Continuous Optimization for Feature Selection with Permutation-Invariant Embedding and Policy-Guided Search

Researchers propose a new framework for feature selection that uses permutation-invariant embedding and reinforcement learning to address limitations in current methods. The approach combines an encoder-decoder paradigm to preserve feature relationships without order bias and employs policy-based RL to explore embedding spaces without convexity assumptions.

• →New framework addresses permutation sensitivity issues in feature selection embedding processes.
• →Encoder-decoder paradigm captures feature interactions through pairwise relationships while maintaining order independence.
• →Policy-based reinforcement learning approach eliminates reliance on convexity assumptions for better optimization.
• →Inducing point mechanism introduced to accelerate pairwise relationship computations for efficiency.
• →Extensive experiments demonstrate improved effectiveness, efficiency, robustness and explicitness compared to existing methods.