Auto-Encoding Variational Bayes
By Diederik P. Kingma, Max Welling. International Conference on Learning Representations, 2013
16,816 citations recorded by the source corpus.
Abstract: How can we perform efficient inference and learning in directed probabilistic models, in the presence of continuous latent variables with intractable posterior distributions, and large datasets? We introduce a stochastic variational inference and learning algorithm that scales to large datasets and, under some mild differentiability conditions, even works in the intractable case. Our contributions is two-fold. First, we show that a reparameterization of the variational lower bound yields a lower bound estimator that can be straightforwardly optimized using standard stochastic gradient methods. Second, we show that for i.i.d. datasets with continuous latent variables per datapoint, posterior inference can be made especially efficient by fitting an approximate inference model (also called a recognition model) to the intractable posterior using the proposed lower bound estimator. Theoretical advantages are reflected in experimental results.
Subfields: Artificial Intelligence, Generative Models, Probabilistic & Bayesian Methods
View Paper: https://arxiv.org/abs/1312.6114
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Yoshua Bengio
- Sergey Levine
- Mohit Bansal
- Jianfeng Gao
- Quoc V. Le
- Percy Liang
- Ruslan Salakhutdinov
- Kyunghyun Cho
In Generative Models
- Stefano Ermon
- Yoshua Bengio
- Jun-Yan Zhu
- Arash Vahdat
- Jiaming Song
- Richard Zhang
- Jan Kautz
- Aaron C. Courville
In Probabilistic & Bayesian Methods
- Yoshua Bengio
- Stefano Ermon
- David M. Blei
- Max Welling
- Zoubin Ghahramani
- Richard E. Turner
- Diederik P. Kingma
- Yarin Gal
Citations in the corpus
- Adam: A Method for Stochastic Optimization (2014, methodology): “We vary the β 1 and β 2 when training a variational auto-encoder (VAE) with the same architecture as in (Kingma & Welling, 2013) with a single hidden layer with 500 hidden units with softplus nonlinearities and a 50-dimensional spherical Gaussian latent variable.”
- Generative Adversarial Nets (2014, methodology): “We were unaware at the time we developed this work that Kingma and Welling [18] and Rezende et al.”
- InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets (2016, methodology): “DC-IGN [7] extends this “clamping” idea to VAE and successfully learns graphics codes that can represent pose and light in 3D rendered images.”
- Neural Variational Inference and Learning in Belief Networks (2014, methodology): “In contrast to other work on scaling up variational inference, NVIL can handle both discrete and continuous latent variables (unlikeKingma & Welling(2013); Rezende et al. (2014)) as well variational posteriors with complex dependency structures (unlikeRanganath et al.(2013)).”
- Denoising Diffusion Probabilistic Models (2020, background): “…by optimizing the usual variational bound on negative log likelihood: The forward process variances β t can be learned by reparameterization [31] or held constant as hyperparameters, and expressiveness of the reverse process is ensured in part by the choice of Gaussian conditionals in p θ (…”
- High-Resolution Image Synthesis with Latent Diffusion Models (2021, background): “, imposes a slight KL-penalty towards a standard normal on the learned latent, similar to a VAE [45, 67], whereas VQ-reg.”
- What makes ImageNet good for transfer learning? (2016, background): “The most well-known such criteria are image reconstruction [7, 38, 31, 29, 34, 23] (see [6] for a comprehensive overview) and feature slowness [48, 18].”
- Generating Sentences from a Continuous Space (2015, methodology): “…circum-vents these difficulties using the architecture of a variational autoencoder and takes advantage of recent advances in variational inference (Kingma and Welling, 2015; Rezende et al., 2014) that introduce a practical training technique for powerful neural network generative models with…”
- Breaking the Softmax Bottleneck: A High-Rank RNN Language Model (2017, methodology): “Except for (Chung et al., 2016), these structures all define a continuous latent variable for each step of the RNN computation, and rely on the SGVB estimator (Kingma & Welling, 2013) to optimize a variational lower bound of the log-likelihood.”
- Pixel Recurrent Neural Networks (2016, background): “Most work focuses on stochastic latent variable models such as VAE’s (Rezende et al., 2014; Kingma & Welling, 2013) that aim to extract meaningful representations, but often come with an intractable inference step that can hinder their performance.”
- MADE: Masked Autoencoder for Distribution Estimation (2015, background): “…algorithms for a variety of neural network models have been proposed (Bengio & Bengio, 2000; Larochelle & Murray, 2011; Gregor & LeCun, 2011; Uria et al., 2013; 2014; Kingma & Welling, 2014; Rezende et al., 2014; Bengio et al., 2014; Gregor et al., 2014; Goodfellow et al., 2014; Dinh et al., 2014).”
- The State of Sparsity in Deep Neural Networks (2019, methodology): “For each training step, we sample weights from this distribution and use the reparameterization trick (Kingma & Welling, 2013; Rezende et al., 2014) to differentiate the loss w.r.t. the parameters through the sampling operation.”
- Generating Sentences by Editing Prototypes (2017, methodology): “aining p(xjx0) as a conditional variational autoencoder (C-VAE) (it is conditioned on x0). We maximize the objective via SGD, approximating the gradient using the usual Monte Carlo estimate for VAEs (Kingma and Welling, 2014). Neural editor p edit(xjx0;z): We implement our neural editor as a left-to”
- A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmark (2019, methodology): “For the generative models, we evaluate GANs and VAEs.”
- Learning Representations by Maximizing Mutual Information Across Views (2019, background): “For clipping score s to range [ − c, c ] , we applied the non-linearity s (cid:48) = c tanh( ) , which is linear around 0 and saturates towards ± c . that We suspect there may be interesting formal and practical connections between regularization restricts the variance/range/etc of scores that go in”
- Data-Efficient Image Recognition with Contrastive Predictive Coding (2019, background): “ntations that resemble simple cells in the brain [48]. The idea was later formalized in the context of probabilistic generative models [27], which have become a staple of modern unsupervised learning [28,22,35,56]. A key problem that generative models must overcome, however, is that not all low-leve”
- Weight Normalization: A Simple Reparameterization to Accelerate Training of Deep Neural Networks (2016, methodology): “Next, we test the effect of weight normalization applied to deep convolutional variational auto-encoders (CVAEs) [13, 24, 25], trained on the MNIST data set of images of handwritten digits and the CIFAR-10 data set of small natural images.”