Dimensions overall score 3.0

GitHub Code Pulse

No public code linked for this paper yet.

• Instruction-based unlearning fails to suppress targeted concepts in diffusion-based image generation models when guided solely by natural-language unlearning instructions.result 90%
• Instruction-based unlearning has proven effective for modifying behavior of large language models at inference time.technical 85%
• Unlearning instructions do not induce sustained reductions in attention to targeted concept tokens in diffusion models.technical 85%
• Targeted concept representations persist throughout generation in diffusion models despite unlearning instructions.result 85%
• Effective unlearning in diffusion models requires interventions beyond inference-time language control.limitation 80%
• The study used controlled experiments across multiple concepts and prompt variants to investigate instruction-based unlearning in diffusion models.method 80%
• The analysis examined both CLIP text encoder and cross-attention dynamics during the denoising process.method 80%
• There is a fundamental limitation of prompt-level instruction in diffusion models for unlearning tasks.limitation 75%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card