Designing Perceptual Puzzles by Differentiating Probabilistic Programs
Designing Perceptual Puzzles by Differentiating Probabilistic Programs
复制标题
DOI：
10.1145/3528233.3530715
复制
发表时间：
2022-04
期刊：
影响因子：
--
通讯作者：
Kartik Chandra;Tzu-Mao Li;J. Tenenbaum;Jonathan Ragan-Kelley
中科院分区：
文献类型：
--
作者：
Kartik Chandra;Tzu-Mao Li;J. Tenenbaum;Jonathan Ragan-Kelley
我们通过为人类感知的主要模型找到“对抗性示例”来设计新的视觉幻象，特别是对于概率模型，它们将视觉视为有效执行此搜索的视力，我们设计了一种可不同的概率编程语言，其API揭示了MCMC推论作为一流的函数。恒定，大小恒定和面部感知。
We design new visual illusions by finding “adversarial examples” for principled models of human perception — specifically, for probabilistic models, which treat vision as Bayesian inference. To perform this search efficiently, we design a differentiable probabilistic programming language, whose API exposes MCMC inference as a first-class differentiable function. We demonstrate our method by automatically creating illusions for three features of human vision: color constancy, size constancy, and face perception.