Mitigates overoptimization in RLHF by reformulating SFT loss as a preference optimization loss.
problem Overoptimization in RLHF where reward model misguides generative model.
method Proposes a theoretical algorithm that minimizes maximum likelihood estimation and reward penalty, reformulates as simple objective combining preference optimization and supervised learning losses.
result Improved performance of RPO compared to DPO baselines in aligning LLMs.
Improved DPO framework penalizes preference uncertainty to avoid overoptimization.
problem Aligning LLMs to human preferences is challenging due to varied, context-dependent, and ambiguous preferences.
method Developed a pessimistic framework for DPO by introducing preference uncertainty penalization schemes.
result Improved overall performance and better completions on high-uncertainty responses compared to vanilla DPO.
Study scaling laws of reward model overoptimization in reinforcement learning.
problem Reward model overoptimization hinders true performance in reinforcement learning.
method Synthetic setup with fixed gold reward model; optimization using RL or best-of-n sampling; analysis of scaling laws. result Scaling laws of reward model overoptimization differ based on optimization method and scale smoothly with model parameters.
IDS improves RLHF by smoothing reward data, enhancing model performance.
problem Reward model performance degrades and overoptimization hinders true objective.
method Iterative Data Smoothing (IDS) updates model and data labels during each epoch.
result IDS outperforms traditional methods in RLHF.
Improved inference-time alignment using Best-of-N and smoothing.
problem Reward overoptimization in Best-of-N (BoN) due to poor proxy reward models.
method Introduced Soft Best-of-N (SBoN) and analyzed its performance through KL divergence and regret analysis.
result Smoothing helps SBoN mitigate reward overoptimization, especially when proxy reward quality is low.
Dan Lovallo and Daniel Kahneman must be commended for their clear identification of causes and cures to the planning fallacy in "Delusions of Success: How Optimism Undermines Executives' Decisions" (HBR July 2003). Their look at overoptimism, anchoring, competitor neglect, and the outside view in forecasting is highly …
Monitoring the biomedical literature for cases of Adverse Drug Reactions (ADRs) is a critically important and time consuming task in pharmacovigilance. The development of computer assisted approaches to aid this process in different forms has been the subject of many recent works. One particular area that has shown pro…
There are several distinct failure modes for overoptimization of systems on the basis of metrics. This occurs when a metric which can be used to improve a system is used to an extent that further optimization is ineffective or harmful, and is sometimes termed Goodhart's Law. This class of failure is often poorly unders…
Paper proposes COM-QEL to avoid overoptimistic solutions in offline optimization.
problem Incorrect extrapolation of objective values in unexplored regions.
method Integrates quantum extremal learning with conservative objective models.
result COM-QEL finds higher true objective values compared to QEL.
EC method calibrates neural networks by matching average confidence to correct label proportion.
problem Overoptimism in neural network prediction confidence.
method Expectation consistency (EC) post-training rescaling of weights.
result EC achieves similar calibration performance to temperature scaling (TS) but is based on a principled Bayesian principle.
Fine-tunes diffusion models to generate diverse samples with high genuine rewards.
problem Reward collapse in finetuning diffusion models.
method Entropy-regularized control against pretrained diffusion models.
result Efficient generation of diverse samples with high genuine rewards.
BRAID fine-tunes diffusion models to optimize reward models in offline scenarios.
problem Combining generative modeling and model-based optimization in offline scenarios.
method Conservative fine-tuning of diffusion models using RL to optimize reward models.
result BRAID outperforms existing methods in offline data, avoiding invalid designs.
DRO-REBEL improves LLM alignment by robustly updating models online.
problem Overfitting and drifting of LLMs during RLHF.
method DRO-REBEL uses type-p Wasserstein, KL, and χ2 ambiguity sets for robust online updates. result DRO-REBEL achieves faster convergence and better performance than prior methods.
Proposes a new RL method to fine-tune flow-based models with arbitrary rewards.
problem Challenges in fine-tuning continuous flow-based generative models with arbitrary reward functions.
method Online Reward-Weighted Conditional Flow Matching with Wasserstein-2 Regularization (ORW-CFM-W2)
result Achieves optimal policy convergence with controllable trade-offs between reward maximization and diversity preservation.