PromptSplit: 暴露生成模型中的提示级分歧
PromptSplit: Revealing Prompt-Level Disagreement in Generative Models
AI总结:
PromptSplit通过核方法检测生成模型在提示层面的分歧,提供可解释的分析工具以识别模型行为差异。
摘要:
基于提示的生成AI模型在视觉和语言领域迅速扩展,能够从文本输入中生成逼真且多样的输出。随着此类模型种类的增加,训练数据和架构各不相同,需要系统的方法来识别哪些类型的提示会导致不同的模型行为。在本工作中,我们提出了PromptSplit,一种基于核的框架,用于检测和分析生成模型之间的提示依赖性分歧。对于每对比较的模型,PromptSplit通过将提示和图像(或文本)特征形成张量积嵌入,构建联合提示-输出表示,然后计算相应的核协方差矩阵。我们利用这些矩阵加权差的特征空间来识别提示跨行为差异的主要方向。为了确保可扩展性,我们采用随机投影近似,将计算复杂度降低到$O(nr^2 + r^3)$,其中r为投影维度。我们进一步提供了一项理论分析,表明这种近似会产生一个特征结构估计,其预期偏离完整维度结果的误差被限制在$O(1/r^2)$以内。在文本到图像、文本到文本和图像描述设置中的实验表明,PromptSplit能够准确检测真实的行为差异并隔离负责的提示,提供了一个可解释的工具来检测生成模型在何处存在分歧。
英文摘要:
Prompt-guided generative AI models have rapidly expanded across vision and language domains, producing realistic and diverse outputs from textual inputs. The growing variety of such models, trained with different data and architectures, calls for principled methods to identify which types of prompts lead to distinct model behaviors. In this work, we propose PromptSplit, a kernel-based framework for detecting and analyzing prompt-dependent disagreement between generative models. For each compared model pair, PromptSplit constructs a joint prompt--output representation by forming tensor-product embeddings of the prompt and image (or text) features, and then computes the corresponding kernel covariance matrix. We utilize the eigenspace of the weighted difference between these matrices to identify the main directions of behavioral difference across prompts. To ensure scalability, we employ a random-projection approximation that reduces computational complexity to $O(nr^2 + r^3)$ for projection dimension $r$. We further provide a theoretical analysis showing that this approximation yields an eigenstructure estimate whose expected deviation from the full-dimensional result is bounded by $O(1/r^2)$. Experiments across text-to-image, text-to-text, and image-captioning settings demonstrate that PromptSplit accurately detects ground-truth behavioral differences and isolates the prompts responsible, offering an interpretable tool for detecting where generative models disagree.
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。