Initial State Distribution

Initial state distribution research focuses on optimizing the starting conditions for various systems, particularly in reinforcement learning and dynamical systems, to improve efficiency and performance. Current efforts involve developing advanced algorithms, such as those incorporating explanation methods or curriculum learning, to strategically select or modify initial states, often leveraging techniques like experience replay and neural networks. This work is significant because improved control over initial states can lead to faster training, enhanced performance in complex tasks (e.g., robotics), and more accurate modeling of physical processes, impacting fields ranging from artificial intelligence to chemical reaction dynamics.

Papers

May 5, 2024

RICE: Breaking Through the Training Bottlenecks of Reinforcement Learning with Explanation
Zelei Cheng, Xian Wu, Jiahao Yu, Sabrina Yang, Gang Wang, Xinyu Xing
Reinforcement Learning Deep Reinforcement Learning Line by Line Explanation Sparse Reward Agent Performance Training Bottleneck Rice Kernel Initial State Distribution

December 14, 2023

HiER: Highlight Experience Replay and Easy2Hard Curriculum Learning for Boosting Off-Policy Reinforcement Learning Agents
Dániel Horváth, Jesús Bujalance Martín, Ferenc Gábor Erdős, Zoltán Istenes, Fabien Moutarde
Robotics Domain Action Space Policy Reinforcement Learning Experience Replay Hindsight Experience Replay Initial State Distribution

April 26, 2023

Controlled density transport using Perron Frobenius generators
Jake Buzhardt, Phanindra Tallapragada
Vector Field Particle Density Non Linear Operator Control Problem Density Control Initial State Distribution

January 29, 2023

Sample Efficient Deep Reinforcement Learning via Local Planning
Dong Yin, Sridhar Thiagarajan, Nevena Lazic, Nived Rajaraman, Botao Hao, Csaba Szepesvari
Reinforcement Learning Uncertainty Aware Sample Efficient Reinforcement Learning Double DQN Local Planning Initial State Distribution

November 5, 2021

Machine Learning Product State Distributions from Initial Reactant States for a Reactive Atom-Diatom Collision System
Julian Arnold, Juan Carlos San Vicente Veliz, Debasish Koner, Narendra Singh, Raymond J. Bemish, Markus Meuwly
Product Distribution Molecular Simulation Initial State Distribution

Initial State Distribution

Papers

RICE: Breaking Through the Training Bottlenecks of Reinforcement Learning with Explanation

HiER: Highlight Experience Replay and Easy2Hard Curriculum Learning for Boosting Off-Policy Reinforcement Learning Agents

Controlled density transport using Perron Frobenius generators

Sample Efficient Deep Reinforcement Learning via Local Planning

Machine Learning Product State Distributions from Initial Reactant States for a Reactive Atom-Diatom Collision System