Abstract
Large Language Model (LLM) unlearning is essential for removing sensitive or copyrighted knowledge while preserving general utility. Existing methods often leave residual knowledge in intermediate representations, which can still be recovered. To address this, we propose Cascade, a hierarchical recoverability control framework that minimizes the internal identifiability of target knowledge. Cascade combines three complementary controls: path-level routing to suppress privacy-associated activation routes, representation-level compression to reduce geometric separability, and decoding-level intervention to limit residual recovery. Experiments on TOFU, MUSE-News, and WMDP, including robustness tests with query reformulation and extraction-style prompts, show that Cascade effectively reduces recoverability while maintaining stable model utility.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Yu, Q., Duan, S., Li, X., Wang, Y., Li, Z., Zhou, S., Sun, Y., & Fan, Z. (2026). Cascade: Hierarchical Recoverability Control for Large Language Model Unlearning. https://omanscience.com/en/articles/cascade-hierarchical-recoverability-control-for-large-language-model-unlearning
MLA 9
Yu, Qingchen, et al. "Cascade: Hierarchical Recoverability Control for Large Language Model Unlearning." https://omanscience.com/en/articles/cascade-hierarchical-recoverability-control-for-large-language-model-unlearning.
Chicago (author–date)
Yu, Qingchen, Shiying Duan, Xiaodong Li, Yuhua Wang, Zhiyu Li, Shiji Zhou, Yifan Sun, and Zhaoxin Fan. 2026. "Cascade: Hierarchical Recoverability Control for Large Language Model Unlearning." https://omanscience.com/en/articles/cascade-hierarchical-recoverability-control-for-large-language-model-unlearning.
Harvard
Yu, Q., Duan, S., Li, X., Wang, Y., Li, Z., Zhou, S., Sun, Y. and Fan, Z. (2026) 'Cascade: Hierarchical Recoverability Control for Large Language Model Unlearning', Available at: https://omanscience.com/en/articles/cascade-hierarchical-recoverability-control-for-large-language-model-unlearning.
Vancouver
Yu Q, Duan S, Li X, Wang Y, Li Z, Zhou S, et al. Cascade: Hierarchical Recoverability Control for Large Language Model Unlearning. https://omanscience.com/en/articles/cascade-hierarchical-recoverability-control-for-large-language-model-unlearning
IEEE
Q. Yu, S. Duan, X. Li, Y. Wang, Z. Li, S. Zhou, Y. Sun, and Z. Fan, "Cascade: Hierarchical Recoverability Control for Large Language Model Unlearning," https://omanscience.com/en/articles/cascade-hierarchical-recoverability-control-for-large-language-model-unlearning.