%0 Journal Article %1 grathwohl2020cutting %A Grathwohl, Will %A Wang, Kuan-Chieh %A Jacobsen, Jorn-Henrik %A Duvenaud, David %A Zemel, Richard %D 2020 %K bayesian energy-models generative-models uncertainty %T Cutting out the Middle-Man: Training and Evaluating Energy-Based Models without Sampling %U http://arxiv.org/abs/2002.05616 %X We present a new method for evaluating and training unnormalized density models. Our approach only requires access to the gradient of the unnormalized model's log-density. We estimate the Stein discrepancy between the data density p(x) and the model density q(x) defined by a vector function of the data. We parameterize this function with a neural network and fit its parameters to maximize the discrepancy. This yields a novel goodness-of-fit test which outperforms existing methods on high dimensional data. Furthermore, optimizing $q(x)$ to minimize this discrepancy produces a novel method for training unnormalized models which scales more gracefully than existing methods. The ability to both learn and compare models is a unique feature of the proposed method.

@article{grathwohl2020cutting, abstract = {We present a new method for evaluating and training unnormalized density models. Our approach only requires access to the gradient of the unnormalized model's log-density. We estimate the Stein discrepancy between the data density p(x) and the model density q(x) defined by a vector function of the data. We parameterize this function with a neural network and fit its parameters to maximize the discrepancy. This yields a novel goodness-of-fit test which outperforms existing methods on high dimensional data. Furthermore, optimizing $q(x)$ to minimize this discrepancy produces a novel method for training unnormalized models which scales more gracefully than existing methods. The ability to both learn and compare models is a unique feature of the proposed method.}, added-at = {2020-02-14T16:12:05.000+0100}, author = {Grathwohl, Will and Wang, Kuan-Chieh and Jacobsen, Jorn-Henrik and Duvenaud, David and Zemel, Richard}, biburl = {https://www.bibsonomy.org/bibtex/2d2b270c2cd1f2cb238df1cc8a67bb97b/kirk86}, description = {[2002.05616] Cutting out the Middle-Man: Training and Evaluating Energy-Based Models without Sampling}, interhash = {8e1d16c79e8c628775c9f18a5760bd1e}, intrahash = {d2b270c2cd1f2cb238df1cc8a67bb97b}, keywords = {bayesian energy-models generative-models uncertainty}, note = {cite arxiv:2002.05616}, timestamp = {2020-02-14T16:12:05.000+0100}, title = {Cutting out the Middle-Man: Training and Evaluating Energy-Based Models without Sampling}, url = {http://arxiv.org/abs/2002.05616}, year = 2020 }