Learning from Sparse Offline Datasets via Conservative Density Estimation
The author proposes Conservative Density Estimation (CDE) as a novel training algorithm to address challenges in offline reinforcement learning, achieving state-of-the-art performance on the D4RL benchmark by overcoming limitations of existing approaches.
Score Regularized Policy Optimization: Efficient Diffusion Behavior Modeling for Offline RL
Efficiently optimize policies using diffusion behavior modeling in offline RL.
Challenges of Offline Reinforcement Learning with Diverse Data
The key factor impacting the performance of offline reinforcement learning algorithms on diverse data is the scale of the network architecture.
Uncertainty-aware Distributional Offline Reinforcement Learning Study
Proposing a novel uncertainty-aware distributional offline RL method to address epistemic uncertainty and environmental stochasticity simultaneously.
Mutual Information Regularized Offline Reinforcement Learning: A Novel Framework
Die MISA-Plattform bietet eine innovative Lösung für das Offline-Reinforcement-Learning durch direkte Regularisierung der Policy-Verbesserung und -Bewertung mithilfe der gegenseitigen Information zwischen Zuständen und Aktionen im Datensatz.
Testen der Stationarität und Erkennung von Änderungspunkten im Reinforcement Learning
Entwicklung eines Modells zur Testung der Stationarität des optimalen Q-Werts und Erkennung von Änderungspunkten in nicht-stationären Umgebungen.
SCOPE-RL: Python Library for Offline Reinforcement Learning and Off-Policy Evaluation
SCOPE-RL integriert Offline-RL und OPE nahtlos für umfassende Implementierungen.
Uncertainty-Aware Decision Transformer (UNREST) for Offline Reinforcement Learning in Stochastic Driving Environments: Addressing Over-Optimism in Decision Transformers
Decision Transformers (DTs) struggle in stochastic environments like autonomous driving because they are overly optimistic, assuming actions that succeed once will always succeed. UNREST, a novel uncertainty-aware decision transformer, addresses this by estimating uncertainty and segmenting trajectories to learn from actual decision outcomes rather than unreliable future returns.
Pessimistic Nonlinear Least-Squares Value Iteration for Offline Reinforcement Learning (Published at ICLR 2024)
This research paper introduces PNLSVI, a novel algorithm for offline reinforcement learning with non-linear function approximation that achieves near-optimal regret bounds by employing pessimistic value iteration, variance-weighted regression, and a novel D2-divergence measure for uncertainty quantification.
Solving Offline Reinforcement Learning Problems Using Decision Tree Regression: RCDTP and RWDTP Frameworks
This paper introduces two novel offline reinforcement learning frameworks, RCDTP and RWDTP, which reframe RL problems as regression tasks solvable by decision trees, achieving comparable performance to established methods while offering faster training and inference, and enhanced explainability.
© 2024 by Linnk AI