NAOMI: Non-Autoregressive Multiresolution Sequence Imputation

NeurIPS 2019  ยท  Yukai Liu, Rose Yu, Stephan Zheng, Eric Zhan, Yisong Yue ยท

Missing value imputation is a fundamental problem in spatiotemporal modeling, from motion tracking to the dynamics of physical systems. Deep autoregressive models suffer from error propagation which becomes catastrophic for imputing long-range sequences. In this paper, we take a non-autoregressive approach and propose a novel deep generative model: Non-AutOregressive Multiresolution Imputation (NAOMI) to impute long-range sequences given arbitrary missing patterns. NAOMI exploits the multiresolution structure of spatiotemporal data and decodes recursively from coarse to fine-grained resolutions using a divide-and-conquer strategy. We further enhance our model with adversarial training. When evaluated extensively on benchmark datasets from systems of both deterministic and stochastic dynamics. NAOMI demonstrates significant improvement in imputation accuracy (reducing average prediction error by 60% compared to autoregressive counterparts) and generalization for long range sequences.

PDF Abstract NeurIPS 2019 PDF NeurIPS 2019 Abstract

Datasets


  Add Datasets introduced or used in this paper
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Multivariate Time Series Imputation Basketball Players Movement NAOMI Path Length 0.573 # 1
OOB Rate (10^โˆ’3) 1.733 # 2
Step Change (10^โˆ’3) 2.565 # 1
Path Difference 0.581 # 2
Player Distance 0.423 # 4
Multivariate Time Series Imputation PEMS-SF NAOMI L2 Loss (10^-4) 3.54 # 1

Methods


No methods listed for this paper. Add relevant methods here