BRUNO: A Deep Recurrent Model for Exchangeable Data
This work addresses the challenge of scalable and exact Bayesian inference for exchangeable data, benefiting researchers and practitioners in machine learning, particularly in domains requiring robust probabilistic modeling with limited data.
The authors tackled the problem of performing exact Bayesian inference on sets of high-dimensional, complex observations by introducing a novel deep recurrent model that is provably exchangeable, enabling efficient conditional sampling with linear cost. They demonstrated its advantages in tasks like conditional image generation, few-shot learning, and anomaly detection, showing improved generalization from short observed sequences.
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.