This paper compares expected and distributional reinforcement learning methods.
problem Understanding why distributional reinforcement learning performs better than expected reinforcement learning.
method Analyzes differences in tabular, linear, and non-linear approximation settings.
result Distributional RL can hurt performance if it does not induce identical behavior.
MaxEnt RL optimizes decision-making in uncertain environments.
problem MaxEnt RL optimizes decision-making in uncertain environments.
method Formal analysis of MaxEnt RL in POMDPs and adversarial reward games.
result MaxEnt RL optimally solves certain classes of control problems with uncertain reward functions.
Introduces a new RL formalism focusing on event desirability.
problem Limited consideration of performance distribution in RL.
method Introduces micro-objective reinforcement learning.
result New formalism allows for prior knowledge and event desirability.
PDCA algorithm learns policies for RL with constraints using a primal-dual approach.
problem Offline constrained reinforcement learning with general function approximation.
method Primal-Dual-Critic Algorithm (PDCA) using a primal-dual approach.
result PDCA finds a near saddle point of the Lagrangian, nearly optimal for constrained RL.
Simpler one-step distributional RL framework for control.
problem Lack of a unified theory for DistrRL in control.
method One-step distributional reinforcement learning (OS-DistrRL) framework.
result Unified theory for policy evaluation and control.
New RL formulation for maximizing maximum reward in molecule generation.
problem Traditional RL frameworks do not fit real-world applications like drug discovery.
method Formulated a new objective function to maximize maximum reward, derived Bellman equation, introduced operators, and proved convergence.
result Achieved state-of-the-art results in molecule generation.
Study on generalization in reparameterizable RL, deriving new guarantees.
problem Understanding generalization in reparameterizable RL.
method Using supervised learning and transfer learning theory, derived guarantees on the gap between expected and empirical return.
result Generalization capability of reparameterizable RL is related to multiple factors including smoothness of the environment transition, reward, and policy function class.
New RL algorithm minimizes distributional learning error.
problem Improving distributional reinforcement learning for better error minimization.
method Proposes a new model-based algorithm with theoretical minimax optimality.
result Proves minimax optimality for approximating return distributions.
Develops DPG methods for continuous-time RL with deterministic policies.
problem High variance and slow convergence in stochastic policy RL methods.
method Derives continuous-time policy gradient formula and proposes CT-DDPG algorithm.
result CT-DDPG achieves superior stability and faster convergence in continuous-time RL.
Paper develops NPG for risk-averse RL with ECRMs, proving global convergence.
problem Ensuring reliable performance in stochastic RL problems with risk-averse policies.
method Developed natural policy gradient updates for ECRMs-based RL problems, proving global optimality and iteration complexity.
result Global convergence of risk-averse NPG algorithm with ECRMs.
Study finds methods to learn multiple solutions from single task in offline RL.
problem Learning multiple solutions from a single task in offline RL.
method Proposed algorithms for offline RL.
result Empirical results show learning of multiple solutions in offline RL.
Risk-sensitive RL aims to optimize performance while considering risk constraints or measures.
problem Optimizing expected value alone may not be sufficient; incorporating risk measures is necessary.
method Policy gradient search is used to solve risk-sensitive RL problems, considering various risk measures.
result Policy gradient methods can be adapted for risk-sensitive RL, addressing challenges and future directions.
This work analyzes the value of future reward information in RL.
problem Analyzing the impact of knowing future rewards in reinforcement learning.
method Competitive analysis and worst-case reward distribution.
result Exact ratios between standard RL agents and those with future-reward lookahead.
Model-based reinforcement learning (RL) is considered to be a promising approach to reduce the sample complexity that hinders model-free RL. However, the theoretical understanding of such methods has been rather limited. This paper introduces a novel algorithmic framework for designing and analyzing model-based RL algo…
Data-driven RL solves Merton's expected utility problem via policy randomization.
problem Maximizing expected utility in an incomplete market with unknown primitives.
method Policy randomization in continuous-time reinforcement learning.
result RL algorithms solve Merton's problem without estimating model primitives.
Simplifies BCQ to match and outperform state-of-the-art in offline RL benchmarks.
problem Sample efficiency in offline reinforcement learning.
method Introduces EMaQ, a novel backup operator for offline RL.
result EMaQ matches and outperforms prior state-of-the-art in offline RL benchmarks.
VIREL uses variational inference for RL, improving policy learning and performance.
problem Challenges in learning optimal policies in reinforcement learning.
method VIREL proposes a novel probabilistic inference framework that uses a parametrised action-value function to capture future dynamics.
result VIREL enables learning deterministic optimal policies naturally and optimises value functions and policies separately.
Most conventional Reinforcement Learning (RL) algorithms aim to optimize decision-making rules in terms of the expected returns. However, especially for risk management purposes, other risk-sensitive criteria such as the value-at-risk or the expected shortfall are sometimes preferred in real applications. Here, we desc…
SIBRE boosts reinforcement learning convergence by rewarding improvement over past performance.
problem Improving the rate of convergence in reinforcement learning.
method SIBRE is a reward shaping approach that rewards improvement over the agent's own past performance.
result SIBRE converges faster and more stably to the optimal policy compared to baseline RL algorithms.
Two-tier approach optimizes RL hyper-parameters for better agent learning.
problem Optimizing hyper-parameters in reinforcement learning to improve agent performance.
method Two-step optimization: first categorical hyper-parameters, then solution-level hyper-parameters.
result Promising results in simulated control tasks, suggesting user-independent reinforcement learning applications.
Meta-gradient D4PG optimizes performance and constraint adherence in RL.
problem Balancing performance and adherence to complex constraints in RL.
method Uses meta-gradients to find a balance between expected return and minimizing constraint violations.
result Meta-gradient D4PG consistently outperforms baselines across MuJoCo domains.
Direct and indirect RL methods classified and compared.
problem Classifying RL methods for sequential decision making.
method Direct RL solves optimal policy directly, indirect RL solves Bellman equation.
result Direct and indirect RL methods are equivalent and can be unified.
A method for Bayes-Adaptive Deep RL using meta-learning.
problem Maximizing expected return in unknown environments with uncertainty.
method variBAD: meta-learning for approximate inference and task uncertainty.
result variBAD achieves higher online return than existing methods in MuJoCo domains.
Paper proposes a new method for better estimating continuous distributions in RL.
problem Challenges in parameterizing estimated distributions for better approximation of true continuous distribution.
method Proposes fully parameterized quantile function with fraction and value networks.
result Significantly outperforms existing distributional RL algorithms on 55 Atari Games.
Active inference enhances RL by balancing exploration and exploitation.
problem Traditional RL's balance between exploration and exploitation is often suboptimal.
method Developed a new decision-making objective based on active inference.
result The new algorithm successfully balances exploration and exploitation on various RL benchmarks.
Kernel-based function approximation improves reinforcement learning performance.
problem Average reward reinforcement learning in infinite horizon settings.
method Optimistic algorithm based on kernel ridge regression.
result No-regret performance guarantees and confidence intervals for kernel-based predictions.
Bayesian RL enhances LLMs to reflectively explore and correct errors.
problem LLMs trained via RL lack reflective behaviors like rethinking and error correction.
method Bayesian RL framework that optimizes expected return under posterior distribution over Markov decision processes.
result BARL algorithm improves LLM performance in reasoning tasks.
A new EM framework for goal-conditioned RL improves performance on sparse reward tasks.
problem Handling sparse rewards in goal-conditioned reinforcement learning.
method A graphical model framework with an EM algorithm that includes a learning-in-hindsight E-step and a supervised M-step.
result hEM significantly outperforms model-free baselines on goal-conditioned benchmarks with sparse rewards.
Extends Thompson sampling for RL with fewer episodes.
problem Limited episodes in RL settings.
method Batch Bayesian optimization over episodes to learn action bias terms.
result Significantly outperforms standard Thompson sampling.
Robo-advisor uses ML to optimize investment performance.
problem Maximizing investment performance with historical data.
method Inverse optimization and deep reinforcement learning.
result Robo-advisor consistently outperformed S&P 500.
Paper introduces DTAE to optimize RL algorithms, balancing exploration and exploitation.
problem Balancing exploration and exploitation in reinforcement learning.
method Soft policy optimization with entropy and dual-track advantage estimator (DTAE).
result DTAE accelerates RL algorithm convergence and improves performance.
This paper shows RL with KL penalties is equivalent to Bayesian inference for fine-tuning LMs.
problem Fine-tuning large language models to avoid undesirable features.
method Analyzed KL-regularized RL and showed it's equivalent to variational inference.
result KL-regularized RL avoids distribution collapse and is more insightful as Bayesian inference.
Survey of offline RL theory and practical algorithm design challenges.
problem Optimizing return from fixed agent trajectories without additional interactions.
method Theoretical insights and practical algorithm design.
result Conditions for practical offline RL algorithms and their limitations.
RePULSe improves language model alignment by reducing undesired outputs without sacrificing overall performance.
problem Aligning language models with human preferences while minimizing undesired outputs.
method Integrates probabilistic inference into RL training to reduce undesired outputs.
result RePULSe achieves a better balance between expected reward and undesired output probability.
Regularization improves generalization in Bayesian RL, shown through algorithmic stability.
problem Ensuring good generalization in Bayesian reinforcement learning.
method Algorithmic stability, using regularization and fast convergence rates for mirror descent.
result Regularization makes the optimal policy stable, improving generalization.
Enhances RL agents with predictive internal representations.
problem Improving model-free reinforcement learning agents.
method Introduces Deep InfoMax (DIM) objective to train predictive internal representations.
result Successfully learned predictive representations in synthetic settings.
New method identifies critical states to improve RL agent explainability and speed.
problem Challenges in RL agent explainability and action selection timing.
method Identify critical states based on action-based variance in Q-function, prioritize exploitation on these states.
result Identified critical states accelerate RL in grid worlds and deep RL tasks.
This work develops efficient methods for continuous-time distributional reinforcement learning.
problem Continuous-time reinforcement learning with return distributions.
method Parameterizing return distributions using quantile representation and showing topological properties.
result Efficient approximation algorithm for continuous-time distributional reinforcement learning.
Paper characterizes optimal language model alignment methods.
problem Aligning language models to maximize reward while keeping them close to the original model.
method KL-constrained reinforcement learning and best-of-N methods.
result Optimal KL-constrained RL solution has a large deviation principle rate function.
Policy optimization is a core component of reinforcement learning (RL), and most existing RL methods directly optimize parameters of a policy based on maximizing the expected total reward, or its surrogate. Though often achieving encouraging empirical success, its underlying mathematical principle on {\em policy-distri…
Paper establishes DRL for high-dimensional rewards.
problem Intractable reinforcement learning with high-dimensional rewards.
method Theoretical foundations and a novel DRL algorithm.
result Bellman operator contraction in high-dimensional spaces.
CQL learns conservative Q-functions to improve offline RL performance.
problem Leveraging large, static datasets in reinforcement learning without further interaction.
method Conservative Q-learning (CQL) which learns a conservative Q-function to lower-bound policy values.
result CQL substantially outperforms existing offline RL methods, often achieving 2-5 times higher final returns.
Framework for robust RL in continuous control with model misspecification.
problem Model misspecification in reinforcement learning for continuous control.
method Integrates robustness into MPO algorithm through worst-case expected return objective and entropy regularization.
result Robust and soft-robust policies outperform non-robust policies in various domains.
Improved RL algorithm with linear MDPs for offline learning with partial data coverage.
problem Efficient offline RL with linear MDPs under partial data coverage.
method Primal-dual algorithm with O(ε−2) sample complexity. result First computationally efficient algorithm with O(ε−2) sample complexity for offline RL with linear MDPs under partial data coverage. GPPSTD uses Gaussian Processes for efficient RL in continuous states.
problem Efficient reinforcement learning in continuous state spaces.
method Gaussian Process Posterior Sampling Reinforcement Learning (GPPSTD) algorithm.
result Combining demonstration and exploration improves reinforcement learning efficiency.
Proposes an RL algorithm for handling convex constraints.
problem Handling constraints in reinforcement learning.
method Algorithm for convex constraints in RL.
result Algorithm can enforce new properties like diversity.
TDprop uses Jacobi preconditioning to improve adaptive optimizers in Deep RL.
problem Improving performance of adaptive optimizers in Deep RL.
method TDprop computes per-parameter learning rates based on Jacobi preconditioning of the TD update rule.
result TDprop matches or exceeds Adam's performance in Deep RL experiments, suggesting Jacobi preconditioning can improve adaptive methods.
The goal of reinforcement learning (RL) is to let an agent learn an optimal control policy in an unknown environment so that future expected rewards are maximized. The model-free RL approach directly learns the policy based on data samples. Although using many samples tends to improve the accuracy of policy learning, c…