Causal Discovery and Forecasting in Nonstationary Environments with - - PowerPoint PPT Presentation

▶
causal discovery and forecasting in nonstationary
SMART_READER_LITE
LIVE PREVIEW

Causal Discovery and Forecasting in Nonstationary Environments with - - PowerPoint PPT Presentation

Jun 15, 2023 224 likes •326 views

Causal Discovery and Forecasting in Nonstationary Environments with State - Space Models Biwei Huang 1 , Kun Zhang 1 , Mingming Gong 1,2 , Clark Glymour 1 1. Department of Philosophy, Carnegie Mellon University 2. Department of Biomedical

slide-1
SLIDE 1

Causal Discovery and Forecasting in Nonstationary Environments with State-Space Models

Biwei Huang1, Kun Zhang1, Mingming Gong1,2, Clark Glymour1

  • 1. Department of Philosophy, Carnegie Mellon University
  • 2. Department of Biomedical Informatics, University of Pittsburgh

1

slide-2
SLIDE 2

Two tasks:

  • 1. Identify time-varying

causal relations

  • 2. Forecast the values of

variables of interest

  • Forecasting benefits from causal knowledge
  • Each causal module changes independently
  • Causal knowledge makes the forecasts more interpretable

GDP Inflation Economic growth Unemployment

slide-3
SLIDE 3

Time-varying causal model:

3

with

  • and change over time

where

slide-4
SLIDE 4

Causal Model, Identifiability, and Estimation

  • Goal: Find time-varying causal relations & make prediction
  • Causal relations change over time
  • Model: causal coefficients modeled by autoregressive

models

  • Identifiability: The causal model identifiable if the

underlying causal structure is acyclic

  • Model Estimation: A specific nonlinear state-space model
  • Estimated by Stochastic approximation EM with

Conditional particle filter

slide-5
SLIDE 5

Forecasting with time-varying causal model

  • T

reat forecasting as a Bayesian inference problem in the causal model

  • Metropolis-Hastings to forecast

5

slide-6
SLIDE 6

500 1000 1500 2000

Sample size (a) Only b changes

0.3 0.5 0.7 0.9

F1 score

500 1000 1500 2000

Sample size (b) Both b and 2 change

0.3 0.5 0.7 0.9

F1 score

Ours LiNGAM IB MC CD-NOD 500 1000 1500 2000

Sample size (a) Only b changes

0.3 0.4 0.5 0.6 0.7 0.8

RMSE

500 1000 1500 2000

Sample size (b) Both b and 2 change

0.8 1 1.2 1.4 1.6 1.8

RMSE

Ours Lasso Window Kalman SSM(CPF) GP

Causal discovery: Forecasting:

Ours: highest F1 score! Ours: lowest RMSE!

6

slide-7
SLIDE 7

Macroeconomics data

(quarterly data, 1965-2017, USA)

RMSE of the forecasts on inflation (2007 - 2017).

GDP Inflation Economic growth Unemployment

slide-8
SLIDE 8

Conclusion

  • A unified framework for causal discovery and forecasting
  • Establish the identifiability results, even when data is

conditional Gaussian

Future work

  • Improve the scalability
  • Nonlinear causal relationships, partially observable processes,

and causal models with instantaneous cycles…

8

Poster #73