Scalable Joint Models for Reliable Uncertainty-Aware Event Prediction

Hossein Soleimani, James Hensman, Suchi Saria

Research output: Contribution to journalArticle

Abstract

Missing data and noisy observations pose significant challenges for reliably predicting events from irregularly sampled multivariate time series (longitudinal) data. Imputation methods, which are typically used for completing the data prior to event prediction, lack a principled mechanism to account for the uncertainty due to missingness. Alternatively, state-of-the-art joint modeling techniques can be used for jointly modeling the longitudinal and event data and compute event probabilities conditioned on the longitudinal observations. These approaches, however, make strong parametric assumptions and do not easily scale to multivariate signals with many observations. Our proposed approach consists of several key innovations. First, we develop a flexible and scalable joint model based upon sparse multiple-output Gaussian processes. Unlike state-of-the-art joint models, the proposed model can explain highly challenging structure including non-Gaussian noise while scaling to large data. Second, we derive an optimal policy for predicting events using the distribution of the event occurrence estimated by the joint model. The derived policy trades-off the cost of a delayed detection versus incorrect assessments and abstains from making decisions when the estimated event probability does not satisfy the derived confidence criteria. Experiments on a large dataset show that the proposed framework significantly outperforms state-of-the-art techniques in event prediction.

Original languageEnglish (US)
JournalIEEE Transactions on Pattern Analysis and Machine Intelligence
DOIs
StateAccepted/In press - Aug 20 2017

Fingerprint

Joint Model
Uncertainty
Prediction
Time series
Innovation
Decision making
Non-Gaussian Noise
Joint Modeling
Multivariate Time Series
Imputation
Large Data
Longitudinal Data
Optimal Policy
Time Series Data
Missing Data
Gaussian Process
Large Data Sets
Confidence
Costs
Trade-offs

Keywords

  • Computational modeling
  • Data models
  • Detectors
  • Joint Modeling
  • Missing Data
  • Predictive models
  • Reliability
  • Scalable Gaussian Processes
  • Survival Analysis
  • Time Series
  • Time series analysis
  • Uncertainty
  • Uncertainty-Aware Prediction

ASJC Scopus subject areas

  • Software
  • Computer Vision and Pattern Recognition
  • Computational Theory and Mathematics
  • Artificial Intelligence
  • Applied Mathematics

Cite this

Scalable Joint Models for Reliable Uncertainty-Aware Event Prediction. / Soleimani, Hossein; Hensman, James; Saria, Suchi.

In: IEEE Transactions on Pattern Analysis and Machine Intelligence, 20.08.2017.

Research output: Contribution to journalArticle

@article{08b3bf17ee2a4a4e9022ff9d67fe3b08,
title = "Scalable Joint Models for Reliable Uncertainty-Aware Event Prediction",
abstract = "Missing data and noisy observations pose significant challenges for reliably predicting events from irregularly sampled multivariate time series (longitudinal) data. Imputation methods, which are typically used for completing the data prior to event prediction, lack a principled mechanism to account for the uncertainty due to missingness. Alternatively, state-of-the-art joint modeling techniques can be used for jointly modeling the longitudinal and event data and compute event probabilities conditioned on the longitudinal observations. These approaches, however, make strong parametric assumptions and do not easily scale to multivariate signals with many observations. Our proposed approach consists of several key innovations. First, we develop a flexible and scalable joint model based upon sparse multiple-output Gaussian processes. Unlike state-of-the-art joint models, the proposed model can explain highly challenging structure including non-Gaussian noise while scaling to large data. Second, we derive an optimal policy for predicting events using the distribution of the event occurrence estimated by the joint model. The derived policy trades-off the cost of a delayed detection versus incorrect assessments and abstains from making decisions when the estimated event probability does not satisfy the derived confidence criteria. Experiments on a large dataset show that the proposed framework significantly outperforms state-of-the-art techniques in event prediction.",
keywords = "Computational modeling, Data models, Detectors, Joint Modeling, Missing Data, Predictive models, Reliability, Scalable Gaussian Processes, Survival Analysis, Time Series, Time series analysis, Uncertainty, Uncertainty-Aware Prediction",
author = "Hossein Soleimani and James Hensman and Suchi Saria",
year = "2017",
month = "8",
day = "20",
doi = "10.1109/TPAMI.2017.2742504",
language = "English (US)",
journal = "IEEE Transactions on Pattern Analysis and Machine Intelligence",
issn = "0162-8828",
publisher = "IEEE Computer Society",

}

TY - JOUR

T1 - Scalable Joint Models for Reliable Uncertainty-Aware Event Prediction

AU - Soleimani, Hossein

AU - Hensman, James

AU - Saria, Suchi

PY - 2017/8/20

Y1 - 2017/8/20

N2 - Missing data and noisy observations pose significant challenges for reliably predicting events from irregularly sampled multivariate time series (longitudinal) data. Imputation methods, which are typically used for completing the data prior to event prediction, lack a principled mechanism to account for the uncertainty due to missingness. Alternatively, state-of-the-art joint modeling techniques can be used for jointly modeling the longitudinal and event data and compute event probabilities conditioned on the longitudinal observations. These approaches, however, make strong parametric assumptions and do not easily scale to multivariate signals with many observations. Our proposed approach consists of several key innovations. First, we develop a flexible and scalable joint model based upon sparse multiple-output Gaussian processes. Unlike state-of-the-art joint models, the proposed model can explain highly challenging structure including non-Gaussian noise while scaling to large data. Second, we derive an optimal policy for predicting events using the distribution of the event occurrence estimated by the joint model. The derived policy trades-off the cost of a delayed detection versus incorrect assessments and abstains from making decisions when the estimated event probability does not satisfy the derived confidence criteria. Experiments on a large dataset show that the proposed framework significantly outperforms state-of-the-art techniques in event prediction.

AB - Missing data and noisy observations pose significant challenges for reliably predicting events from irregularly sampled multivariate time series (longitudinal) data. Imputation methods, which are typically used for completing the data prior to event prediction, lack a principled mechanism to account for the uncertainty due to missingness. Alternatively, state-of-the-art joint modeling techniques can be used for jointly modeling the longitudinal and event data and compute event probabilities conditioned on the longitudinal observations. These approaches, however, make strong parametric assumptions and do not easily scale to multivariate signals with many observations. Our proposed approach consists of several key innovations. First, we develop a flexible and scalable joint model based upon sparse multiple-output Gaussian processes. Unlike state-of-the-art joint models, the proposed model can explain highly challenging structure including non-Gaussian noise while scaling to large data. Second, we derive an optimal policy for predicting events using the distribution of the event occurrence estimated by the joint model. The derived policy trades-off the cost of a delayed detection versus incorrect assessments and abstains from making decisions when the estimated event probability does not satisfy the derived confidence criteria. Experiments on a large dataset show that the proposed framework significantly outperforms state-of-the-art techniques in event prediction.

KW - Computational modeling

KW - Data models

KW - Detectors

KW - Joint Modeling

KW - Missing Data

KW - Predictive models

KW - Reliability

KW - Scalable Gaussian Processes

KW - Survival Analysis

KW - Time Series

KW - Time series analysis

KW - Uncertainty

KW - Uncertainty-Aware Prediction

UR - http://www.scopus.com/inward/record.url?scp=85028517618&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85028517618&partnerID=8YFLogxK

U2 - 10.1109/TPAMI.2017.2742504

DO - 10.1109/TPAMI.2017.2742504

M3 - Article

JO - IEEE Transactions on Pattern Analysis and Machine Intelligence

JF - IEEE Transactions on Pattern Analysis and Machine Intelligence

SN - 0162-8828

ER -