Generative Adversarial Nets
Goodfellow, I. J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y. (2014). Generative adversarial networks.
Summary
The document presents a novel framework for estimating generative models through an adversarial process, involving the simultaneous training of a generative model and a discriminative model. The generative model aims to capture the data distribution while the discriminative model assesses the probability that a sample originates from the training data. This approach is characterized as a minimax two-player game, where the unique solution leads to the generative model recovering the training data distribution. The framework allows for efficient training using backpropagation without the need for Markov chains or approximate inference during sample generation, demonstrating its potential through various experiments.
Keywords
Generative Models
Adversarial Process
Deep Learning
Backpropagation
Discriminative Models
Main claims
A new framework for estimating generative models via an adversarial process is proposed.
The framework involves simultaneous training of a generative model G and a discriminative model D.
The training procedure maximizes the probability of D making a mistake.
A unique solution exists in the space of arbitrary functions G and D, allowing G to recover the training data distribution.
The system can be trained using backpropagation without the need for Markov chains or unrolled approximate inference networks.