When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient | Paper Citation Record

Observation 7a34d8fc-b13d-43e3-ac7f-067bd024a5c4 · outbound

This paper cites On the theory of policy gradient methods: Optimality, approximation, and distribution shift.The Journal of Machine Learning Research, 22(1):4431–4506.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient On the theory of policy gradient methods: Optimality, approximation, and distribution shift.The Journal of Machine Learning Research, 22(1):4431–4506

Reference 1

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.863945Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 25482215-e86d-4465-b6c4-70102db495c1 · outbound

This paper cites Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs

Reference 2

Resolution: verified exact; arxiv_id, observed 2026-05-13T01:42:22.769824Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation bd5e698c-3480-452c-9ca7-4047323ee798 · outbound

This paper cites Understanding the impact of entropy on policy optimization.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Understanding the impact of entropy on policy optimization

Reference 3

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.904768Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0d54c6f2-c896-41a5-ac7e-88e78269e706 · outbound

This paper cites Concrete Problems in AI Safety.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Concrete Problems in AI Safety

Reference 4

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.510706Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 246d52d9-a8e3-4c28-b4bb-2de9932b416f · outbound

This paper cites Potential-based shaping in model-based reinforce- ment learning.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Potential-based shaping in model-based reinforce- ment learning

Reference 5

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.984806Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation b373cf31-2bdd-4a9f-9bfa-2a15ebe38687 · outbound

This paper cites InfAlign: Inference-aware language model alignment.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient InfAlign: Inference-aware language model alignment

Reference 6

Resolution: verified exact; arxiv_id, observed 2026-05-11T23:41:19.360622Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 053b26ba-67ef-4828-a9e4-f1c62adc4f01 · outbound

This paper cites Dota 2 with Large Scale Deep Reinforcement Learning.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Dota 2 with Large Scale Deep Reinforcement Learning

Reference 7

Resolution: verified exact; arxiv_id, observed 2026-05-12T22:18:16.018301Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 51e7cdf1-4e82-42b6-803e-d770919c1c2b · outbound

This paper cites an unresolved cited work.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Unresolved cited work

Reference 8

Resolution: unresolved; raw_fallback, observed 2026-05-27T02:18:25.100242Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation b77832eb-85f9-4636-a428-0ba3f8298037 · outbound

This paper cites The accuracy paradox in rlhf: When better reward models don’t yield better language models.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient The accuracy paradox in rlhf: When better reward models don’t yield better language models

Reference 9

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.072863Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation b0803899-d69e-443e-b11f-e67df679bff4 · outbound

This paper cites Heuristic-guided reinforcement learning.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Heuristic-guided reinforcement learning

Reference 10

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.859056Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation d4899a4c-28d0-444c-a2f5-3d47850b65af · outbound

This paper cites Learning navigation behaviors end-to-end with autorl.IEEE Robotics and Automation Letters, 4(2):2007–2014.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Learning navigation behaviors end-to-end with autorl.IEEE Robotics and Automation Letters, 4(2):2007–2014

Reference 11

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.013493Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 2df8d76b-484d-47ff-a1e4-6730ad9ade43 · outbound

This paper cites More is less: inducing sparsity via overparameteriza- tion.Information and Inference: A Journal of the IMA, 12(3):1437–1460.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient More is less: inducing sparsity via overparameteriza- tion.Information and Inference: A Journal of the IMA, 12(3):1437–1460

Reference 12

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.175394Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0bafe182-0e35-4c0b-bd22-7149147b4048 · outbound

This paper cites Reward model ensembles help mitigate overoptimization.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Reward model ensembles help mitigate overoptimization

Reference 13

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.068336Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 001ce548-e2a0-4a8a-b017-38e03f8407c4 · outbound

This paper cites Ultrafeedback: Boosting language models with high-quality feedback.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Ultrafeedback: Boosting language models with high-quality feedback

Reference 14

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.141787Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation b6bba76d-fb51-4eb3-8356-1311f0257689 · outbound

This paper cites Maximum expected hitting cost of a markov decision process and informativeness of rewards.Advances in Neural Information Processing Systems.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Maximum expected hitting cost of a markov decision process and informativeness of rewards.Advances in Neural Information Processing Systems

Reference 15

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.885611Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 3195c83e-82e7-4a9b-983a-e246baccc8b9 · outbound

This paper cites Exploration-guided reward shaping for reinforcement learning under sparse rewards.Advances in Neural Information Processing Systems.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Exploration-guided reward shaping for reinforcement learning under sparse rewards.Advances in Neural Information Processing Systems

Reference 16

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.838961Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 61d4bf89-748f-4f6b-87dc-723058a0a7dd · outbound

This paper cites Continuous vs.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Continuous vs

Reference 17

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.868010Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 77b8edde-0e2a-42b1-b3da-fffbe35e77f9 · outbound

This paper cites The perils of optimizing learned reward functions: Low training error does not guarantee low regret.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient The perils of optimizing learned reward functions: Low training error does not guarantee low regret

Reference 18

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.895553Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation e2304c86-3d52-4ac1-9fc0-d76092cecfea · outbound

This paper cites Is a good foundation necessary for efficient reinforcement learning? the computational role of the base model in exploration.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Is a good foundation necessary for efficient reinforcement learning? the computational role of the base model in exploration

Reference 19

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.786800Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation d14e7f16-b218-42ac-aab0-be81f3b28c37 · outbound

This paper cites How to evaluate reward models for rlhf.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient How to evaluate reward models for rlhf

Reference 20

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.123615Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 39e03859-ef9c-4662-8d30-f9a6d8735a40 · outbound

This paper cites Scaling laws for reward model overoptimization.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Scaling laws for reward model overoptimization

Reference 21

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.805765Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation bacdaebb-550e-4a6d-8693-a5602a9b3cf6 · outbound

This paper cites An alternate policy gradient estimator for softmax policies.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient An alternate policy gradient estimator for softmax policies

Reference 22

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.018960Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 00c18eea-a1cd-4f85-95c7-8248d629d8c7 · outbound

This paper cites Quantifying differences in reward functions.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Quantifying differences in reward functions

Reference 23

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.828078Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 5ff05225-ae09-41cd-b833-66e2010ba483 · outbound

This paper cites The Llama 3 Herd of Models.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient The Llama 3 Herd of Models

Reference 24

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.354997Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation cd9ba846-5e88-440f-81d5-a62981433e19 · outbound

This paper cites Reward shaping in episodic reinforcement learning.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Reward shaping in episodic reinforcement learning

Reference 25

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.109371Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0cee9084-8afc-4ecf-b712-e6f53d50705b · outbound

This paper cites DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

Reference 26

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.569181Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation b2b53894-57f5-4717-aae4-3f45bd19d9af · outbound

This paper cites Unpacking reward shaping: Understanding the benefits of reward engineering on sample complexity.Advances in Neural Information Processing Systems.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Unpacking reward shaping: Understanding the benefits of reward engineering on sample complexity.Advances in Neural Information Processing Systems

Reference 27

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.795897Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation e8bd2e1e-9a1d-4d8e-bf3e-e7939a5ca7fe · outbound

This paper cites Neural Replicator Dynamics.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Neural Replicator Dynamics

Reference 28

Resolution: verified exact; arxiv_id, observed 2026-05-11T23:41:19.370655Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation a9db3d6b-33f0-4820-b168-cb647a65ce88 · outbound

This paper cites Is best-of-n the best of them? coverage, scaling, and optimality in inference-time alignment.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Is best-of-n the best of them? coverage, scaling, and optimality in inference-time alignment

Reference 29

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.843538Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 2e768850-5d4d-4dc9-a3bf-9d122f133052 · outbound

This paper cites On the Emergence of Implicit Curriculum in RLVR Learning Dynamics.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient On the Emergence of Implicit Curriculum in RLVR Learning Dynamics

Reference 30

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.411893Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f4292780-bbe9-47f2-8cb6-4447e8fb1132 · outbound

This paper cites Pitfalls of rule- and model-based verifiers–a case study on mathematical reasoning.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Pitfalls of rule- and model-based verifiers–a case study on mathematical reasoning

Reference 31

Resolution: verified exact; arxiv_id, observed 2026-05-11T23:41:19.516586Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 79356e28-dcfb-46ca-b1ba-7c2e125cd99e · outbound

This paper cites Goodhart’s law in reinforcement learning.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Goodhart’s law in reinforcement learning

Reference 32

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.119332Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 9ee1562b-2bad-4ee7-994d-05965bcfd311 · outbound

This paper cites Beyond stationarity: Convergence analysis of stochastic softmax policy gradient methods.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Beyond stationarity: Convergence analysis of stochastic softmax policy gradient methods

Reference 33

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.890286Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 7b1c7b91-3a83-4ce4-819c-d1c6b11ad14f · outbound

This paper cites Buy 4 reinforce samples, get a baseline for free!Deep Reinforcement Learning Meets Structured Prediction ICLR Workhsop.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Buy 4 reinforce samples, get a baseline for free!Deep Reinforcement Learning Meets Structured Prediction ICLR Workhsop

Reference 34

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.800342Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation c313ba31-c8b0-4040-8b8d-569a560a30e6 · outbound

This paper cites A neural collapse perspective on feature evolution in graph neural networks.Advances in Neural Information Processing Systems.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient A neural collapse perspective on feature evolution in graph neural networks.Advances in Neural Information Processing Systems

Reference 35

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.791469Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 9f74ff2c-b947-4f68-b9c1-2b78bd4fd77a · outbound

This paper cites Correlated proxies: A new definition and improved mitigation for reward hacking.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Correlated proxies: A new definition and improved mitigation for reward hacking

Reference 36

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.777821Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation b79c9aee-6548-486f-8dd0-87ec58fd2409 · outbound

This paper cites Tulu 3: Pushing Frontiers in Open Language Model Post-Training.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Tulu 3: Pushing Frontiers in Open Language Model Post-Training

Reference 37

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.452354Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 6b3e9138-9638-4f18-9945-7bf498b74769 · outbound

This paper cites Rewardbench: Evaluating reward models for language modeling.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Rewardbench: Evaluating reward models for language modeling

Reference 38

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.880935Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation a9957a1d-b670-4804-bde9-34413deea7cb · outbound

This paper cites The influence of reward on the speed of reinforcement learning: An analysis of shaping.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient The influence of reward on the speed of reinforcement learning: An analysis of shaping

Reference 39

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.810202Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation c5b2414b-01f3-4776-957a-58931e1a29a7 · outbound

This paper cites Softmax policy gradient methods can take exponential time to converge.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Softmax policy gradient methods can take exponential time to converge

Reference 40

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.160033Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 3acb3b2f-9de6-41fa-b43c-54d351479c5a · outbound

This paper cites Hashimoto.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Hashimoto

Reference 41

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.170643Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 054f8504-b2e6-4b03-a827-9ab741972c4c · outbound

This paper cites Rethinking the Global Convergence of Softmax Policy Gradient with Linear Function Approximation.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Rethinking the Global Convergence of Softmax Policy Gradient with Linear Function Approximation

Reference 42

Resolution: verified exact; arxiv_id, observed 2026-05-11T23:41:19.689881Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 8e41e6ee-74ea-4de0-96cb-53143922b22e · outbound

This paper cites Skywork-Reward-V2: Scaling Preference Data Curation via Human-AI Synergy.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Skywork-Reward-V2: Scaling Preference Data Curation via Human-AI Synergy

Reference 43

Resolution: verified exact; arxiv_id, observed 2026-05-16T22:17:43.404381Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 78f1a5c5-0921-403e-ae92-c241984b705b · outbound

This paper cites Elementary Analysis of Policy Gradient Methods.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Elementary Analysis of Policy Gradient Methods

Reference 44

Resolution: verified exact; arxiv_id, observed 2026-05-11T23:41:19.581704Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation dcfc11c6-1ee1-4546-b85c-2d855129d2c7 · outbound

This paper cites RLTF: Reinforce- ment learning from unit test feedback.Transactions on Machine Learning Research.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient RLTF: Reinforce- ment learning from unit test feedback.Transactions on Machine Learning Research

Reference 45

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.150697Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 6c70cdc9-b1a0-4538-86f7-feb0e02f2e9c · outbound

This paper cites Rm-bench: Benchmarking reward mod- els of language models with subtlety and style.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Rm-bench: Benchmarking reward mod- els of language models with subtlety and style

Reference 46

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.050469Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation d310456e-86a5-4dd7-a59d-3dad4320a502 · outbound

This paper cites RewardBench 2: Advancing Reward Model Evaluation.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient RewardBench 2: Advancing Reward Model Evaluation

Reference 47

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.467386Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation da614b5e-a305-461e-a951-b4d3dc04a718 · outbound

This paper cites Reward engineering for reinforcement learning in software tasks.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Reward engineering for reinforcement learning in software tasks

Reference 48

Resolution: verified exact; arxiv_id, observed 2026-05-11T23:41:19.440658Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0c30e2b3-8a65-4ef6-aad2-cc8f10b05a47 · outbound

This paper cites Reward functions for accelerated learning.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Reward functions for accelerated learning

Reference 49

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.059742Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation c95685b4-ae34-43b6-b82a-ae9695cbb204 · outbound

This paper cites Escaping the gravitational pull of softmax.Advances in Neural Information Processing Systems.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Escaping the gravitational pull of softmax.Advances in Neural Information Processing Systems

Reference 50

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.999383Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 3d0dd837-3d61-420f-bbb6-f7eb95c65c0d · outbound

This paper cites On the global convergence rates of softmax policy gradient methods.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient On the global convergence rates of softmax policy gradient methods

Reference 51

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.146068Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 960fb550-3502-4e22-9e08-65740ce9b899 · outbound

This paper cites Leveraging non-uniformity in first-order non-convex optimization.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Leveraging non-uniformity in first-order non-convex optimization

Reference 52

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.782435Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 248c8549-954e-416d-8029-11395b13baf0 · outbound

This paper cites Ordering-based conditions for global convergence of policy gradient methods.Advances in Neural Information Processing Systems.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Ordering-based conditions for global convergence of policy gradient methods.Advances in Neural Information Processing Systems

Reference 53

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.872459Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 33d3b4f0-5ba3-434f-ae45-e150ee40740d · outbound

This paper cites Stochastic gradient succeeds for bandits.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Stochastic gradient succeeds for bandits

Reference 54

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.823342Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 987f6b8a-5dfd-40a6-9e92-7c999a5a1dde · outbound

This paper cites Policy invariance under reward transformations: Theory and application to reward shaping.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Policy invariance under reward transformations: Theory and application to reward shaping

Reference 55

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.032695Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation e28f85ac-e154-4a89-a61c-b8fa25db28e3 · outbound

This paper cites 2 OLMo 2 Furious.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient 2 OLMo 2 Furious

Reference 56

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.431341Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation a170f578-7634-4aa8-bc62-0b5865648f16 · outbound

This paper cites Olmo 3.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Olmo 3

Reference 57

Resolution: metadata mismatch; local_arxiv, observed 2026-05-11T23:41:19.524345Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 519a7f06-fdb2-4a40-8947-523801103c68 · outbound

This paper cites Training language models to follow instructions with human feedback.Advances in Neural Information Processing Systems.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Training language models to follow instructions with human feedback.Advances in Neural Information Processing Systems

Reference 58

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.055485Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation e0aef6b5-3da2-4eea-b445-f7fab22c60b0 · outbound

This paper cites Reward gaming in conditional text generation.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Reward gaming in conditional text generation

Reference 59

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.900348Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 30751c67-af25-4434-8dad-9db33a339f83 · outbound

This paper cites Automatic differentiation in pytorch.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Automatic differentiation in pytorch

Reference 60

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.082466Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 57dbbfe6-b229-48be-a261-b000c3d43132 · outbound

This paper cites Amp: Adversarial motion priors for stylized physics-based character control.ACM Transactions on Graphics (ToG), 40(4):1–20.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Amp: Adversarial motion priors for stylized physics-based character control.ACM Transactions on Graphics (ToG), 40(4):1–20

Reference 61

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.155802Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation dd783c71-682e-4090-993b-c31105667067 · outbound

This paper cites Generalizing verifiable instruction following.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Generalizing verifiable instruction following

Reference 62

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.037042Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 097d2b6e-9754-4d08-8403-6e69c16330f1 · outbound

This paper cites Outcome-Based RL Provably Leads Transformers to Reason, but Only With the Right Data.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Outcome-Based RL Provably Leads Transformers to Reason, but Only With the Right Data

Reference 63

Resolution: verified exact; arxiv_id, observed 2026-06-04T02:07:08.062014Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 1760c012-5233-4b15-9b96-c2cf885b0f2f · outbound

This paper cites Learning to drive a bicycle using reinforcement learning and shaping.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Learning to drive a bicycle using reinforcement learning and shaping

Reference 64

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.834343Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation ad7c4af3-c797-4a68-afb8-233e215bcba2 · outbound

This paper cites Implicit regularization in deep learning may not be explainable by norms.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Implicit regularization in deep learning may not be explainable by norms

Reference 65

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.994339Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 10457d44-eb20-4523-a1c2-b0ab66501224 · outbound

This paper cites Implicit regularization in tensor factorization.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Implicit regularization in tensor factorization

Reference 66

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.091425Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation efc34028-c14f-4e67-a47b-ed586a5b2417 · outbound

This paper cites Implicit regularization in hierarchical tensor factorization and deep convolutional neural networks.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Implicit regularization in hierarchical tensor factorization and deep convolutional neural networks

Reference 67

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.004683Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 471d2902-ec56-4d25-a6a4-00650a010efd · outbound

This paper cites Susskind, and Etai Littwin.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Susskind, and Etai Littwin

Reference 68

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.028147Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 5b1a8b75-d679-4728-8d7b-6f986db2992e · outbound

This paper cites What makes a reward model a good teacher? an optimization perspective.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient What makes a reward model a good teacher? an optimization perspective

Reference 69

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.095922Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation d585a040-940a-4d8f-be86-bd62a25510d7 · outbound

This paper cites Why is your language model a poor implicit reward model? InInternational Conference on Learning Representations.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Why is your language model a poor implicit reward model? InInternational Conference on Learning Representations

Reference 70

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.854294Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation cee29f65-6e2d-4f9f-aa72-c6d6a227f58c · outbound

This paper cites On the effective number of linear regions in shallow univariate relu networks: Convergence guarantees and implicit bias.Advances in Neural Information Processing Systems.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient On the effective number of linear regions in shallow univariate relu networks: Convergence guarantees and implicit bias.Advances in Neural Information Processing Systems

Reference 71

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.773208Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 4b9d81ba-92ac-41be-82b4-16e2a588389a · outbound

This paper cites Exact solutions to the nonlinear dynamics of learning in deep linear neural networks.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Exact solutions to the nonlinear dynamics of learning in deep linear neural networks

Reference 72

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.848707Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 903a3577-8159-47d5-953b-5ea29d577059 · outbound

This paper cites Ray Interference: a Source of Plateaus in Deep Reinforcement Learning.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Ray Interference: a Source of Plateaus in Deep Reinforcement Learning

Reference 73

Resolution: verified exact; arxiv_id, observed 2026-05-11T23:41:19.487655Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 16b258bc-2c08-45a3-bc04-c0cba41e2226 · outbound

This paper cites Proximal Policy Optimization Algorithms.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Proximal Policy Optimization Algorithms

Reference 74

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.387711Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 4d69d36c-3d2d-4992-a032-4116c8f57c0b · outbound

This paper cites DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

Reference 75

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.624857Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 3c3ce965-d6d4-4970-8910-bca64027c993 · outbound

This paper cites Intrinsically motivated reinforce- ment learning: An evolutionary perspective.IEEE Transactions on Autonomous Mental Development, 2 (2):70–82.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Intrinsically motivated reinforce- ment learning: An evolutionary perspective.IEEE Transactions on Autonomous Mental Development, 2 (2):70–82

Reference 76

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.132804Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f18a4585-de3d-4add-b1dd-e5088a989381 · outbound

This paper cites Defining and characterizing reward gaming.Advances in Neural Information Processing Systems.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Defining and characterizing reward gaming.Advances in Neural Information Processing Systems

Reference 77

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.041345Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 4685644c-8a2c-47f1-a102-03505ab7b8ff · outbound

This paper cites Starc: A general framework for quantifying differences between reward functions.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Starc: A general framework for quantifying differences between reward functions

Reference 78

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.045761Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation ff578be5-2523-4960-adf2-91febc47327c · outbound

This paper cites The implicit bias of structured state space models can be poisoned with clean labels.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient The implicit bias of structured state space models can be poisoned with clean labels

Reference 79

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.814814Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 8052fc45-4fd5-4da8-9a7f-298f85c40e55 · outbound

This paper cites Reward design via online gradient ascent.Advances in Neural Information Processing Systems, 23.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Reward design via online gradient ascent.Advances in Neural Information Processing Systems, 23

Reference 80

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.127911Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 5ad401cd-72af-4fef-8bb7-e731b91238ce · outbound

This paper cites Learning to summarize with human feedback.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Learning to summarize with human feedback

Reference 81

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.876597Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 2ae292cf-cc5a-4eb9-afaf-454659e56a41 · outbound

This paper cites Rl grokking recipe: How does rl unlock and transfer new algorithms in llms? InInternational Conference on Learning Representations.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Rl grokking recipe: How does rl unlock and transfer new algorithms in llms? InInternational Conference on Learning Representations

Reference 82

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.105202Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 33501b71-3632-4fad-8b33-1198ac9fb464 · outbound

This paper cites All roads lead to likelihood: The value of reinforcement learning in fine-tuning.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient All roads lead to likelihood: The value of reinforcement learning in fine-tuning

Reference 83

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.063971Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 82c9f1d3-92ad-4d9d-ba59-3a1cf0f06f06 · outbound

This paper cites Understanding the performance gap between online and offline alignment algorithms.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Understanding the performance gap between online and offline alignment algorithms

Reference 84

Resolution: verified exact; arxiv_id, observed 2026-05-11T23:41:19.645340Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 530bd498-9511-42bb-ac2e-50bc6c38d329 · outbound

This paper cites InThe Thirty-ninth Annual Conference on Neural Information Process- ing Systems.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient InThe Thirty-ninth Annual Conference on Neural Information Process- ing Systems

Reference 85

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T23:41:19.591630Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation e9020eb9-ad60-4f07-bbc6-55043e5a9723 · outbound

This paper cites Perturbation analysis of neural collapse.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Perturbation analysis of neural collapse

Reference 86

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.137475Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f6f5f133-7298-4f34-8121-1faf2ea73373 · outbound

This paper cites Implicit regularization in relu networks with the square loss.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Implicit regularization in relu networks with the square loss

Reference 87

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:24.819104Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 78c93807-1306-4273-b3fb-328262e3caf1 · outbound

This paper cites Interpretable Preferences via Multi-Objective Reward Modeling and Mixture-of-Experts.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Interpretable Preferences via Multi-Objective Reward Modeling and Mixture-of-Experts

Reference 88

Resolution: verified exact; arxiv_id, observed 2026-05-11T23:41:19.698135Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 97aadcd7-abbf-4da2-a2fc-00e035fbd907 · outbound

This paper cites Rethinking reward model evaluation: Are we barking up the wrong tree? InInternational Conference on Learning Representations.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Rethinking reward model evaluation: Are we barking up the wrong tree? InInternational Conference on Learning Representations

Reference 89

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.114951Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation a299a1bd-a796-466a-bc73-4d8fd1093a67 · outbound

This paper cites Simple statistical gradient-following algorithms for connectionist reinforcement learning.Machine learning, 8(3):229–256.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Simple statistical gradient-following algorithms for connectionist reinforcement learning.Machine learning, 8(3):229–256

Reference 90

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.165478Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation ee639010-6b2f-4c5c-81d1-f0e14cb02f1c · outbound

This paper cites HuggingFace's Transformers: State-of-the-art Natural Language Processing.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient HuggingFace's Transformers: State-of-the-art Natural Language Processing

Reference 91

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.611570Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation df978ab5-46a8-4420-8cba-df6b7d5ccc91 · outbound

This paper cites Kernel and rich regimes in overparametrized models.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Kernel and rich regimes in overparametrized models

Reference 92

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.008874Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 91cb503f-62a6-4488-aa6a-22baec79bc27 · outbound

This paper cites Dynamics-aware comparison of learned reward functions.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Dynamics-aware comparison of learned reward functions

Reference 93

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.077423Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation cf047b6a-c400-4d90-8d3d-0e3bfbe6ea2d · outbound

This paper cites Is DPO Superior to PPO for LLM Alignment? A Comprehensive Study.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Is DPO Superior to PPO for LLM Alignment? A Comprehensive Study

Reference 94

Resolution: verified exact; arxiv_id, observed 2026-05-11T23:41:19.562482Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f3fcb075-38e6-4eaf-b9b1-ee1850569394 · outbound

This paper cites Qwen3 Technical Report.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Qwen3 Technical Report

Reference 95

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.605231Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 59af3c34-be70-43a5-9e92-5b9434de1e4d · outbound

This paper cites DAPO: An Open-Source LLM Reinforcement Learning System at Scale.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient DAPO: An Open-Source LLM Reinforcement Learning System at Scale

Reference 96

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.708864Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f7c9d62b-ed80-410c-8034-fa70ad788a31 · outbound

This paper cites On learning intrinsic rewards for policy gradient methods.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient On learning intrinsic rewards for policy gradient methods

Reference 97

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.024062Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 75db98d7-8b6b-4d13-8254-9d0ce84bccde · outbound

This paper cites Rmb: Comprehensively benchmarking reward models in llm alignment.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Rmb: Comprehensively benchmarking reward models in llm alignment

Reference 98

Resolution: verified fuzzy; raw_fallback, observed 2026-05-27T02:18:25.087015Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 86b2e17f-577a-4982-8958-8f23b466519c · outbound

This paper cites Fine-Tuning Language Models from Human Preferences.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient Fine-Tuning Language Models from Human Preferences

Reference 99

Resolution: verified exact; local_arxiv, observed 2026-05-11T23:41:19.598253Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation ae83b582-bad8-46ba-b797-a75062ed3730 · outbound

This paper cites partial rewards.

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient partial rewards

Reference 100

Resolution: malformed identifier; arxiv_id, observed 2026-05-11T23:41:19.586625Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Coverage vector

Reference resolution

Outbound references

Pith citing papers