MONTRAGE: Monitoring Training for Attribution of Generative Diffusion Models

Diffusion models, which revolutionized image generation, are facing challenges related to intellectual property. These challenges arise when a generated image is influenced by copyrighted images from the training data, a plausible scenario in internet-collected data. Hence, pin-pointing influential images from the training dataset, a task known as data attribution, becomes crucial for transparency of content origins. We introduce MONTRAGE, a pioneering data attribution method. Unlike existing approaches that analyze the model post-training, MONTRAGE integrates a novel technique to monitor generations throughout the training via internal model representations. It is tailored for customized diffusion models, where training dynamics access is a practical assumption. This approach, coupled with a new loss function, enhances performance while maintaining efficiency. The advantage of MONTRAGE is evaluated in two granularity-levels: Between-concepts and within-concept, outperforming current state-of-the-art methods for high accuracy. This substantiates MONTRAGE's insights on diffusion models and its contribution towards copyright solutions for AI digital-art.