无需数据即可训练生成模型，支持连续、离散及混合模态数据生成。
Energy-based generator matching: A neural sampler for general state space
- 基于能量函数构建神经采样器，通过自归一化重要性采样优化损失
- 在100维离散和20维混合模态任务上验证，支持扩散、流与跳跃过程
- 适用于无数据场景，适合研究生成模型基础机制的学者
我们提出能量基生成器匹配（EGM），一种无需数据即可从能量函数训练生成模型的通用方法。该方法扩展了最近提出的生成器匹配，能够训练任意连续时间马尔可夫过程，如扩散、流和跳跃过程，并可生成连续、离散以及两者的混合模态数据。为此，我们提出使用自归一化重要性采样估计生成器匹配损失，并引入额外的自助技巧以降低重要性权重的方差。我们在离散任务（最高100维）和多模态任务（最高20维）上验证了EGM的有效性。
原文摘要 · Abstract (English)
We propose Energy-based generator matching (EGM), a modality-agnostic approach to train generative models from energy functions in the absence of data. Extending the recently proposed generator matching, EGM enables training of arbitrary continuous-time Markov processes, e.g., diffusion, flow, and jump, and can generate data from continuous, discrete, and a mixture of two modalities. To this end, we propose estimating the generator matching loss using self-normalized importance sampling with an additional bootstrapping trick to reduce variance in the importance weight. We validate EGM on both discrete and multimodal tasks up to 100 and 20 dimensions, respectively.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。