The paper systematically analyzes the phenomenon of varying sparsity ratios across views in multi-view learning, and proposed a targeted data-driven network architecture based on Sparse Autoencoder with Adaptive Constraints.

Analogy-Based Multi-Turn Jailbreak Method for LLMs