ICML Reinforcement Learning Article Classification

serial number	article	Key words	roughly means
61	Reinforcement Learning with General Utilities: Simpler Variance Reduction and Large State-Action Space	General Utilities, PG	Introduced a gradient descent method for General Utilities, that is, the objective function is a nonlinear function of state-action pair distribution
62	Trajectory-Aware Eligibility Traces for Off-Policy Reinforcement Learning	off-policy, Important sampling	An importance sampling algorithm RBIS for offline learning is introduced, which expresses trace as a function of t instead of the product of IS ratio
63	Semi-Offline Reinforcement Learning for Optimized Text Generation	semi-offline, LLM	Introduces a semi-offline reinforcement learning method for training language models, that is, using the training data of the model, only one step of model reasoning is required
64	StriderNet: A Graph Reinforcement Learning Approach to Optimize Atomic Structures on Rough Energy Landscapes	GNN, Atomic structs, RL applications	The atomic structure is optimized using reinforcement learning, and a GNN is used to extract the characteristics of the atomic structure
65	Reinforcement Learning Can Be More Efficient with Multiple Rewards	-
66	LESSON: Learning to Integrate Exploration Strategies for Reinforcement Learning via an Option Framework	-
67	Flipping Coins to Estimate Pseudocounts for Exploration in Reinforcement Learning	Exploration	Use a network to estimate whether each step in the trajectory visits the state, and calculate a corresponding bonus obtained by exploration
68	Interactive Object Placement with Reinforcement Learning	-
69	Oracles and Followers: Stackelberg Equilibria in Deep Multi-Agent Reinforcement Learning	Stackelberg Equilibria, game theory, RL applications	Introduces a framework for implementing Stackelberg equilibrium problems with multiple agents
70	Non-stationary Reinforcement Learning under General Function Approximation	Non-stationary	Introduces SW-OPEA, a reinforcement learning algorithm for unstable environments, using sliding window and confidence-based conditions when screening policy set functions
71	Multi-task Hierarchical Adversarial Inverse Reinforcement Learning	IL, IRL，Muti-task	A layered imitation learning algorithm MH-AIRL is introduced, which improves the AIRL algorithm and can be used in multi-task
72	Local Optimization Achieves Global Optimality in Multi-Agent Reinforcement Learning	Multi-Agent	A PPO algorithm for multi-agent tasks is proposed. The agents are updated in a certain order, and the action of the previous agent is used as the condition when updating.
73	Entity Divider with Language Grounding in Multi-Agent Reinforcement Learning	Muti-Agent, LLM	Introduces a multi-agent reinforcement learning framework, EnDi, which avoids sub-goal conflicts and improves generalization by dividing the entities that agents need to interact with.
74	Parallel $Q$-Learning: Scaling Off-policy Reinforcement Learning under Massively Parallel Simulation	Parallel	Introduced a parallel reinforcement learning algorithm PQL, based on DDQN extension, parallel update Q function and strategy.
75	Simple Embodied Language Learning as a Byproduct of Meta-Reinforcement Learning	-
76	Language Instructed Reinforcement Learning for Human-AI Coordination	LLM	Introduced a reinforcement learning framework instructRL, which uses human instructions to correct the Q function, improves Q-learning and PPO, and improves the human-machine coordination ability
77	Representation-Driven Reinforcement Learning	Exploration	Introduces ReRL, a reinforcement learning framework, which transforms the exploration problem into a representation problem by representing the parameter of the policy as a value for exploration
78	Efficient Online Reinforcement Learning with Offline Data	offline	A new framework RLPD is introduced, which utilizes offline data by mixing offline data and online data, adding LayerNorm and other methods
79	Reinforcement Learning with History Dependent Dynamic Contexts	Non-stationary	介绍了一种动态的马尔科夫决策过程DCMDP，采用特征映射来获取历史向量，以及一种采用最大似然法求解特征映射的方法LDC-UCB，以及一种基于模型的方法DCZero
80	Improved Regret for Efficient Online Reinforcement Learning with Linear Function Approximation	Exploration, adversarial cost	介绍了一种基于最小二乘法的强化学习算法PO-LSBE，用于鼓励在可变损失环境中的进行探索。