Search papers, labs, and topics across Lattice.
This paper introduces MapRoute++, an enhanced method for visual concept unlearning that incorporates task-specific training objectives and semantic routing for selecting concept-specific mappers. By improving the robustness of concept removal while maintaining the integrity of unrelated and semantically adjacent concepts, the approach addresses critical challenges in the field. The method achieves a significant 12.1% improvement over the state-of-the-art baseline on the Erasing-Retention-Robustness (ERR) metric, demonstrating its effectiveness across five concept categories in the Gen渭 2.0 Challenge.
Achieving a 12.1% boost in robust concept removal, MapRoute++ redefines the boundaries of visual concept unlearning.
We present our submission to Task 3 of the Gen$\mu$ 2.0 Challenge on visual concept unlearning. Building on MapRoute, we introduce task-specific training objectives, richer concept representations, and semantic routing for concept-specific mapper selection. Our approach improves robust concept removal while preserving unrelated and semantically adjacent concepts. On the official benchmark, evaluated using the Erasing-Retention-Robustness (ERR) metric on Stable Diffusion v1.4, our method outperforms the state-of-the-art baseline by 12.1\% on average across the five concept categories, achieving substantial gains.
