BRPO optimizes batch RL policies to better exploit state-action differences.
problem Batch RL's conservatism limits exploitation of state-action differences.
method Proposes residual policies and derives BRPO to maximize policy performance.
result BRPO achieves state-of-the-art performance in various tasks.
We make policy optimization algorithms batch size-invariant by decoupling proximal and behavior policies.
problem Some policy optimization algorithms do not have batch size-invariance, leading to inefficiencies.
method We decouple the proximal policy from the behavior policy to achieve batch size-invariance.
result Our approach makes policy optimization algorithms more efficient and allows them to use stale data more effectively.
This paper benchmarks batch RL algorithms on Atari, finding DQN and partially-trained policies perform best.
problem Deep RL algorithms fail in batch setting.
method Benchmarked batch RL algorithms on Atari using a single partially-trained policy.
result Many batch RL algorithms underperform DQN and partially-trained policies.
The paper proposes an iterative approach to batch reinforcement learning for safer and more informative data collection.
problem Learning policies that are too rigid and do not adapt to new data.
method Safe diversified model-based policy search in an iterative batch reinforcement learning framework.
result Improved learned policies through continuous data collection and adaptation.
Researchers develop methods to protect reinforcement learning and control systems from policy poisoning attacks.
problem Attacks on reinforcement learning and control systems that manipulate learned policies.
method Unified framework for solving policy poisoning attacks, demonstrating global optimality and feasibility.
result Policy poisoning attacks are feasible and can be defended with a convex optimization approach.
Solves batch policy learning with constraints using flexible meta-algorithm and OPE.
problem Efficiently use pre-collected behavior data and mediate among competing objectives and constraints.
method Flexible meta-algorithm with any batch RL and online learning subroutines, specific instantiation, and OPE method.
result Achieves strong empirical results and OPE performance in various domains, including car driving.
ABS dynamically adjusts batch size based on policy stability, improving RL performance.
problem Diminishing returns with large batch sizes in RL due to non-stationary data.
method Adaptive Batch Scaling (ABS) with Behavioral Divergence metric.
result Larger batch sizes can improve RL performance, contrary to conventional wisdom.
The paper tackles batch policy learning in Markov Decision Processes, focusing on average reward maximization.
problem Maximizing long-term average reward in Markov Decision Processes with batch learning.
method Doubly robust estimator for average reward, optimization algorithm for optimal policy, finite-sample regret guarantee.
result The proposed method achieves semiparametric efficiency and provides a finite-sample regret guarantee.
Study model selection in batch policy optimization with three error sources.
problem Learn a policy competitive with the best model class in batch policy optimization.
method Formalized in contextual bandit setting with linear model classes, addressing approximation error, statistical complexity, and dataset shift.
result No algorithm can optimally trade-off all three error sources, but relaxing any one enables near-oracle inequalities for the others.
A novel approach learns goal-conditioned policies for locomotion using batch RL.
problem Training goal-conditioned policies for rotation invariant locomotion.
method Data augmentation and Siamese framework for invariance.
result Our approach outperforms existing RL algorithms on 3D locomotion agents.
New algorithm learns expert reward structures from batch data.
problem Learning expert reward structures from batch data without dynamics models.
method Deep Successor Feature Networks (DSFN) and transition-regularized imitation network.
result Superior performance on control benchmarks and sepsis management.
New batch RL method avoids overly optimistic policies.
problem Challenges in applying RL to large domains with limited data.
method Modified Bellman optimality and evaluation back-up for conservative updates.
result Can find approximately best policies within explored state-action space.
Many practical applications of reinforcement learning constrain agents to learn from a fixed batch of data which has already been gathered, without offering further possibility for data collection. In this paper, we demonstrate that due to errors introduced by extrapolation, standard off-policy deep reinforcement learn…
Sparse feature selection improves batch RL efficiency.
problem High-dimensional batch RL with many features.
method Sparse linear function approximation, Lasso, group Lasso, fitted Q-evaluation, fitted Q-iteration.
result Sparse feature selection makes batch RL more sample efficient.
We analyze how batch learning impacts bandit problems.
problem Impact of batch learning in stochastic bandits.
method Policy-agnostic regret analysis, upper and lower bounds demonstration.
result The impact of batch learning can be measured in terms of online behavior.
BAIL learns high-performing policies from batch data without additional interactions.
problem Learning high-performing policies from batch data without additional interactions.
method BAIL learns a V function, selects high-performing actions, and trains a policy network using those actions.
result BAIL outperforms other batch Q-learning and imitation-learning schemes in MuJoCo benchmark.
Paper proposes a method to improve off-policy reinforcement learning in batch settings.
problem Challenges in applying off-policy reinforcement learning to batch data.
method Uses a learned prior, the advantage-weighted behavior model (ABM), to bias RL policies.
result Improves performance on various RL tasks, including robot control.
A new algorithm learns policies from batch data in hierarchical RL.
problem Learning policies from fixed batches of data without full exploration.
method Modeling RL as a two-player game with a leader-follower structure, proposing StackelbergLearner.
result StackelbergLearner achieves competitive performance in batch RL and real-world datasets.
New algorithm learns optimal policy with multi-step lookahead information.
problem Learning optimal policy in reinforcement learning with multi-step lookahead information is NP-hard.
method Adaptive batching policies that process lookahead in state-dependent chunks.
result Order-optimal regret bounds up to a constant factor of lookahead horizon.
New algorithm for sequential off-policy learning improves performance over batch methods.
problem Training policies from logged interaction data in a sequential setting.
method Combines Logarithmic Smoothing with online PAC-Bayesian tools.
result Improves performance and accelerates convergence in sequential off-policy learning.
New RL algorithm learns good actions from offline data, reducing uncertainty and divergence.
problem Limited applicability of current RL algorithms in real-world settings due to high costs of exploration.
method Proposes an algorithm for batch RL using a fixed offline dataset, with penalties for policy and value constraints.
result Compared favorably to state-of-the-art methods on 32 continuous-action benchmarks.
When learning from a batch of logged bandit feedback, the discrepancy between the policy to be learned and the off-policy training data imposes statistical and computational challenges. Unlike classical supervised learning and online learning settings, in batch contextual bandit learning, one only has access to a colle…
Improved TD learning reduces batch sampling error.
problem TD learning can be inaccurate in batch settings due to action weighting.
method Introduced PSEC-TD(0) to correct action weighting using importance sampling.
result PSEC-TD(0) produces more accurate value function estimates.
Novel RL algorithms learn from human interaction data without exploration.
problem Efficiently learning from off-policy data in reinforcement learning.
method Developed off-policy batch RL algorithms using KL-control and dropout-based uncertainty.
result Successfully learned multiple reward functions from human interaction data.
Efficient RL algorithms for linear function approximation with limited adaptivity constraints.
problem Limited adaptivity in reinforcement learning with linear function approximation.
method Proposed two efficient online RL algorithms for episodic linear Markov decision processes under batch learning and rare policy switch models.
result Achieved efficient regret bounds for both batch learning and rare policy switch models, with substantial reduction in adaptivity.
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.
Study nonparametric contextual bandits with batched updates, achieving optimal regret.
problem Optimal regret in nonparametric contextual bandits with batch constraints.
method Dynamic binning of covariate space, optimal regret achieved.
result Achieves optimal regret (up to logarithmic factors) for nonparametric contextual bandits.
Batch Reinforcement Learning (RL) algorithms attempt to choose a policy from a designer-provided class of policies given a fixed set of training data. Choosing the policy which maximizes an estimate of return often leads to over-fitting when only limited data is available, due to the size of the policy class in relatio…
This paper unifies three regularization methods in batch reinforcement learning.
problem Learning overly-complex models in batch reinforcement learning.
method Unified weighted average transition matrix framework for three regularization methods.
result Empirical evaluation confirms intuitions about regularization methods' performance.
BaSE policy optimizes multi-armed bandits with batched data.
problem Optimizing multi-armed bandits with batched data.
method BaSE (batched successive elimination) policy for batched multi-armed bandits.
result Achieves rate-optimal regrets with adaptive batch sizes.
Active search is a learning paradigm for actively identifying as many members of a given class as possible. A critical target scenario is high-throughput screening for scientific discovery, such as drug or materials discovery. In this paper, we approach this problem in Bayesian decision framework. We first derive the B…
A new method for learning policies from demonstrations without reinforcement.
problem Learning policies from demonstrations without access to reinforcement signals.
method Energy-based distribution matching (EDM) to learn policy parameters and state marginals.
result EDM yields consistent performance gains over existing algorithms for strictly batch imitation learning.
STORM-PG uses momentum for faster policy gradient updates.
problem Improving policy gradient methods for reinforcement learning.
method Introduces STORM-PG, a SARAH-based algorithm with exponential moving average.
result Achieves O ( 1 / ε 3 ) O(1/ε^3) O ( 1/ ε 3 ) sample complexity, matching best-known rate. Batched Neural Bandits reduces policy updates in sequential decision-making.
problem Sequential decision-making with batched policy changes.
method BatchNeuralUCB algorithm combining neural networks and optimism.
result Achieves similar regret as fully sequential version with fewer policy updates.
We study reinforcement learning of chatbots with recurrent neural network architectures when the rewards are noisy and expensive to obtain. For instance, a chatbot used in automated customer service support can be scored by quality assurance agents, but this process can be expensive, time consuming and noisy. Previous …
Paper addresses OPE for dependent bandit samples using MDS and batch updates.
problem Evaluating policies from non-i.i.d. historical data in contextual bandits.
method Constructs an MDS-based estimator for dependent samples, solves batch update and deficient support issues.
result Derives an asymptotically normal estimator for evaluation policy value.
LF-IBIS learns optimal policies online without explicit likelihood.
problem Bayesian RL challenges due to intractable likelihood functions.
method Combines ABC with IBIS for online belief updates.
result Approximates posterior distributions for policies and parameters.
The paper proposes a method to explain expert decisions by modeling preferences with 'what if' outcomes.
problem Interpreting and auditing decision-making policies in institutions.
method Integrating counterfactual reasoning into batch inverse reinforcement learning.
result The method effectively recovers accurate and interpretable descriptions of expert behavior.
We develop an off-policy actor-critic algorithm for learning an optimal policy from a training set composed of data from multiple individuals. This algorithm is developed with a view towards its use in mobile health.
GN algorithm solves batched bandit for nondegenerate functions near-optimally.
problem Batched bandit learning for nondegenerate functions.
method Introduces Geometric Narrowing (GN) algorithm with a O ~ ( A + d T ) \widetilde{\mathcal{O}} ( A_{+}^d \sqrt{T} ) O ( A + d T ) regret bound and O ( log log T ) \mathcal{O} (\log \log T) O ( log log T ) batches. result GN achieves near optimal regret with minimal number of batches.
Critic-regularized regression improves offline RL performance.
problem Poor performance of off-policy algorithms in offline RL.
method Critic-regularized regression (CRR) for policy learning from fixed datasets.
result CRR outperforms state-of-the-art offline RL algorithms significantly.
Most prior work on active learning of classifiers has focused on sequentially selecting one unlabeled example at a time to be labeled in order to reduce the overall labeling effort. In many scenarios, however, it is desirable to label an entire batch of examples at once, for example, when labels can be acquired in para…
MBML improves multi-task RL by inferring task identity from state-action pairs.
problem Multi-task batch reinforcement learning with unseen tasks.
method MBML uses triplet loss and relabeling to robustify task inference.
result Significantly faster convergence on unseen tasks compared to random initialization.
Proposes a method to imitate active learning heuristics for better performance.
problem The performance of active learning heuristics depends on the classifier model and data structure.
method Imitates the selection of the best active learning heuristic using DAGGER.
result Outperforms state-of-the-art imitation learners and heuristics on well-known datasets.
A new approach to reinforcement learning improves policy performance by adjusting control frequency.
problem Improving reinforcement learning performance by optimizing control frequency.
method Introducing action persistence and a novel algorithm, PFQI, to learn optimal value function at a given persistence.
result PFQI effectively learns optimal value function with action persistence, improving reinforcement learning performance.
This paper considers Safe Policy Improvement (SPI) in Batch Reinforcement Learning (Batch RL): from a fixed dataset and without direct access to the true environment, train a policy that is guaranteed to perform at least as well as the baseline policy used to collect the data. Our approach, called SPI with Baseline Boo…
Batch Thompson Sampling reduces exploration-exploitation trade-off in online decision making.
problem Balancing exploration and exploitation in online decision making.
method Introducing a batch Thompson Sampling framework for stochastic multi-arm bandit and linear contextual bandit problems.
result Achieves asymptotic regret bound with O ( log T ) O(\log T) O ( log T ) batch queries, significantly reducing interactions. We study the problem of multi-agent reinforcement learning (MARL) with adaptivity constraints -- a new problem motivated by real-world applications where deployments of new policies are costly and the number of policy updates must be minimized. For two-player zero-sum Markov Games, we design a (policy) elimination base…