Introspective Generative Modeling: Decide Discriminatively
This addresses unsupervised learning problems for researchers in generative modeling, but it appears incremental as it builds on existing discriminative and generative approaches.
The paper tackles unsupervised learning by proposing introspective generative modeling (IGM), which uses a generator that acts as a discriminator to self-evaluate differences between generated and training data, achieving encouraging results in applications like texture modeling, artistic style transferring, face modeling, and semi-supervised learning.
We study unsupervised learning by developing introspective generative modeling (IGM) that attains a generator using progressively learned deep convolutional neural networks. The generator is itself a discriminator, capable of introspection: being able to self-evaluate the difference between its generated samples and the given training data. When followed by repeated discriminative learning, desirable properties of modern discriminative classifiers are directly inherited by the generator. IGM learns a cascade of CNN classifiers using a synthesis-by-classification algorithm. In the experiments, we observe encouraging results on a number of applications including texture modeling, artistic style transferring, face modeling, and semi-supervised learning.