Study on tracking preference shifts in dueling bandits problems.
problem Tracking significant preference shifts in dueling bandits problems.
method Analysis of dueling bandits with distribution shifts, focusing on significant shifts (Suk and Kpotufe, 2022).
result Design of adaptive algorithms with O ( K i l d e L T ) O(\sqrt{K ilde{L}T}) O ( K i l d e L T ) dynamic regret for certain preference distribution classes. A new policy for contextual bandits adapts to reward vector shifts.
problem Learning under reward vector shifts with ordered rewards.
method Adaptive-discretization and optimistic elimination policy.
result Established upper bounds on preference-based regret.
Training models to prefer certain responses can unintentionally shift probability to harmful ones.
problem Likelihood displacement in DPO models, leading to unintended unalignment.
method Characterized and mitigated likelihood displacement using CHES score.
result Training models to prefer certain responses can unintentionally shift probability mass to harmful responses.
LoCo-RLHF models diverse human feedback with contextual information.
problem Heterogeneous human feedback from diverse contexts and preferences.
method Low-rank contextual preference model, PRS policy.
result LoCo-RLHF achieves tighter sub-optimality gap than existing methods.
Theory for RLHF generalization under reward shift and clipped KL.
problem Theoretical understanding of RLHF generalization, especially with reward shift and clipped KL.
method Developed generalization theory for RLHF, accounting for reward shift and clipped KL.
result Presented generalization bounds for RLHF, suggesting generalization error from sampling, reward shift, and KL clipping.
Federated learning method improves covariate shift adaptation for missing target values.
problem Missing target values in federated learning.
method Federated covariate shift adaptation algorithm for missing target output values.
result Asymptotically unbiased and efficient algorithm for federated learning.
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.
New method adapts to structural shifts in graph data for better label prevalence estimation.
problem Structural shifts in graph data affect label prevalence estimation.
method Importance sampling variant of KDEy quantification approach.
result Adapts to structural shifts and outperforms standard approaches.
Citizen science projects are successful at gathering rich datasets for various applications. However, the data collected by citizen scientists are often biased --- in particular, aligned more with the citizens' preferences than with scientific objectives. We propose the Shift Compensation Network (SCN), an end-to-end l…
New method encodes function preferences into neural nets for better generalization.
problem Challenges in encoding explicit function preferences in neural network training.
method Function-space empirical Bayes (FSEB) regularization.
result FSEB leads to near-perfect semantic shift detection and improved generalization.
LLMs prefer Bitcoin under crisis frames, affecting financial decisions.
problem Testing whether LLMs have built-in biases towards specific financial assets.
method Developed a three-level audit protocol to examine Bitcoin's representation and influence in LLMs.
result An identifiable internal feature in LLMs can be perturbed to move financial choices, but only within measurable limits.
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.
Proposes robust assortment optimization from observational data.
problem Real-world scenarios often violate assumptions of stable customer preferences and correct choice models.
method Develops a robust framework that accounts for potential distributional shifts in customer choice behavior.
result Uncovered the notion of ``robust item-wise coverage'' as the minimal data requirement for sample-efficient robust assortment learning.
The paper analyzes how mutable blockchain protocols affect miner behavior and strategic stability.
problem The mutability of blockchain protocols undermines long-term planning and cooperative equilibria.
method Integrates Austrian capital theory with repeated game theory to examine miner behavior under different institutional conditions.
result Effective time preference increases when protocol rules are mutable, leading to political rent-seeking and undermining strategic coherence.
The paper proposes a method to learn and leverage contextual preference distributions for better decision-making.
problem Heterogeneous and context-dependent human preferences in decision-making problems.
method A sequential learning-and-optimization pipeline using a bounded-variance score function gradient estimator to train a predictive model mapping contextual features to preference distributions.
result The approach reduces average post-decision surprise by up to 25 times compared to risk-averse baselines in a ridesharing environment.
New method improves consistency in preference learning for neural networks.
problem Inconsistent surrogate losses in preference learning for neural networks.
method Formulated a margin-shifted ranking framework and introduced Structure-Aware H H H -consistency. result Proved superior consistency guarantees for capacity-bounded models using heavy-tailed surrogates.
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.
Bayes predictor remains robust to ignorable missingness shifts.
problem Challenges in prediction with missing covariates and shifts in missingness reasons.
method Bayesian approach and different prediction methods.
result Bayes predictor remains unchanged by ignorable shifts, but robust prediction requires disregarding missingness for non-ignorable shifts.
New framework shifts bandit algorithms from expected reward to preference metrics, optimizing mixtures of arms.
problem Traditional bandit algorithms focus on expected rewards, ignoring variability and risk.
method Introduces preference metrics (PMs) and designs algorithms to optimize mixtures of arms.
result Optimal policy selects mixtures of arms based on specific weights, not a single best arm.
Proposes a dynamic matching algorithm for two-sided online markets.
problem Dynamic preferences in two-sided online matching platforms.
method Dynamic Matching Bandit Algorithm with statistical preference ranking estimation.
result Agent-optimal stable matching result with logarithmic regret bound.
New method improves accuracy of quantized neural networks.
problem Accuracy drop in quantized neural networks, especially MobileNet family.
method Weight equalizing shift scaler, binary shifting to recover output range.
result Top-1 accuracy improved from 0.1% to 69.78% ~ 70.96% in MobileNets.
OpFlow predicts robust OD flows by learning choice potentials conditioned on spatial exposures.
problem Deep models trained on raw counts are vulnerable to distribution shift.
method OpFlow learns row-centered choice potentials and reconstructs flows by combining them with a calibrated origin scale.
result OpFlow improves robustness under environment shifts, as shown by controlled synthetic shifts and a real-world experiment.
RAVEN improves weak-to-strong generalization under distribution shifts.
problem Weak models fail to supervise strong models effectively under distribution shifts.
method RAVEN dynamically learns optimal combinations of weak models and strong model parameters.
result RAVEN outperforms existing methods by over 30% on out-of-distribution tasks.
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 ) T^{(β+1)/(β+2)} T ( β + 1 ) / ( β + 2 ) for online RLHF, with polynomial scaling in all model parameters. Novel hyperparameter optimization for target tasks under covariate shift.
problem Hyperparameter optimization under multi-source covariate shift.
method Construct variance reduced estimator to unbiasedly approximate target objective; propose no-regret hyperparameter optimization procedure.
result Proposed framework broadens applications of automated hyperparameter optimization.
Automates debiasing for large language model evaluations through Fisher random walk.
problem Rigorous and scalable evaluation of large language models.
method Semiparametric efficient estimator using Fisher random walk for weighted residual balancing.
result Efficient estimation of contextual preference scores for large language models.
This study develops a dynamic inverse optimization framework to recover hidden, time-varying preferences from observed allocation trajectories.
problem The gap between classical optimization theory and real-world practice, especially in the presence of drift and shocks.
method Dynamic inverse optimization framework using a drift-aware estimator grounded in convex analysis and online learning theory.
result Sharp static and dynamic regret bounds for the framework, demonstrating its responsiveness to gradual drift and sudden shocks.
The paper models market dynamics using a limit order book system to explain slippage and inefficiency.
problem Inefficiency in matching markets due to structural liquidity constraints and slippage.
method Introduces a market microstructure framework with a latent preference state matrix and a dynamic discrete choice execution model.
result Persistent slippage and regional invariance of preference orderings are explained by liquidity thresholds.
This work analyzes IRM and ERM from sample complexity perspective, revealing different behaviors under various distribution shifts.
problem Choosing between IRM and ERM for OOD generalization.
method Sample complexity analysis comparing IRM and ERM under different data generation mechanisms.
result IRM is preferred over ERM for certain distribution shifts, leading to better OOD generalization.
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.
New term ADS describes how machine learning can change user behavior.
problem Machine learning systems can unintentionally change user behavior, affecting performance.
method Introduced `unit tests` and mitigation strategy for hidden incentives in auto-induced distributional shift.
result Meta-learning and Q-learning sometimes fail unit tests but pass with mitigation strategy.
Q-SAVI model improves drug discovery accuracy with prior knowledge of chemical space.
problem Challenges in drug discovery due to covariate shift and limited labeled data.
method Probabilistic model with domain-informed prior distributions over functions.
result Q-SAVI outperforms state-of-the-art techniques in predictive accuracy and calibration.
SAIL improves online alignment of large language models with minimal feedback.
problem Offline RLHF methods often lead to sub-optimal performance due to fixed preference datasets.
method SAIL uses bilevel optimization and a single-level first-order method to iteratively refine model alignment.
result SAIL significantly improves alignment performance on open-sourced datasets with minimal computational overhead.
Randomly initialized transformers show extreme token preferences.
problem Structural biases in randomly initialized transformers.
method Dissection of transformer architecture at initialization.
result Initialization-induced biases persist throughout training.
DRO-REBEL improves LLM alignment by robustly updating models online.
problem Overfitting and drifting of LLMs during RLHF.
method DRO-REBEL uses type- p p p Wasserstein, KL, and χ 2 χ^2 χ 2 ambiguity sets for robust online updates. result DRO-REBEL achieves faster convergence and better performance than prior methods.
The sharp decline in Ireland's economic performance in recent years has coincided with a recent fall in asylum applications. Simultaneously countries such as Switzerland are seeing increases in asylum numbers with evidence for greater numbers of Nigerian applicants, a group that have for some time been the largest nati…
Gradient equilibrium improves online learning performance without requiring sublinear regret.
problem Achieving sublinear regret in online learning.
method Gradient equilibrium: average of gradients converges to zero.
result Gradient equilibrium can be achieved by standard online learning methods.
Foreign policy analysis has been struggling to find ways to measure policy preferences and paradigm shifts in international political systems. This paper presents a novel, potential solution to this challenge, through the application of a neural word embedding (Word2vec) model on a dataset featuring speeches by heads o…
Domain Adaptation in 6G wireless networks: When is it green?
problem Energy consumption of Domain Adaptation (UDA) compared to single-task training in 6G wireless networks.
method Investigate energy consumption and propose a method to determine the minimum number of target domains for UDA to be more energy-efficient than retraining.
result Proposed a method to determine the minimum number of target domains for UDA to be more energy-efficient than retraining.
ADPO optimizes relative advantage in reinforcement learning from human feedback.
problem Optimizing policy alignment in reinforcement learning from human preferences.
method ADPO explicitly parameterizes the optimal structure through anchored logits, decoupling response quality from prior popularity.
result Empirically, ADPO achieves state-of-the-art performance on reasoning tasks, outperforming GRPO by 30.9 percent.
Adaptive algorithms minimize regret in matching markets with contextual arm preferences.
problem Minimizing regret in matching markets with context-dependent player utilities.
method Developed adaptive algorithms for stochastic and adversarial contexts, providing upper and lower bounds.
result Achieved sublinear regret bounds for both stochastic and adversarial contexts.
This paper improves recommender systems by handling dynamic user preferences and item popularity.
problem Dynamic user preferences and changing item popularity in recommender systems.
method Developed a Thompson sampling-based policy for a high-dimensional linear bandit problem, reducing feature vector dimensionality and using exponentially increasing weights.
result Proved a regret bound that scales with the reduced dimension, demonstrating effectiveness in trade-off between computational complexity and regret performance.
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.
Paper improves recommendation systems by optimizing sequence of items for clicks.
problem Improving recommendation systems robustness against bots and clicks.
method Minimizing pairwise ranking loss over sequences of items, with thresholds to prevent bot influence.
result The proposed algorithms converge and outperform existing methods in various ranking measures.
Python library for causal discovery from observational data.
problem Revealing causal relations from observational data.
method Comprehensive collection of causal discovery methods in Python.
result Ease of use for non-specialists and modular building blocks for developers.
Optimizes molecular generation for chemist preferences.
problem Models lack inherent preferences for chemist-desired structures.
method Fine-tuning with Direct Preference Optimization.
result Approach is simple, efficient, and highly effective.
New method adapts to user preferences dynamically, improving recommendation models.
problem Current recommendation models lack dynamic adaptation to changing user preferences.
method Preference Discerning with LLM-Enhanced Generative Retrieval
result Mender achieves state-of-the-art performance in adapting to evolving user 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…