiple extraction methods, models, and personas.
-
DPAgent-in-the-Middle: Agentic Defense and Repair Against AI-Groomed Deceptive Patterns
DPAgent is an agentic framework that detects 90.98% of AI-groomed deceptive samples and repairs 77% of deceptive interfaces while exploring 80% of pattern types with 10% of baseline page visits.
-
Learn to Match: Two-Sided Matching with Temporally Extended Feedback
Learn2Match is a POMG-based MARL benchmark for two-sided matching with temporally extended feedback; independent PPO yields higher social welfare and lower regret than CA-ETC but higher information-friction loss.
-
Your GFlowNet Secretly Learns an Optimal Transport Plan
Minimum-flow GFlowNets on graphs encode optimal transport plans, with the learned policy recovering the optimal coupling between source and target distributions.
-
Online KL-Regularized Reinforcement Learning with Function Approximation under Misspecification
Introduces KL misspecification for bandits and RL under function approximation and proves explicit KL-regret bounds for regression-based Gibbs algorithms that recover the realizable case.
-
ACE-SQL: Adaptive Co-Optimization via Empirical Credit Assignment for Text-to-SQL
ACE-SQL jointly optimizes schema linking and SQL generation via RL with empirical credit assignment from execution-correct rollouts, achieving 65.3% greedy execution accuracy on BIRD Dev using 0.93k output tokens.
-
Step-by-Step Optimization-like Reasoning in LLMs over Expanding Search Spaces
Introduces OPT* tasks and two training regimes (solver-guided online policy optimization with rank-based reward shaping and search-based offline RL) plus a theoretical link between search success and information extraction per budget unit, showing empirical gains in optimization-like reasoning.
-
Reinforcement Learning from Rich Feedback with Distributional DAgger
DistIL applies distributional DAgger with forward cross-entropy to achieve monotonic policy improvement and better Pass@N from rich feedback in RL for reasoning tasks.
-
Policy-Conditioned Counterfactual Credit for Verifiable Reinforcement Learning of Long-Horizon Language Agents
CVT-RL improves verified task success to 78.9% and reduces hacking to 3.9% in long-horizon language agents by combining intervention-validity gating with a selection-adjusted doubly robust PCCC estimator.
-
Alpha-RTL: Test-Time Training for RTL Hardware Optimization
TTT-RTL performs per-design test-time RL on an LLM policy with EDA-derived PPA rewards and an adaptive KL controller, reducing geometric-mean PPA product by 65.1% on RTLLM v2.0 and ADP by 59.4% on an industrial FPU unit.