The field of optimization and feature transformation is experiencing significant advancements, driven by the integration of reinforcement learning, evolutionary algorithms, and graph-driven path optimization. Novel approaches, such as collaborative multi-agent reinforcement learning and hierarchical reinforcement learning, are being explored to improve the efficiency and adaptability of optimization processes. These innovations aim to address the challenges of complex datasets and high-dimensional feature spaces, enhancing the performance of downstream machine learning tasks. Notable developments include the use of learning-based cooperative coevolution, dynamic decomposition strategies, and graph-driven path optimization to automate feature transformation and improve optimization effectiveness. Noteworthy papers in this area include:

• Unlearning Works Better Than You Think, which introduces Local Reinforcement-Based Selection of Auxiliary Objectives, a novel approach that selects auxiliary objectives using reinforcement learning to support optimization processes.
• Collaborative Multi-Agent Reinforcement Learning for Automated Feature Transformation with Graph-Driven Path Optimization, which proposes a framework that automates feature engineering through graph-driven path optimization.
• Comprehend, Divide, and Conquer: Feature Subspace Exploration via Multi-Agent Hierarchical Reinforcement Learning, which presents a novel approach that employs a Large Language Model-based hybrid state extractor to capture feature characteristics and constructs hierarchical agents for each cluster and sub-cluster.
• Advancing CMA-ES with Learning-Based Cooperative Coevolution for Scalable Optimization, which introduces a learning-based cooperative coevolution framework that dynamically schedules decomposition strategies during optimization processes.