The Variational Homoencoder: Learning to learn high capacity generative models from few examples

By Luke B. Hewitt, Maxwell Nye, Andreea Gane, Tommi S. Jaakkola, Josh Tenenbaum. 2018

70 citations recorded by the source corpus.

Hierarchical Bayesian methods can unify many related tasks (e.g. k-shot classification, conditional and unconditional generation) as inference within a single generative model. However, when this generative model is expressed as a powerful neural network such as a PixelCNN, we show that existing learning techniques typically fail to effectively use latent variables. To address this, we develop a modification of the Variational Autoencoder in which encoded observations are decoded to new elements from the same class. This technique, which we call a Variational Homoencoder (VHE), produces a hierarchical latent variable model which better utilises latent variables. We use the VHE framework to learn a hierarchical PixelCNN on the Omniglot dataset, which outperforms all existing models on test set likelihood and achieves strong performance on one-shot generation and classification tasks. We additionally validate the VHE on natural images from the YouTube Faces database. Finally, we develop extensions of the model that apply to richer dataset structures such as factorial and hierarchical categories.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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

In Computer Vision and Pattern Recognition

Citations in the corpus

Conditional Neural Processes (2018, methodology): “A model that is conceptually very similar to CNPs (and in particular the latent variable version) is the “neural statis-tician” paper (Edwards & Storkey, 2016) and the related variational homoencoder (Hewitt et al., 2018).”
Neural Processes (2018, methodology): “Another variant of the ELBO is presented in the variational homoencoder (Hewitt et al., 2018), a model that is very similar to the neural statistician but uses a separate subset of data points for the context and predictions.”
Generalizing from a Few Examples (2019, methodology): “(VAE) [68] is used in [34, 57, 109], autoregressive model [135] is used in [107], generative adversarial networks (GAN) [46] is used in [159], and a combination of VAE and GAN is proposed in [47].”
Meta-Amortized Variational Inference and Learning (2019, background): “The most relevant prior works include the Neural Statistician (Edwards and Storkey 2016) (NS) and the Variational Homoencoder (Hewitt et al. 2018) (VHE), two very similar models that study inference over sets of observations.”
A Meta-Learning Framework for Generalized Zero-Shot Learning (2019, background): “Finally, although our focus here has been on ZSL and generalized ZSL, our meta-learning based adversarial generation model can be useful for the problem of distribution learning and generation tasks as well [37, 17].”
Variational Few-Shot Learning (2019, methodology): “Recent works [8, 21] have already imposed variational inference on fewshot scenario.”
Attentive Neural Processes (2019, background): “Especially relevant are the Neural Statistician (Edwards & Storkey, 2017) and the Variational Homoencoder (Hewitt et al., 2018) who have a similar permutation invariant encoder (that outputs summaries of a data set), but use local latents on top of a global latent.”
SketchEmbedNet: Learning Novel Concepts by Imitating Drawings (2020, unknown): “(Snell et al., 2017; Lake et al., 2019) Conv4 Omniglot - - 86.30 - RCN (Supervised) (George et al., 2017; Lake et al., 2019) N/A Omniglot - - 92.70 - VHE (Supervised) (Hewitt et al., 2018; Lake et al., 2019) N/A Omniglot - - 81.30 -”