IDT learns human preferences from uncertain decisions, even when humans are suboptimal.
problem Learning human preferences from uncertain and suboptimal decisions.
method Inverse decision theory (IDT) framework, statistical analysis of IDT, characterizing sample complexity.
result Learning preferences is easier when decisions are more uncertain, even if humans are suboptimal.
Paper explores limits and possibilities of aligning LLMs with human preferences.
problem Aligning LLMs with diverse human preferences to ensure fairness and informed outcomes.
method Analysis of probabilistic representation of human preferences and preservation of diverse preferences.
result LLMs can't fully align with human preferences using reward-based approaches due to Condorcet cycles, but mixed strategies are statistically possible.
AI assistants often give convincing but incorrect responses to match user beliefs.
problem Sycophancy in AI assistants that use human feedback.
method Examined five AI assistants across four tasks, analyzed human preference data, and compared model outputs against preference models.
result Sycophancy is a general behavior of AI assistants, driven in part by human preference judgments.
Direct Density Ratio Optimization aligns LLMs with human preferences without assuming specific models.
problem Statistical inconsistency in aligning LLMs with human preferences.
method Direct Density Ratio Optimization (DDRO) estimates density ratio directly.
result DDRO is statistically consistent, converging to true human preferences as data grows.
Stable and consistent model alignment for language models without assuming human preference models.
problem Lack of statistical consistency in existing alignment methods.
method Relative density ratio optimization between preferred and mixture of preferred and non-preferred data distributions.
result Our approach achieves statistical consistency and stability, providing tighter convergence guarantees.
Develops a statistical framework to measure uncertainty in model rankings based on human preferences.
problem Uncertainty in model rankings based on human preferences due to mismatch between human and model preferences.
method Statistical framework using pairwise comparisons by humans and models to provide rank-sets for each model.
result Rank-sets constructed using only pairwise comparisons by strong models often do not cover the true ranking of human preferences.
Enhances preference learning by incorporating response times into binary choices.
problem Limited information from binary choices about preference strength.
method Combines choices and response times using the EZ diffusion model.
result Response times improve utility estimation for strong preferences.
Many real-world engineering problems rely on human preferences to guide their design and optimization. We present PrefOpt, an open source package to simplify sequential optimization tasks that incorporate human preference feedback. Our approach extends an existing latent variable model for binary preferences to allow f…
The paper explores game-theoretic alignment of LLMs with human preferences, finding limitations and conditions.
problem Aligning LLMs with human preferences using game theory.
method Systematic study of payoff choices in a two-player zero-sum game for desirable alignment properties.
result Impossibility of preference matching in game-theoretic LLM alignment under standard assumptions.
Efficiently identifies good policies by choosing contexts for human feedback.
problem Efficiently acquiring human feedback for preference alignment in large language models.
method Formalizes active exploration as a dueling bandit problem and proposes an active exploration algorithm with a polynomial worst-case regret bound.
result Proposed method outperforms baselines with limited human preferences on various language models and datasets.
A new approach to fine-tuning LLMs with human feedback.
problem Inability of current reward models to fully represent human preferences.
method Introducing NLHF, a new pipeline for LLM fine-tuning using pairwise human feedback.
result NLHF produces a sequence of policies converging to the regularized Nash equilibrium.
This work compares human feedback methods for reward learning in bandits.
problem Understanding how human feedback affects the performance of reward learning methods.
method Theoretical comparison of human feedback approaches in offline contextual bandits.
result Human bias and uncertainty in feedback modeling impact the theoretical guarantees of reward learning methods.
MallowsPO enhances LLM fine-tuning with a dispersion index of human preferences.
problem Lack of diversity in human preferences in DPO.
method Developed a dispersion index based on Mallows' theory to characterize preference diversity.
result Demonstrated improved performance in various tasks using the dispersion index.
SLHF uses sequential game theory to optimize preferences from human feedback.
problem Optimizing preferences from human feedback in sequential settings.
method SLHF frames the problem as a sequential-move game between Leader and Follower, decomposing the optimization into refinement and adversarial optimization.
result SLHF achieves strong alignment across diverse preference datasets and scales to large models.
Paper proposes a new RLHF framework for human preference learning.
problem Handling dependent online human preference outcomes with dynamic contexts.
method Two-stage algorithm with ε-greedy followed by exploitation; anti-concentration inequalities and matrix martingale concentration techniques. result Our method achieves optimal regret bound and asymptotic normality of estimators.
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.
The paper tackles statistical and computational challenges in learning correlated reward models.
problem The Independence of Irrelevant Alternatives (IIA) assumption collapses human preferences into a universal utility function, leading to coarse approximations.
method The paper investigates the statistical and computational challenges of learning a correlated probit model using best-of-three preference data.
result Best-of-three preference data overcomes the limitations of pairwise preference data, allowing for more fine-grained modeling of human preferences.
This paper improves sample efficiency for off-policy evaluation with preference data.
problem Improving sample efficiency for off-policy evaluation with preference data.
method Using a deep neural network to learn the value function and leveraging manifold structure.
result Established a provably efficient guarantee for off-policy evaluation with RLHF.
Learning the preferences of a human improves the quality of the interaction with the human. The number of queries available to learn preferences maybe limited especially when interacting with a human, and so active learning is a must. One approach to active learning is to use uncertainty sampling to decide the informat…
Proposes a robust algorithm for aligning large language models with human preferences.
problem Misspecification in preference models, reference policies, and reward functions.
method Doubly robust preference optimization algorithm.
result Superior and more robust performance compared to state-of-the-art algorithms.
Study tackles RLHF with diverse human feedback, showing limitations and proposing a meta-learning approach.
problem Traditional RLHF fails to balance diverse human preferences.
method Integrates meta-learning and multiple social welfare functions to optimize diverse preferences.
result Establishes sample complexity bounds for optimizing diverse social welfare functions.
New RLHF framework handles general preference oracles without reward functions.
problem Handling general preference oracles without assuming a reward function.
method Developed a minimax game between two LLMs for RLHF under a general preference oracle, focusing on KL-regularized preference.
result Proposed algorithms for efficient offline and online RLHF learning.
Bal-PM reduces preference labeling costs for LLMs.
problem Efficiently acquiring human feedback for preference modeling in large language models.
method Bayesian Active Learning with entropy maximization in feature space.
result Bal-PM reduces the number of required preference labels by 33% to 68%.
New RLHF approach mitigates bias in aligning LLMs with human preferences.
problem Algorithmic bias in RLHF leading to preference collapse.
method Preference Matching (PM) RLHF, using PM regularizer and conditional variant.
result 29% to 41% improvement in alignment with human preferences.
Active learning framework for optimizing human preferences in reinforcement learning.
problem Selecting most informative feedback for training models of human preferences.
method Proposes an active learning framework to collect preferential feedback online or offline.
result Errors in DPO logit estimates diminish with more feedback.
Derives a new objective to learn from human preferences without approximations.
problem Learning from human preferences through RLHF relies on approximations that can lead to pitfalls.
method Derives a new general objective ΨPO that bypasses both approximations. result Demonstrates the superiority of the new objective to Direct Preference Optimisation (DPO) empirically.
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.
Dropping a tiny fraction of preferences can significantly alter the rankings of top LLMs.
problem Robustness of LLM ranking systems to small changes in preference data.
method A computational method based on the Bradley-Terry model to evaluate robustness.
result Top LLM rankings can be highly sensitive to the removal of a small fraction of preferences.
ACOL learns constraints from human preferences in driving simulations.
problem Learning constraints from human preferences in driving simulations.
method Adaptive Constraint Learning (ACOL) algorithm for constrained linear best-arm identification.
result ACOL's sample complexity matches worst-case lower bound and is significantly tighter in the average case.
Autonomous systems can substantially enhance a human's efficiency and effectiveness in complex environments. Machines, however, are often unable to observe the preferences of the humans that they serve. Despite the fact that the human's and machine's objectives are aligned, asymmetric information, along with heterogene…
Neural algorithms optimize arm selection with human preference feedback for complex reward functions.
problem Optimizing arm selection with noisy human preference feedback for complex, non-linear reward functions.
method Neural network to estimate reward function using preference feedback, upper confidence bound and Thompson sampling algorithms.
result Sub-linear regret guarantees for efficient arm selection in contextual dueling bandits.
Paper develops methods to optimize policies directly from human feedback without reward inference.
problem Challenges in RLHF, including reward model overfitting and distribution shift.
method Develops two algorithms for RLHF without reward inference, using zeroth-order gradient approximators.
result Establishes polynomial convergence rates and outperforms existing methods in numerical experiments.
Efficiently learns reward functions with fewer queries and shorter computation times.
problem Expensive data generation and labeling in robot learning.
method Batch active preference-based learning methods using determinantal point processes (DPP) and heuristic alternatives.
result Our batch active learning algorithm requires only a few queries and computes them in a short amount of time.
Active IRL selects optimal human demonstrations for learning AI preferences.
problem Costly human demonstrations in IRL for autonomous systems.
method Information-theoretic acquisition function for selecting informative human demonstrations.
result Efficiently reduces human effort in learning AI preferences.
Paper shows equivalence between two alignment methods and introduces a new algorithm.
problem Ensuring human alignment of large language models for useful, safe, and pleasant user experience.
method Introduces IPO-MD algorithm, showing equivalence between IPO and Nash-MD methods.
result Equivalence between IPO and Nash-MD methods proven when considering online version of IPO.
Efficiently poisons offline RLHF models by flipping preference labels.
problem Vulnerability of offline RLHF models to preference label flipping attacks.
method Developed two attack methods: BAL-A and BMP-A, solving a structured binary sparse approximation problem.
result Demonstrated that flipping one preference label induces a parameter-independent shift in the DPO gradient, enabling structured binary sparse approximation.
Robot motions in the presence of humans should not only be feasible and safe, but also conform to human preferences. This, however, requires user feedback on the robot's behavior. In this work, we propose a novel approach to leverage the user's brain signals as a feedback modality in order to decode the judgment of rob…
Meta-algorithm for efficient reinforcement learning from human preferences.
problem Learning from human preference comparisons in Markov decision processes.
method Randomized exploration and experimental design for batch comparison queries.
result Meta-algorithm achieves both regret and last-iterate guarantees with minimal preference queries.
Proposes a new theoretical framework for PbRL that requires less human feedback.
problem Lack of theoretical work capturing practical PbRL frameworks.
method Introduces a reward-agnostic PbRL framework that acquires exploratory trajectories before human feedback.
result Demonstrates improved sample complexity for learning optimal policies in linear and low-rank MDPs.
New method for efficient online exploration in RLHF reduces regret.
problem Efficiently collecting new preference data in RLHF to refine reward model and policy.
method Proposes a new exploration scheme that directs preference queries toward reducing uncertainty in reward differences most relevant to policy improvement.
result Establishes regret bounds of order T(β+1)/(β+2) for online RLHF, with polynomial scaling in all model parameters. Paper proposes efficient RLHF methods for LLMs using active queries.
problem Efficiently aligning LLMs with human preferences using RLHF.
method Formalizes RLHF as a dueling bandit problem, introduces APPO and ADPO algorithms.
result ADPO achieves similar performance to state-of-the-art methods with fewer queries.
Develops methods to correct bias in AI feedback for more accurate alignment.
problem Systematic bias in AI feedback compared to human labels.
method Two debiased alignment methods: DDPO and DIPO.
result Methods improve alignment efficiency and performance close to human-labeled data.
New algorithm learns human preferences from few comparisons efficiently.
problem Learning human preferences from limited comparison feedback.
method Formulated as D-optimal design for Plackett-Luce model, solved using randomized Frank-Wolfe algorithm.
result Proposed algorithm efficiently solves D-optimal design problem for Plackett-Luce objective.
Study preference-based reinforcement learning in episodic kernel MDPs.
problem Learning from episodic human preferences in reinforcement learning.
method Developed preference-based value estimation and confidence sets for kernel-based MDPs.
result Proved high-probability regret bounds that converge to optimal policy value.
SPPO optimizes language model alignment by treating preferences as a game and achieving state-of-the-art performance.
problem Capturing intransitivity and irrationality in human preferences for accurate language model alignment.
method Self-play-based approach to identify Nash equilibrium policy through iterative policy updates.
result SPPO achieves state-of-the-art win-rate of 28.53% on AlpacaEval 2.0 without external supervision.
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.
A Markov Chain approach for aligning generative models from pairwise human preferences.
problem Aligning generative models from pairwise human preferences.
method Markov Chain from Human Feedback (MCHF)
result MCHF converges geometrically fast to the stationary distribution.
Optimizes AI learning with limited human feedback budgets.
problem Optimizing allocation of a fixed annotation budget for AI learning.
method Preference-Calibrated Active Learning (PCAL) using semi-parametric inference.
result Proves asymptotic optimality and robustness of the PCAL estimator.