EraseFlow: Learning Concept Erasure Policies via GFlowNet-Driven Alignment

NeurIPSSpotlight2025

Authors
Naga Sai Abhiram kusumba, Maitreya Patel, Kyle Min, Changhoon Kim, Chitta Baral, Yezhou Yang
Venue
NeurIPS 2025
Track
Spotlight

TL;DR

Erasing harmful or proprietary concepts from powerful text‑to‑image generators is an emerging safety requirement, yet current ``concept erasure'' techniques either collapse image quality, rely on brit…

Opening excerpt from the authors’ abstract. source

Read the paper

Topics

alignment safety

Related NeurIPS papers

← All NeurIPS 2025 Spotlight papers · Browse the whole archive