Reward models need more than just accuracy for effective RLHF.
problem The effectiveness of reward models in RLHF is not fully understood.
method An optimization perspective to evaluate reward models.
result Reward models with low reward variance can lead to a flat optimization landscape, hindering performance.
Reward collapse occurs when ranking-based reward models yield uniform rewards for different prompts.
problem Reward collapse in aligning large language models with human preferences.
method Introduced a prompt-aware optimization scheme to derive closed-form expressions for reward distributions.
result Our prompt-aware utility functions significantly alleviate reward collapse during training.
A new method resolves non-identifiability in reward modeling using anchor labels.
problem Non-identifiability in reward modeling from pairwise preferences alone.
method Anchor-guided Variance-aware Reward Modeling (AVRM) framework.
result AVRM resolves non-identifiability and improves reward modeling performance.
The study categorizes reward errors in reinforcement learning, finding some can be beneficial.
problem Training language models with imperfect proxy rewards.
method Theoretical analysis of policy gradient optimization and categorization of reward errors.
result Reward errors can be benign or even beneficial, preventing policy from stalling.
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.
Reward hacking exploits misspecified rewards, affecting agent capabilities and true performance.
problem Reward hacking in RL models exploiting reward misspecifications.
method Constructed four RL environments with misspecified rewards; analyzed agent capabilities and behavior.
result More capable agents exploit reward misspecifications, achieving higher proxy reward but lower true reward.
Learning reward functions from data is a promising path towards achieving scalable Reinforcement Learning (RL) for robotics. However, a major challenge in training agents from learned reward models is that the agent can learn to exploit errors in the reward model to achieve high reward behaviors that do not correspond …
This work shows how approximate reward models can significantly improve inference-time scaling.
problem Improving the efficiency of inference for large language models.
method Identifying the Bellman error of approximate reward models and using Sequential Monte Carlo (SMC) for inference.
result Approximate reward models can reduce computational complexity from exponential to polynomial in T. 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.
We here adopt Bayesian nonparametric mixture models to extend multi-armed bandits in general, and Thompson sampling in particular, to scenarios where there is reward model uncertainty. In the stochastic multi-armed bandit, the reward for the played arm is generated from an unknown distribution. Reward uncertainty, i.e.…
Method tackles uncertainty in reward models for LLMs from heterogeneous human feedback.
problem Uncertainty in reward models for LLMs from heterogeneous human feedback.
method Heterogeneous preference framework and alternating gradient descent algorithm.
result Established theoretical guarantees for estimator convergence and asymptotic distribution.
Iterative tilting fine-tunes diffusion models for reward-tilted distributions.
problem Fine-tuning diffusion models for reward-tilted distributions.
method Decomposes large reward tilts into smaller, tractable tilts via first-order Taylor expansion, avoiding backpropagation.
result Validated on a two-dimensional Gaussian mixture, achieving exact closed-form solutions.
The paper proposes a model reward scheme for collaborative ML based on Shapley value and information gain.
problem Designing fair incentives for collaborative machine learning.
method The paper proposes a reward scheme based on Shapley value and information gain, with properties like fairness and stability.
result The proposed reward scheme satisfies fairness and trade-offs between desirable properties via an adjustable parameter.
Learning reward functions can lead to poor policy performance despite low error.
problem Low error in learned reward functions does not guarantee low regret in policy performance.
method Mathematical analysis of reward learning and policy optimization.
result A low expected test error of the reward model guarantees low worst-case regret, but error-regret mismatch can occur with certain data distributions.
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.
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.
Unified framework improves diffusion model rewards without full trajectories.
problem Limited theoretical understanding of guided diffusion samplers.
method Developed a unified algorithmic and theoretical framework for diffusion guidance and reward-guided diffusion.
result Framework shows CFG decreases expected reciprocal of classifier probability.
Reinforcement Learning (RL) agents require the specification of a reward signal for learning behaviours. However, introduction of corrupt or stochastic rewards can yield high variance in learning. Such corruption may be a direct result of goal misspecification, randomness in the reward signal, or correlation of the rew…
EBRM improves robustness and generalization of language model rewards.
problem Challenges in capturing complex human preferences and generalizing to unseen data in reward models.
method Energy-Based Reward Model (EBRM) that models reward distribution explicitly, using conflict-aware data filtering, label-noise-aware contrastive training, and hybrid initialization.
result Significant improvements in robustness and generalization, up to 5.97% improvement in safety-critical alignment tasks.
One obstacle to applying reinforcement learning algorithms to real-world problems is the lack of suitable reward functions. Designing such reward functions is difficult in part because the user only has an implicit understanding of the task objective. This gives rise to the agent alignment problem: how do we create age…
We seek to align agent behavior with a user's objectives in a reinforcement learning setting with unknown dynamics, an unknown reward function, and unknown unsafe states. The user knows the rewards and unsafe states, but querying the user is expensive. To address this challenge, we propose an algorithm that safely and …
A novel framework refines diffusion models iteratively for better downstream reward optimization.
problem Optimizing reward functions during inference of diffusion models.
method Iterative refinement process with noising and reward-guided denoising steps.
result Superior empirical performance in protein and DNA design.
Scalable model for slate recommendation learns reward probabilities.
problem Scalable personalized slate recommendation in large action spaces.
method Probabilistic Rank and Reward (PRR) model combining reward, interaction, and rank.
result PRR outperforms existing methods and is scalable to large action spaces.
Language models perform worse with implicit reward models than explicit ones.
problem Understanding why implicit reward models generalize worse than explicit ones.
method Investigated the root cause of the generalization gap between IM-RMs and EX-RMs.
result Implicit reward models rely more on superficial token-level cues, leading to worse generalization.
New models for bandit problems with fidelity rewards are introduced and analyzed.
problem Fidelity rewards in bandit problems to incentivize loyalty.
method Two models (loyalty-points and subscription) for fidelity rewards; stochastic and adversarial settings considered.
result Sublinear regret bounds for some models, worst case lower bounds for others.
Trade-R1 bridges verifiable rewards to stochastic financial markets via process-level reasoning verification.
problem Extending RL to financial markets where rewards are verifiable but noisy.
method A verification method that transforms reasoning over financial documents into a structured RAG task, using a triangular consistency metric.
result DSR achieves superior cross-market generalization while maintaining reasoning consistency.
To solve complex real-world problems with reinforcement learning, we cannot rely on manually specified reward functions. Instead, we can have humans communicate an objective to the agent directly. In this work, we combine two approaches to learning from human feedback: expert demonstrations and trajectory preferences. …
Extends reinforcement learning alignment to scalar rewards, improving math reasoning.
problem Designing reinforcement learning algorithms for general LLM alignment.
method Introduces f-GRPO and f-HAL, estimating f-divergences between reward-aligned and unaligned distributions.
result Improves math reasoning RLVR tasks and mitigates reward hacking.
Paper addresses reward hacking in preference optimization, proposing POWER-DL to improve AI alignment.
problem Reward hacking problem in preference optimization, leading to undesired behaviors.
method POWER-DL combines robust reward maximization and dynamic label updates to mitigate reward hacking.
result POWER-DL consistently outperforms state-of-the-art methods on alignment benchmarks.
Study optimal reward schemes for inducing desired player performance in risky contests.
problem Designing optimal rewards to encourage desired performance levels in risky contests.
method Analyzed the optimal reward schemes for inducing average and specific rank performance.
result Optimal reward schemes can have surprising shapes, not just related to inequality.
Reinforcement learning usually uses the feedback rewards of environmental to train agents. But the rewards in the actual environment are sparse, and even some environments will not rewards. Most of the current methods are difficult to get good performance in sparse reward or non-reward environments. Although using shap…
Improves generative models by optimizing rewards and sample editing.
problem Efficiently generating high-reward samples with structural constraints.
method Introduces MDM-VGB, a discrete diffusion sampler that augments unmasking generation with reward-guided remasking.
result MDM-VGB achieves quadratic complexity and robustness to noise, outperforming heuristics like best-of-N. The study analyzes how neural reward models learn features for policy optimization in a Gaussian single-index model.
problem Reward modeling in policy optimization and its impact on downstream value.
method Two-stage neural reward model: first learns hidden direction, then fits readout layer.
result For any feature-learning temperature above a dimension-free threshold, a constant fraction of neurons recover the hidden direction.
PILAF optimizes reward models from human feedback for better policy alignment.
problem Creating accurate reward models from human feedback for policy optimization.
method Policy-Interpolated Learning for Aligned Feedback (PILAF) that explicitly aligns preference learning with maximizing underlying oracle reward.
result PILAF is optimal from both optimization and statistical perspectives, demonstrating strong performance in RLHF settings.
DDO-RM improves reward-based policies by converting reward scores into a target distribution.
problem Improving reward-based policies when the reward function is simpler than the policy.
method Converts reward scores into a target distribution and uses KL-regularized mirror-descent updates.
result DDO-RM outperforms DPO in pair accuracy and mean margin.
DPA aligns LLMs with multi-objective rewards for diverse user preferences.
problem Fine-grained control over LLMs for diverse user needs.
method Integrates multi-objective reward modeling and directional preference control.
result DPA offers better performance trade-offs and intuitive user control over LLM generation.
Mitigates biases in reward models using variational inference.
problem Spurious correlations in reward models that align large language models with human preferences.
method Formulates data-generating process, identifies non-spurious latent variables, and uses variational inference to recover them.
result Effective mitigation of spurious correlation issues, yielding more robust reward models.
A fair reward system boosts participation in federated learning.
problem Fairness in federated learning among competitive agents with siloed data.
method Hierarchically fair federated learning (HFFL) framework with proportional rewards based on contribution levels.
result Efficacy of HFFL in maintaining fairness and facilitating federated learning in competitive settings.
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.
Off-policy evaluation for MNAR rewards in MDPs
problem Off-policy evaluation in MDPs with MNAR rewards
method Formalizing a reward-dependent propensity model and using future states as shadow variables
result Proposed an Fitted-Q-Evaluation-style estimator that propagates recovered rewards while allowing target policies to depend on past missingness indicators
Extends linear MDP to handle nonlinear rewards.
problem Restrictive linear MDP assumption limits real-world applicability.
method Proposes Generalized Linear MDP (GLMDP) with GLMs for rewards.
result Develops offline RL algorithms achieving suboptimality guarantees.
This paper formalizes reward training in RLHF and provides theoretical guarantees.
problem Costly human feedback collection in RLHF.
method Linear contextual dueling bandit method for regret minimization.
result Derives bounds on simple regret for offline reward training.
Paper shows RLHF can be solved similarly to standard RL.
problem Difficulty of RLHF compared to standard RL.
method Reduction to reward-based RL techniques.
result RLHF can be solved using existing algorithms for reward-based RL.
CausalRM models rewards from user feedback, overcoming noise and bias.
problem Aligning language models with user preferences from noisy, biased feedback.
method Causal-theoretic reward modeling framework addressing noise and bias in observational feedback.
result CausalRM learns accurate reward signals from noisy and biased observational feedback.
Adaptive reward models capture individual preferences from human feedback.
problem Learning a reward model that can be specialised to a user.
method Empirical risk minimisation and PAC bound analysis.
result Adaptive reward models benefit from the heterogeneity of user preferences.
Adjoint Matching improves flow and diffusion models with reward fine-tuning.
problem Improving generative models with reward fine-tuning.
method Casting reward fine-tuning as stochastic optimal control (SOC) and enforcing a specific noise schedule.
result Adjoint Matching outperforms existing SOC algorithms.
Align-RUDDER improves reinforcement learning with few demonstrations by redistributing rewards.
problem Learning complex tasks with sparse and delayed rewards using few demonstrations.
method Align-RUDDER uses a profile model for reward redistribution based on multiple sequence alignment of demonstrations.
result Align-RUDDER outperforms competitors on complex tasks with few demonstrations.
A new reward shaping method balances learning efficiency and effectiveness for robot manipulation.
problem Efficient and effective learning in robot manipulations with system uncertainty.
method Dense2Sparse reward shaping method combining dense and sparse rewards.
result Dense2Sparse method achieves higher expected reward and better system uncertainty tolerance.