BRUNO: A Deep Recurrent Model for Exchangeable Data
arXiv:1802.07535
Abstract
We present a novel model architecture which leverages deep learning tools to perform exact Bayesian inference on sets of high dimensional, complex observations. Our model is provably exchangeable, meaning that the joint distribution over observations is invariant under permutation: this property lies at the heart of Bayesian inference. The model does not require variational approximations to train, and new samples can be generated conditional on previous samples, with cost linear in the size of the conditioning set. The advantages of our architecture are demonstrated on learning tasks that require generalisation from short observed sequences while modelling sequence variability, such as conditional image generation, few-shot learning, and anomaly detection.
arXiv commentNIPS 2018
Cited by in corpus (10)
- Hopfield Networks is All You Need
- Janossy Pooling: Learning Deep Permutation-Invariant Functions for Variable-Size Inputs
- Equivariant Energy Flow Networks for Jet Tagging
- SurVAE Flows: Surjections to Bridge the Gap between VAEs and Flows
- Guided Variational Autoencoder for Disentanglement Learning
- Exchangeable Neural ODE for Set Modeling
- DuMLP-Pin: A Dual-MLP-dot-product Permutation-invariant Network for Set Feature Extraction
- Set Flow: A Permutation Invariant Normalizing Flow
- Learning Permutation Invariant Representations using Memory Networks
- Kernel Mean Matching for Content Addressability of GANs