论文标题
部分可观测时空混沌系统的无模型预测
Indeterminacy in Generative Models: Characterization and Strong Identifiability
论文作者
论文摘要
大多数现代概率的生成模型，例如变异自动编码器（VAE），即使有大量的数据也无法解决某些不确定的不确定。不同的任务可以容忍不同的不确定，但是最近的应用表明需要强烈可识别的模型，其中观察结果与唯一的潜在代码相对应。在保持灵活性的同时，取得了进步，在减少模型不确定性的同时，最近的工作不包括很多 - 但并非全部 - 要求。在这项工作中，我们可以从任务可识别性方面激励模型可识别性，然后构建一个理论框架，用于分析潜在变量模型的不确定性，这可以在生成器函数和先前的分布空间方面进行精确的表征。我们透露，即使使用高灵活的非线性发电机，也可能实现强可识别性，并给出两个这样的示例。一种是对IVAE的直接修改（Arxiv：1907.04809 [Stat.ml]）;另一个使用三角形单调图，从而导致最佳运输与可识别性之间的新连接。
Most modern probabilistic generative models, such as the variational autoencoder (VAE), have certain indeterminacies that are unresolvable even with an infinite amount of data. Different tasks tolerate different indeterminacies, however recent applications have indicated the need for strongly identifiable models, in which an observation corresponds to a unique latent code. Progress has been made towards reducing model indeterminacies while maintaining flexibility, and recent work excludes many--but not all--indeterminacies. In this work, we motivate model-identifiability in terms of task-identifiability, then construct a theoretical framework for analyzing the indeterminacies of latent variable models, which enables their precise characterization in terms of the generator function and prior distribution spaces. We reveal that strong identifiability is possible even with highly flexible nonlinear generators, and give two such examples. One is a straightforward modification of iVAE (arXiv:1907.04809 [stat.ML]); the other uses triangular monotonic maps, leading to novel connections between optimal transport and identifiability.
