arXiv 2504.081492025-04-10cs.CV

LoRAX：用于持续合成图像溯源的 LoRA 可扩展网络

LoRAX: LoRA eXpandable Networks for Continual Synthetic Image Attribution

MIT Lincoln Laboratory（麻省理工学院林肯实验室）；Yale University（耶鲁大学）

Danielle Sullivan-Pao, Nicole Tian, Pooya Khorrami

发表信息：: Proceedings of the Media Authenticity in the Age of Artificial Intelligence Workshop at the 35th British Machine Vision Conference (BMVC), Glasgow, UK, 2024

AI总结：

针对持续合成图像溯源中模型难以泛化到新生成模型的问题，本文提出参数高效的类增量算法 LoRAX，通过低秩适配训练任务特定特征提取器，在 Continual Deepfake Detection 基准上以不到 3% 的可训练参数取得领先性能。

摘要：

随着生成式 AI 图像技术日益普及和先进，对强溯源模型的需求不断增长。这些模型对于验证图像真实性以及识别其原始生成模型的架构至关重要，是维护媒体完整性的关键。然而，溯源模型难以泛化到未见过的模型，而用于更新这些模型的传统微调方法在实际环境中已被证明不切实际。为应对这些挑战，我们提出了 LoRA 可扩展网络（LoRAX），这是一种参数高效的类增量算法，无需完全重新训练即可适应新型生成图像模型。我们的方法通过低秩适配（Low Rank Adaptation）为每个持续学习任务训练一个参数极度高效的特征提取器。每个任务特定的特征提取器学习不同的特征，同时仅需底层特征提取器骨干模型中极少一部分参数。我们的大量实验表明，在 Continual Deepfake Detection 基准上，LoRAX 在所有训练场景和记忆设置下均优于或保持与最先进的类增量学习算法相当的竞争力，同时每个特征提取器所需可训练参数数量不到全秩实现的 3%。LoRAX 代码可在以下地址获取：https://github.com/mit-ll/lorax_cil。

英文摘要：

As generative AI image technologies become more widespread and advanced, there is a growing need for strong attribution models. These models are crucial for verifying the authenticity of images and identifying the architecture of their originating generative models-key to maintaining media integrity. However, attribution models struggle to generalize to unseen models, and traditional fine-tuning methods for updating these models have shown to be impractical in real-world settings. To address these challenges, we propose LoRA eXpandable Networks (LoRAX), a parameter-efficient class incremental algorithm that adapts to novel generative image models without the need for full retraining. Our approach trains an extremely parameter-efficient feature extractor per continual learning task via Low Rank Adaptation. Each task-specific feature extractor learns distinct features while only requiring a small fraction of the parameters present in the underlying feature extractor's backbone model. Our extensive experimentation shows LoRAX outperforms or remains competitive with state-of-the-art class incremental learning algorithms on the Continual Deepfake Detection benchmark across all training scenarios and memory settings, while requiring less than 3% of the number of trainable parameters per feature extractor compared to the full-rank implementation. LoRAX code is available at: https://github.com/mit-ll/lorax_cil.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。