One-step policy improvement outperforms iterative RL methods on D4RL.
problem Improving offline RL without off-policy evaluation.
method One-step constrained/regularized policy improvement using on-policy Q estimates.
result One-step algorithm outperforms iterative algorithms on D4RL benchmark.
A key problem in reinforcement learning for control with general function approximators (such as deep neural networks and other nonlinear functions) is that, for many algorithms employed in practice, updates to the policy or Q-function may fail to improve performance---or worse, actually cause the policy performance …
A new algorithm improves policy discovery in uncertain environments.
problem Discovering optimal policies in uncertain environments.
method Bayesian least-squares policy iteration (BLSPI) with randomised variant (RBLSPI).
result RBLSPI effectively discovers optimal policies by quantifying uncertainty.
Paper proves exponential lower bounds for policy iteration in MDPs.
problem Proving lower bounds for policy iteration in multi-action MDPs.
method Generalized previous results to k-action MDPs, constructed families of MDPs.
result Proved novel exponential lower bound of (3+k)2^(N/2-3) iterations.
Combines soft greediness with Modified Policy Iteration for more efficient deep reinforcement learning.
problem Improving sample efficiency in deep reinforcement learning.
method Combines soft greediness with Modified Policy Iteration (MPI) for off-policy learning.
result The proposed algorithm is more sample efficient than original PPO and competitive with SAC.
RRPI improves offline RL by optimizing policies against worst-case dynamics.
problem Offline RL's performance degrades under distribution shift and transition uncertainty.
method Formulates offline RL as robust policy optimization, treating transition kernel as decision variable.
result RRPI achieves strong average performance on D4RL benchmarks, outperforming recent baselines.
Study shows PI for LQR requires fewer policy improvement steps than policy evaluation steps.
problem Understanding the sample complexity of RL algorithms for continuous control tasks.
method Finite-time analysis of approximate policy iteration for LQR, quantifying policy improvement and evaluation complexities.
result Policy evaluation is the dominant factor in sample complexity, requiring (n+d)3/ε2 samples per step. This paper improves self-play learning in games by manipulating experience distributions.
problem Improving self-play learning in games through better experience sampling.
method Three approaches: weighted sampling, Prioritized Experience Replay, and diversifying trajectories.
result Major improvements in early training performance in some games, minor improvements overall.
Lower bounds for PI on multi-action MDPs are established, showing complexity grows with action count.
problem Establishing the minimum number of iterations for PI to converge on MDPs with multiple actions.
method Developed lower bounds for a specific PI variant on multi-action MDPs, scaling with action count.
result A particular PI variant can take Ω(kn/2) iterations to terminate, scaling with action count. Normalizing flows policy improves trust region policy optimization.
problem Improving exploration and avoiding local optima in policy optimization.
method Constructing trust region with KL divergence constraints and using normalizing flows policy.
result Normalizing flows policy significantly improves policy optimization, especially on high-dimensional tasks.
Improved API to achieve optimal error bound and query complexity in local planning.
problem Efficient local planning in discounted MDPs with linear approximation.
method Confident Approximate Policy Iteration (CAPI) for stationary policies, applying to local access simulators.
result Achieves optimal accuracy and query complexity bounds, improving over API.
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.
New RL algorithms improve performance using multi-step greedy policies.
problem Improving model-free reinforcement learning performance.
method Developed multi-step greedy κ-Policy Iteration and κ-Value Iteration algorithms for model-free RL. result Multi-step greedy algorithms outperform DQN and TRPO on various benchmark tasks.
Recently, a novel class of Approximate Policy Iteration (API) algorithms have demonstrated impressive practical performance (e.g., ExIt from [2], AlphaGo-Zero from [27]). This new family of algorithms maintains, and alternately optimizes, two policies: a fast, reactive policy (e.g., a deep neural network) deployed at t…
This paper applies CPI to deep RL, improving stability and performance.
problem Improving stability and performance in deep reinforcement learning.
method Combines Conservative Policy Iteration with deep neural networks and adaptive mixture rates.
result Demonstrates improved stability and performance in deep RL algorithms.
NeuPL learns diverse policies in strategy games efficiently.
problem Iterative training of policies in strategy games leads to under-trained good-responses and wasteful repetition.
method NeuPL uses a single conditional model to represent a population of policies, offering convergence guarantees and transfer learning.
result NeuPL achieves better performance and efficiency across various domains, enabling access to novel strategies.
KL-constrained API shows optimization issues and improved with regularization.
problem Optimization issues in KL-constrained API algorithms.
method Comparison of KL divergence as a constraint vs. regularizer, empirical evaluation.
result KL-constrained API is not guaranteed to converge and incurs linear regret.
Accelerates policy optimization in RL with optimistic and adaptive updates.
problem Improving policy optimization methods in reinforcement learning.
method Integrates foresight into policy improvement step via optimistic and adaptive updates.
result Designs an optimistic policy gradient algorithm, adaptive via meta-gradient learning.
Mixed RL improves RL efficiency with dual representations.
problem Poor sampling efficiency in RL methods.
method Uses dual representations of environmental dynamics to search optimal policies.
result Proves convergence and recursive stability of the mixed RL.
PGS uses neural networks to improve policies online without search trees.
problem Limited scalability of Monte Carlo Tree Search (MCTS) for high branching factor games.
method Adapts a neural network simulation policy via policy gradient updates, avoiding search trees.
result PGS achieves comparable performance to MCTS and defeats strong Hex agents.
Improved Politex algorithm reduces regret bound to O(√T) with experience replay.
problem Learning in infinite-horizon MDPs with function approximation.
method Sharpened regret analysis of Politex algorithm, experience replay implementation.
result First high-probability O(√T) regret bound for computationally efficient algorithm.
DSPI connects natural policy gradient to policy iteration, proving global convergence.
problem Optimizing policies in reinforcement learning.
method DSPI framework, combining smoothed policy iteration and natural policy gradient.
result DSPI achieves geometric convergence and optimal complexity for policy optimization.
VA-OPE improves OPE by incorporating variance information, achieving tighter error bounds.
problem Estimating value function of a target policy from offline data collected by a behavior policy.
method Proposes VA-OPE, an algorithm that reweights Bellman residual using estimated variance of the value function.
result Achieves a tighter error bound than the best-known result.
Improves RL algorithms with two techniques.
problem Enhance off-policy RL performance.
method Formulates RL as proximal point iteration; uses value functions for improved action value estimate.
result Significant performance improvement on RL benchmarks.
MVPI framework optimizes risk in reinforcement learning, improving performance in robot simulations.
problem Optimizing risk in reinforcement learning control problems.
method Mean-Variance Policy Iteration (MVPI) framework for risk-averse control in MDPs.
result Risk-averse TD3 outperforms previous methods in robot simulation tasks.
A lifelong learning architecture improves reinforcement learning policies using simulations and a DNC model.
problem Improving reinforcement learning policies in dynamic environments.
method Iterative training of a Reinforcement Learning agent and a DNC model in conjunction.
result DNC models can continually learn from pixels alone to simulate new tasks.
In this paper we discuss policy iteration methods for approximate solution of a finite-state discounted Markov decision problem, with a focus on feature-based aggregation methods and their connection with deep reinforcement learning schemes. We introduce features of the states of the original problem, and we formulate …
Deep RL agents suffer from transient non-stationarity, which ITER mitigates.
problem Transient non-stationarity in deep RL agents affects generalization.
method Iterated Relearning (ITER) transfers knowledge between networks to reduce non-stationarity.
result ITER improves deep RL agents' performance on generalization benchmarks.
Paper proves convergence of optimistic policy iteration for stochastic shortest path problems.
problem Optimizing policies in stochastic shortest path problems.
method Analyzes optimistic policy iteration algorithm with Monte Carlo and TD(λ) methods.
result Proves convergence of the algorithm under specific conditions.
Policy gradient methods with aggregated states can achieve better performance than approximate policy iteration.
problem Approximation errors in policy and value function approximations.
method State-aggregated representations and policy gradient methods.
result Policy gradient methods can achieve a per-period regret bounded by ε, while approximate policy iteration and value iteration have a higher regret.
Paper proposes a method to learn goal-reaching behaviors from scratch using imitation learning.
problem Current reinforcement learning algorithms are brittle and require expert demonstrations.
method Iterated supervised learning where agents relabel and imitate generated trajectories.
result Improved goal-reaching performance and robustness over current RL algorithms.
New RL algorithms improve average-reward performance.
problem Improving RL algorithms for average-reward criteria.
method Developed novel algorithms addressing average-reward criterion directly.
result ATRPO significantly outperforms TRPO in challenging MuJuCo environments.
A new method improves stability in policy learning for continuous control tasks.
problem Stability issues in policy gradient methods when policies are close to deterministic.
method Target Distribution Learning (TDL) alternates between proposing a target distribution and training the policy network to approach it.
result TDL leads to more stable policy improvements over iterations compared to existing methods.
Adaptive optimal control of nonlinear dynamic systems with deterministic and known dynamics under a known undiscounted infinite-horizon cost function is investigated. Policy iteration scheme initiated using a stabilizing initial control is analyzed in solving the problem. The convergence of the iterations and the optim…
AAPI improves regret bound for undiscounted continuing learning in uniformly ergodic MDPs.
problem Improving regret bounds for undiscounted continuing learning in uniformly ergodic MDPs.
method Adaptive approximate policy iteration (AAPI) with online learning techniques and data-dependent adaptive learning rate.
result AAPI achieves a ildeO(T2/3) regret bound, improving over the best existing bound of ildeO(T3/4). The famous Policy Iteration algorithm alternates between policy improvement and policy evaluation. Implementations of this algorithm with several variants of the latter evaluation stage, e.g, n-step and trace-based returns, have been analyzed in previous works. However, the case of multiple-step lookahead policy impr…
Deep reinforcement learning has led to several recent breakthroughs, though the learned policies are often based on black-box neural networks. This makes them difficult to interpret and to impose desired specification constraints during learning. We present an iterative framework, MORL, for improving the learned polici…
Deep RL improves Diplomacy performance, outperforming previous methods.
problem Applying RL to complex, many-agent, simultaneous-move games like Diplomacy.
method Best response policy iteration and fictitious play approximation.
result Deep RL agents convincingly outperform previous Diplomacy agents.
Paper analyzes convergence of entropy-regularized reinforcement learning methods.
problem Ensuring sub-optimality control in entropy-regularized RL.
method Unified analysis of regularized MPI and VI schemes, providing convergence rates.
result Established sufficient conditions for convergence and provided explicit rates.
We propose a policy improvement algorithm for Reinforcement Learning (RL) which is called Rerouted Behavior Improvement (RBI). RBI is designed to take into account the evaluation errors of the Q-function. Such errors are common in RL when learning the Q-value from finite past experience data. Greedy policies or even …
Paper uses RL to optimize multi-asset portfolios in fluctuating markets.
problem Optimizing multi-asset portfolios in time-varying financial markets.
method Soft Actor-Critic (SAC) algorithm for policy learning, policy iteration process.
result SAC algorithm outperforms in various criteria in simulated and real financial markets.
Unified framework for policy improvement in RL with benefits in data efficiency and computation.
problem Improving data efficiency and computation in reinforcement learning for continuous control.
method Local, regularized policy improvement with tree search for continuous action spaces.
result Improves data efficiency and reduces wall-clock time in high-dimensional domains.
SIMPOL solves complex economic models using numerical methods.
problem Optimizing consumption and savings under uncertainty.
method SIMPOL uses a modular numerical framework combining policy iteration and finite difference schemes.
result SIMPOL produces solutions consistent with economic and mathematical theory.
Self-guided ALPs improve MDP policies without domain knowledge.
problem Improving MDP policies with minimal domain knowledge.
method Self-guided sequence of ALPs with random basis functions and state-relevance distribution.
result High probability error bounds and improved policy performance.
We consider the infinite-horizon discounted optimal control problem formalized by Markov Decision Processes. We focus on several approximate variations of the Policy Iteration algorithm: Approximate Policy Iteration, Conservative Policy Iteration (CPI), a natural adaptation of the Policy Search by Dynamic Programming a…
Introduces GAMPS for better model-based policy learning.
problem Misspecified model classes lead to poor policy estimates.
method Exploits current policy to learn approximate transition model, focusing on relevant parts of the environment.
result Empirically validated GAMPS on benchmark domains, demonstrating improved properties.
RANDPOL uses randomized networks for efficient reinforcement learning in continuous state and action MDPs.
problem Efficient reinforcement learning in environments with continuous state and action spaces.
method RANDPOL uses randomized function approximation to represent policy and value functions, providing finite time guarantees and improved numerical performance.
result RANDPOL achieves better numerical performance and provides finite time guarantees compared to deep neural network based algorithms.
RLCFR improves CFR's generalization in imperfect information games.
problem Improving CFR's performance in large-scale, imperfect information games.
method RLCFR integrates CFR with deep reinforcement learning to update strategies dynamically.
result RLCFR significantly enhances CFR's generalization ability in various games.