Imagen de Portada

Correlated Proxies: A New Definition and Improved Mitigation for Reward Hacking

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Laidlaw, Cassidy, Singhal, Shivam, Dragan, Anca
Formato:	Preprint
Publicado:	2024
Materias:	Machine Learning Artificial Intelligence
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

The Effective Horizon Explains Deep RL Performance in Stochastic Environments
por: Laidlaw, Cassidy, et al.
Publicado: (2023)

Bridging RL Theory and Practice with the Effective Horizon
por: Laidlaw, Cassidy, et al.
Publicado: (2023)

Benchmarking and Improving Monitors for Out-Of-Distribution Alignment Failure in LLMs
por: Feng, Dylan, et al.
Publicado: (2026)

AssistanceZero: Scalably Solving Assistance Games
por: Laidlaw, Cassidy, et al.
Publicado: (2025)

Reward Hacking Mitigation using Verifiable Composite Rewards
por: Tarek, Mirza Farhan Bin, et al.
Publicado: (2025)

Repairing Reward Functions with Feedback to Mitigate Reward Hacking
por: Hatgis-Kessell, Stephane, et al.
Publicado: (2025)

Adversarial Reward Auditing for Active Detection and Mitigation of Reward Hacking
por: Beigi, Mohammad, et al.
Publicado: (2026)

Reward Shaping to Mitigate Reward Hacking in RLHF
por: Fu, Jiayi, et al.
Publicado: (2025)

Robust Optimization for Mitigating Reward Hacking with Correlated Proxies
por: Liu, Zixuan, et al.
Publicado: (2026)

Distributional Preference Learning: Understanding and Accounting for Hidden Context in RLHF
por: Siththaranjan, Anand, et al.
Publicado: (2023)

ODIN: Disentangled Reward Mitigates Hacking in RLHF
por: Chen, Lichang, et al.
Publicado: (2024)

AI Alignment with Changing and Influenceable Reward Functions
por: Carroll, Micah, et al.
Publicado: (2024)

InfoRM: Mitigating Reward Hacking in RLHF via Information-Theoretic Reward Modeling
por: Miao, Yuchun, et al.
Publicado: (2024)

Iterative Label Refinement Matters More than Preference Optimization under Weak Supervision
por: Ye, Yaowen, et al.
Publicado: (2025)

Hack-Verifiable Environments: Towards Evaluating Reward Hacking at Scale
por: Roth, Amit, et al.
Publicado: (2026)

Mitigating Reward Hacking in RLHF via Advantage Sign Robustness
por: Ono, Shinnosuke, et al.
Publicado: (2026)

IR$^3$: Contrastive Inverse Reinforcement Learning for Interpretable Detection and Mitigation of Reward Hacking
por: Beigi, Mohammad, et al.
Publicado: (2026)

A Generalized Acquisition Function for Preference-based Reward Learning
por: Ellis, Evan, et al.
Publicado: (2024)

Uncertainty-Aware Reward Discounting for Mitigating Reward Hacking
por: Singha, Disha
Publicado: (2026)

MONA: Myopic Optimization with Non-myopic Approval Can Mitigate Multi-step Reward Hacking
por: Farquhar, Sebastian, et al.
Publicado: (2025)

Temper and Tilt Lead to SLOP: Reward Hacking Mitigation with Inference-Time Alignment
por: Wang, Ye, et al.
Publicado: (2026)

Beyond Reward Hacking: Causal Rewards for Large Language Model Alignment
por: Wang, Chaoqi, et al.
Publicado: (2025)

Mitigating Preference Hacking in Policy Optimization with Pessimism
por: Gupta, Dhawal, et al.
Publicado: (2025)

Learning to Assist Humans without Inferring Rewards
por: Myers, Vivek, et al.
Publicado: (2024)

LLMs Gaming Verifiers: RLVR can Lead to Reward Hacking
por: Helff, Lukas, et al.
Publicado: (2026)

Q-SFT: Q-Learning for Language Models via Supervised Fine-Tuning
por: Hong, Joey, et al.
Publicado: (2024)

Adversaries Can Misuse Combinations of Safe Models
por: Jones, Erik, et al.
Publicado: (2024)

Reward Hacking Benchmark: Measuring Exploits in LLM Agents with Tool Use
por: Thaman, Kunvar
Publicado: (2026)

Learning Temporal Distances: Contrastive Successor Features Can Provide a Metric Structure for Decision-Making
por: Myers, Vivek, et al.
Publicado: (2024)

Honesty to Subterfuge: In-Context Reinforcement Learning Can Make Honest Models Reward Hack
por: McKee-Reid, Leo, et al.
Publicado: (2024)

Feedback Loops With Language Models Drive In-Context Reward Hacking
por: Pan, Alexander, et al.
Publicado: (2024)

Interactive Dialogue Agents via Reinforcement Learning on Hindsight Regenerations
por: Hong, Joey, et al.
Publicado: (2024)

Gradient Regularization Prevents Reward Hacking in Reinforcement Learning from Human Feedback and Verifiable Rewards
por: Ackermann, Johannes, et al.
Publicado: (2026)

Countdown-Code: A Testbed for Studying The Emergence and Generalization of Reward Hacking in RLVR
por: Khalifa, Muhammad, et al.
Publicado: (2026)

Benchmarking Reward Hack Detection in Code Environments via Contrastive Analysis
por: Deshpande, Darshan, et al.
Publicado: (2026)

When Your AIs Deceive You: Challenges of Partial Observability in Reinforcement Learning from Human Feedback
por: Lang, Leon, et al.
Publicado: (2024)

On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback
por: Williams, Marcus, et al.
Publicado: (2024)

Training LLM Agents to Empower Humans
por: Ellis, Evan, et al.
Publicado: (2025)

GARDO: Reinforcing Diffusion Models without Reward Hacking
por: He, Haoran, et al.
Publicado: (2025)

Rethinking the Role of Proxy Rewards in Language Model Alignment
por: Kim, Sungdong, et al.
Publicado: (2024)