A method for robust reinforcement learning in large state spaces.
problem Challenges in RL with large state spaces, costly data, and real-world dynamics deviation.
method Distributionally robust Markov decision processes with Gaussian Processes and maximum variance reduction.
result Efficient learning of multi-output nominal transition dynamics with statistical sample complexity bounds.
Risk-averse model uncertainty framework for safe reinforcement learning.
problem Safe decision making in uncertain environments.
method Risk-averse perspective towards model uncertainty using coherent distortion risk measures; equivalent to distributionally robust safe reinforcement learning problems; efficient, model-free implementation.
result Demonstrates robust performance and safety across perturbed test environments.
New algorithm improves RL performance across different environments.
problem Improving reinforcement learning performance across various environments.
method Designing a fully model-free DRRL algorithm that learns from a single trajectory.
result Demonstrates superior robustness and sample efficiency compared to existing methods.
New approach connects reinforcement learning robustness and regularization.
problem Dealing with external uncertainty in reinforcement learning.
method Introducing Wasserstein distributionally robust MDPs and a new regularizer.
result Established a dual relation between robust MDPs and regularization.
Develops robust learning framework under distributional perturbations.
problem Learning robust to data distributional changes.
method Distributionally Robust Optimization (DRO) under Wasserstein metric.
result Establishes performance guarantees and tractable formulations.
Improved reinforcement learning for environments with distributional shifts.
problem Learning optimal policies in environments with distributional shifts.
method Distributionally robust Q-learning with multi-level Monte Carlo estimator.
result Proved upper bound on sample complexity for robust RL.
Paper tackles robust reinforcement learning with minimal data.
problem Learning robust policies from limited data in uncertain environments.
method Distributionally robust formulation, model-based algorithm combining value iteration and pessimism.
result Proves near-optimal sample complexity for robust offline RL.
New algorithms tackle robust RL with linear models, revealing unique challenges.
problem Distributionally robust offline RL with uncertainty in dynamics.
method Proposes minimax optimal and computationally efficient algorithms using novel function approximation mechanisms.
result Function approximation in robust offline RL is distinct and harder than in standard offline RL.
This paper tackles RL issues with robust policies using historical data.
problem Limited data and mismatch between training and testing environments.
method Distributionally robust offline RL with linear function approximation.
result Achieved error bounds for sample complexity in RL.
Study improves reinforcement learning for stable long-term performance.
problem Distributionally robust average-reward reinforcement learning for stable long-term performance.
method Proposes two algorithms to achieve near-optimal sample complexity.
result Achieves a sample complexity of O ( ∣ S ∣ ∣ A ∣ t m i x 2 ε − 2 ) O(|\mathbf{S}||\mathbf{A}| t_{\mathrm{mix}}^2\varepsilon^{-2}) O ( ∣ S ∣∣ A ∣ t mix 2 ε − 2 ) for estimating optimal policy and robust average reward. Improved RL algorithm for robustness against parameter mismatches.
problem Learning robust control policies against parameter mismatches between training and testing environments.
method Formulated as DR-RL problem, proposed RPVL algorithm for tabular episodic learning with four divergences.
result Achieved i l d e O ( ∣ S ∣ ∣ A ∣ H 5 ) ilde{\mathcal{O}}(|\mathcal{S}||\mathcal{A}| H^{5}) i l d e O ( ∣ S ∣∣ A ∣ H 5 ) sample complexity uniformly better than existing results. New framework for robust reinforcement learning policies in uncertain environments.
problem Robust reinforcement learning policies in environments with distributional shifts.
method Comprehensive modeling framework centered around robust Markov decision processes (RMDPs).
result Existence and conditions for the dynamic programming principle (DPP) in RMDPs.
Paper studies S-rectangular DR-RL models for robust reinforcement learning with near-optimal sample complexity.
problem Addressing distributional discrepancies in reinforcement learning environments.
method Empirical value iteration algorithm for divergence-based S-rectangular DR-RL models.
result Near-optimal sample complexity bound of O ( ∣ S ∣ ∣ A ∣ ( 1 − γ ) − 4 ε − 2 ) O(|\mathcal{S}||\mathcal{A}|(1-γ)^{-4}\varepsilon^{-2}) O ( ∣ S ∣∣ A ∣ ( 1 − γ ) − 4 ε − 2 ) . Paper proposes a framework for reliable off-policy evaluation in reinforcement learning.
problem Quantifying uncertainty in off-policy estimates for safe deployment of target policies.
method Distributionally robust optimization for creating confidence bounds.
result Non-asymptotic and asymptotic guarantees for robust cumulative reward estimates.
New algorithms learn robust policies from shifted distributions.
problem Learning robust policies in environments with distributional shifts.
method Two novel model-free algorithms: distributionally robust Q-learning and variance-reduced distributionally robust Q-learning.
result Achieves minimax sample complexity upper bound of i l d e O ( ∣ S ∣ ∣ A ∣ ( 1 − γ ) − 4 ε − 2 ) ilde O(|\mathbf{S}||\mathbf{A}|(1-γ)^{-4}ε^{-2}) i l d e O ( ∣ S ∣∣ A ∣ ( 1 − γ ) − 4 ε − 2 ) . A new algorithm improves offline reinforcement learning robustness.
problem Finding optimal policies in perturbed environments from offline data.
method Doubly Pessimistic Model-based Policy Optimization (P^2MPO) framework.
result Proves sample efficiency with robust partial coverage data.
Study robust control for systems with continuous states using adversarial perturbations.
problem Fragile policies in Markov control models under internal or external perturbations.
method Distributionally robust stochastic control with adaptive adversarial perturbations.
result Optimal robust policies for continuous state systems with uniform learning guarantees.
This work tackles robust RL in multi-agent settings, improving sample efficiency.
problem Overcoming environmental uncertainties in multi-agent reinforcement learning.
method Proposes DRNVI, a sample-efficient algorithm for learning robust equilibria in RMGs.
result Establishes near-optimal sample complexity for solving RMGs.
We develop a robust RL algorithm for off-dynamics environments with improved suboptimality bounds and computational efficiency.
problem Learning policies robust to uncertainties in transition dynamics between training and deployment environments.
method Distributionally robust Markov decision processes (DRMDPs) with a novel algorithm We-DRIVE-U.
result Improved suboptimality bound of O ~ ( d H ⋅ min { 1 / ρ , H } / K ) \widetilde{\mathcal{O}}\big({d H \cdot \min \{1/ρ, H\}/\sqrt{K} }\big) O ( d H ⋅ min { 1/ ρ , H } / K ) , near-optimal up to O ( H ) \mathcal{O}(\sqrt{H}) O ( H ) . New RL method tackles sim-to-real gap using interactive data collection.
problem Sim-to-real gap in reinforcement learning.
method Distributionally robust reinforcement learning with interactive data collection.
result Proves sample-efficient learning is impossible without additional assumptions.
Develops ODRPO to improve RL algorithms with better performance and stability.
problem RL algorithms converge to sub-optimal solutions due to limited policy representation.
method Integrates DRO approach to solve trust region constrained optimization problem without parameterizing policies.
result Achieves globally optimal policy update and higher sample efficiency.
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.
Real-world applications require RL algorithms to act safely. During learning process, it is likely that the agent executes sub-optimal actions that may lead to unsafe/poor states of the system. Exploration is particularly brittle in high-dimensional state/action space due to increased number of low-performing actions. …
Combines adversarial and interventional robustness for machine learning models.
problem Designing robust models for distribution shifts in machine learning.
method RISe formulation using distributionally robust optimization.
result Demonstrates efficacy of RISe approach with synthetic and real-world datasets.
Study online RL with mismatched dynamics, achieving sublinear regret.
problem Exploration challenges in online RL with mismatched training and deployment dynamics.
method Introduce supremal visitation ratio, propose efficient algorithm with f f f -divergence. result Achieves sublinear regret in online RMDPs with optimal dependence on supremal visitation ratio and interaction episodes.
Proposes DRRO to mitigate over-optimization in RLHF from human feedback.
problem Over-optimization due to reward misspecification in RLHF.
method Wasserstein distributionally robust regret optimization (DRRO).
result DRRO mitigates over-optimization more effectively than existing baselines.
DR-RPO optimizes robust policies in RL with limited interaction, achieving sublinear regret.
problem Policy optimization in RL under distribution shift and adversarial dynamics.
method DR-RPO algorithm incorporating reference-policy regularization and upper confidence bonus for exploration.
result DR-RPO achieves sublinear regret and polynomial suboptimality bounds in robust RL.
New model-free DR-RL algorithm with finite sample complexity.
problem Limited model-free DR-RL methods with convergence guarantees or sample complexities.
method Integrates Multi-level Monte Carlo (MLMC) technique with threshold mechanism.
result First model-free DR-RL approach with finite sample complexity for total variation and Chi-square divergence.
New scalable methods for robust model learning from large datasets.
problem Training robust models resistant to data distribution shifts.
method Composite optimization for distributionally robust optimization (DRO).
result Scalable methods for learning robust models from large datasets.
New approach improves model generalization through distributionally robust learning.
problem Improving model generalization in machine learning.
method Stochastic gradient descent applied to the outer minimization problem, with gradient estimation through multi-level Monte Carlo randomization.
result Our approach yields significant benefits over previous work in numerical experiments.
New method improves adversarial robustness of neural networks.
problem Vulnerability of neural networks to adversarial attacks.
method Integrates φ-divergence regularization into distributionally robust risk function.
result Achieves near-optimal sample complexity with efficient stochastic gradient methods.
New regularizer for machine learning using private data.
problem Machine learning with private data.
method Distributionally-robust optimization with locally-differentially-private datasets.
result New regularizer for training linear regression models.
This work bridges offline RL and DRL to address distributional shift.
problem Distributional shift in offline RL due to difference in state-action visitation distributions.
method Proposes offline RL algorithms using DRL framework, characterizes sample complexity under single policy concentrability.
result Demonstrates superior performance of proposed algorithms through simulations.
STEEL tackles batch RL with singularity, improving policy optimization.
problem Existing RL methods assume absolutely continuous data, but STEEL handles non-overlapping regions.
method Proposes STEEL algorithm using maximum mean discrepancy and distributionally robust optimization.
result First finite-sample regret guarantee for batch RL with singularity.
The paper connects three machine learning methods to reduce generalization errors.
problem Reducing generalization errors in machine learning models.
method Distributionally robust optimization, Bayesian methods, and regularization.
result Machine learning models can be characterized using distributional uncertainty and robustness measures.
A general duality proof for Wasserstein distributionally robust optimization.
problem Optimizing under uncertainty with Wasserstein distance.
method One-dimensional convex analysis and interchangeability principle.
result General duality result holds for various distributions and costs.
Improved estimator reduces bias in statistical learning models.
problem Asymptotic bias in classic WDRO estimator.
method Adjusted Wasserstein distributionally robust estimator.
result Asymptotic unbiased estimator with smaller MSE.
Algorithm learns robust equilibrium in online Markov games with interactive data.
problem Sim-to-real gap in reinforcement learning.
method Distributionally robust RL with minimum value assumption, least square value iteration.
result Sample-efficient algorithm for robust equilibrium in online Markov games.
Paper develops a robust Bayesian optimization method for noisy zeroth-order settings.
problem Achieving robustness to distributional shift in machine learning.
method Distributionally robust Bayesian optimization (DRBO) algorithm for noisy zeroth-order optimization.
result DRBO algorithm provably obtains sub-linear robust regret in various settings.
Wasserstein distributionally robust optimization estimators are obtained as solutions of min-max problems in which the statistician selects a parameter minimizing the worst-case loss among all probability models within a certain distance (in a Wasserstein sense) from the underlying empirical measure. While motivated by…
Scaff-PD improves fairness and robustness in federated learning with reduced communication.
problem Improving fairness and robustness in federated learning with limited communication.
method Scaff-PD uses a family of distributionally robust objectives and an accelerated primal dual algorithm with bias-corrected steps.
result Scaff-PD achieves significant gains in communication efficiency and convergence speed while maintaining fairness and robustness.
Tikhonov regularization is robust under specific martingale constraints in distributionally robust optimization.
problem Distributionally robust optimization and regularization of learning models.
method Optimal transport approach with martingale constraints.
result Tikhonov regularization is optimal transport robust under specified martingale constraints.
Proposes using Wasserstein barycenters for robust optimization with multiple data sources.
problem Distributionally robust optimization with multiple heterogeneous data sources.
method Construct nominal distribution through Wasserstein barycenter of multiple data samples, reformulates as a finite convex program.
result Proposed scheme outperforms other estimators in sparse inverse covariance matrix estimation.
Adaptive optimal transport priors improve few-shot learning robustness.
problem Limited supervision and distribution shifts in few-shot learning.
method Prototype-Guided Distributionally Robust Optimization (PG-DRO) framework.
result PG-DRO achieves stronger robust generalization in few-shot scenarios.
Exact generalization guarantees for robust models using Wasserstein distance are established.
problem Capturing data uncertainty and distribution shifts in machine learning models.
method Establishes exact generalization guarantees for robust models based on the Wasserstein distance, covering various cases and transport costs.
result Exact generalization guarantees are provided for a wide range of cases, including deep learning objectives with nonsmooth activations.
Discuss new policy learning objectives and methods.
problem Improving policy learning efficiency and robustness.
method Introducing curvature considerations and calibration data methods.
result Efficient retargeting and distributionally robust policies.
DRDA robustly adapts models across domains with mismatched distributions.
problem Vulnerability of DA methods to noise and inability to generalize to unseen samples.
method DRDA uses distributionally robust optimization (DRO) with MMD metric to learn robust decision functions.
result DRDA outperforms existing robust learning approaches in experiments.
Extends ML fairness to handle minority groups over time.
problem Limitations of existing fairness criteria.
method Performative Distributionally Robust Optimization.
result Improves fairness for minority groups over time.