Dietterich: Machine Unlearning Is a Harder Form of Weight-Space Programming

Thomas Dietterich argued that machine unlearning is essentially a form of weight-space programming, but harder since it aims to fully remove a capability rather than merely alter behavior. He suggested using weight-space programming attacks to recover forgotten abilities as a way to test the robustness of unlearning methods.

2026-11-04 ~ 2026-11-04 · 3 related posts

• Tom Dietterich: Unlearning methods are a form of weight space programming — tdietterich · 2026-11-04
• Dietterich: Unlearning is tougher—it removes capabilities, not just behavior — tdietterich · 2026-11-04
• Dietterich suggests testing unlearning robustness by attacking it — tdietterich · 2026-11-04