Gaussian processes for time-series modelling by S. Roberts, M. Osborne, M. Ebden, S. Reece, N. Gibson, and S. Aigrain Philosophical Transactions A Volume.

Slides:



Advertisements
Similar presentations
Pattern Recognition and Machine Learning
Advertisements

Pattern Recognition and Machine Learning
The Simple Linear Regression Model Specification and Estimation Hill et al Chs 3 and 4.
Philosophical Transactions A
State Estimation and Kalman Filtering CS B659 Spring 2013 Kris Hauser.
Pattern Recognition and Machine Learning
A semi-automated measurement technique for the assessment of radiolucency by E. C. Pegg, B. J. L. Kendrick, H. G. Pandit, H. S. Gill, and D. W. Murray.
Introduction of Markov Chain Monte Carlo Jeongkyun Lee.
Application of interval iterations to the entrainment problem in respiratory physiology by Jacques Demongeot, and Jules Waku Philosophical Transactions.
Introduction to Sampling based inference and MCMC Ata Kaban School of Computer Science The University of Birmingham.
Bayesian Reasoning: Markov Chain Monte Carlo
Autonomous Robot Navigation Panos Trahanias ΗΥ475 Fall 2007.
Predictive Automatic Relevance Determination by Expectation Propagation Yuan (Alan) Qi Thomas P. Minka Rosalind W. Picard Zoubin Ghahramani.
Nonlinear and Non-Gaussian Estimation with A Focus on Particle Filters Prasanth Jeevan Mary Knox May 12, 2006.
Particle Filters for Mobile Robot Localization 11/24/2006 Aliakbar Gorji Roborics Instructor: Dr. Shiri Amirkabir University of Technology.
End of Chapter 8 Neil Weisenfeld March 28, 2005.
Amos Storkey, School of Informatics. Density Traversal Clustering and Generative Kernels a generative framework for spectral clustering Amos Storkey, Tom.
Kernel Methods Part 2 Bing Han June 26, Local Likelihood Logistic Regression.
Particle Filtering. Sensors and Uncertainty Real world sensors are noisy and suffer from missing data (e.g., occlusions, GPS blackouts) Use sensor models.
Bayesian Filtering for Robot Localization
Binary Variables (1) Coin flipping: heads=1, tails=0 Bernoulli Distribution.
Gaussian process regression Bernád Emőke Gaussian processes Definition A Gaussian Process is a collection of random variables, any finite number.
沈致远. Test error(generalization error): the expected prediction error over an independent test sample Training error: the average loss over the training.
Markov Localization & Bayes Filtering
Applications of Bayesian sensitivity and uncertainty analysis to the statistical analysis of computer simulators for carbon dynamics Marc Kennedy Clive.
Model Inference and Averaging
Introduction to MCMC and BUGS. Computational problems More parameters -> even more parameter combinations Exact computation and grid approximation become.
Jamal Saboune - CRV10 Tutorial Day 1 Bayesian state estimation and application to tracking Jamal Saboune VIVA Lab - SITE - University.
Probabilistic Robotics Bayes Filter Implementations.
Stochastic Monte Carlo methods for non-linear statistical inverse problems Benjamin R. Herman Department of Electrical Engineering City College of New.
PATTERN RECOGNITION AND MACHINE LEARNING CHAPTER 3: LINEAR MODELS FOR REGRESSION.
Sea level: measuring the bounding surfaces of the ocean by Mark E. Tamisiea, Chris W. Hughes, Simon D. P. Williams, and Richard M. Bingley Philosophical.
Overview of Supervised Learning Overview of Supervised Learning2 Outline Linear Regression and Nearest Neighbors method Statistical Decision.
Sector and Sphere: the design and implementation of a high-performance data cloud by Yunhong Gu, and Robert L. Grossman Philosophical Transactions A Volume.
Mixture Models, Monte Carlo, Bayesian Updating and Dynamic Models Mike West Computing Science and Statistics, Vol. 24, pp , 1993.
17 May 2007RSS Kent Local Group1 Quantifying uncertainty in the UK carbon flux Tony O’Hagan CTCD, Sheffield.
Mobile Robot Localization (ch. 7)
Hydrocarbon liquefaction: viability as a peak oil mitigation strategy by Mikael Höök, Dean Fantazzini, André Angelantoni, and Simon Snowden Philosophical.
Use of kernel-based Bayesian models to predict late osteolysis after hip replacement by P. Aram, V. Kadirkamanathan, and J. M. Wilkinson Interface Volume.
ECE-7000: Nonlinear Dynamical Systems Overfitting and model costs Overfitting  The more free parameters a model has, the better it can be adapted.
Energy transfer processes along a supramolecular chain of π-conjugated molecules by S. A. Schmid, R. Abbel, A. P. H. J. Schenning, E. W. Meijer, and L.
Lecture 2: Statistical learning primer for biologists
Dropout as a Bayesian Approximation
Bayesian Travel Time Reliability
Gaussian Processes For Regression, Classification, and Prediction.
Particle Filtering. Sensors and Uncertainty Real world sensors are noisy and suffer from missing data (e.g., occlusions, GPS blackouts) Use dynamics models.
Particle Filtering. Sensors and Uncertainty Real world sensors are noisy and suffer from missing data (e.g., occlusions, GPS blackouts) Use sensor models.
1 Chapter 8: Model Inference and Averaging Presented by Hui Fang.
The Unscented Particle Filter 2000/09/29 이 시은. Introduction Filtering –estimate the states(parameters or hidden variable) as a set of observations becomes.
CS Statistical Machine learning Lecture 25 Yuan (Alan) Qi Purdue CS Nov
Fitting normal distribution: ML 1Computer vision: models, learning and inference. ©2011 Simon J.D. Prince.
Sparse Approximate Gaussian Processes. Outline Introduction to GPs Subset of Data Bayesian Committee Machine Subset of Regressors Sparse Pseudo GPs /
RECITATION 2 APRIL 28 Spline and Kernel method Gaussian Processes Mixture Modeling for Density Estimation.
CS Statistical Machine learning Lecture 7 Yuan (Alan) Qi Purdue CS Sept Acknowledgement: Sargur Srihari’s slides.
Probability Theory and Parameter Estimation I
Multiple Imputation using SOLAS for Missing Data Analysis
Alan Qi Thomas P. Minka Rosalind W. Picard Zoubin Ghahramani
Michael Epstein, Ben Calderhead, Mark A. Girolami, Lucia G. Sivilotti 
CSCI 5822 Probabilistic Models of Human and Machine Learning
CSCI 5822 Probabilistic Models of Human and Machine Learning
Volume 6, Issue 5, Pages e13 (May 2018)
by K. Jun Tong, Sebastián Duchêne, Simon Y. W. Ho, and Nathan Lo
Biointelligence Laboratory, Seoul National University
Phase Transitions in Biological Systems with Many Components
Volume 111, Issue 2, Pages (July 2016)
Finding Periodic Discrete Events in Noisy Streams
Volume 23, Issue 21, Pages (November 2013)
Overview: Chapter 2 Localization and Tracking
Evolution-informed forecasting of seasonal influenza A (H3N2)‏
Uncertainty Propagation
Presentation transcript:

Gaussian processes for time-series modelling by S. Roberts, M. Osborne, M. Ebden, S. Reece, N. Gibson, and S. Aigrain Philosophical Transactions A Volume 371(1984): February 13, 2013 ©2013 by The Royal Society

A simple example of curve fitting. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

The conceptual basis of GPs starts with an appeal to simple multi-variate Gaussian distributions. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

The change in distributions on x1 and x2 is here presented in a form more familiar to time-series analysis. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

(a) The posterior distribution (the black line showing the mean and the grey shading ±σ) for a 10 day example, with observations made at locations 2, 6 and 8. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

(a–c) Functions drawn from a GP with a squared exponential covariance function with output scale h=1 and length scales λ=0.1 (a), 1 (b), 10 (c). S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

(a) Given six noisy data points (error bars are indicated with vertical lines), we are interested in estimating a seventh at x*=0.2. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

(a) Estimation of y* (solid line) and ±2σ posterior deviance for a function with short-term and long-term dynamics, and (b) long-term dynamics and a periodic component. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

A simple example of a GP applied sequentially. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

The GP is run sequentially making forecasts until a new datum is observed. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

Random draws from GPs with different kernels. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

Example covariance functions (a) for the modelling of data with changepoints and associated draws (b) from the resultant GPs, indicating what kind of data that they might be appropriate for. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

The effect of including a simple mean function. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

Samples (dots) obtained by optimizing the log-likelihood (grey curve) using a global optimizer, and the maximum-likelihood approximation (vertical line) of the likelihood surface. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

Samples obtained by taking draws from the posterior using a Markov chain Monte Carlo method. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

A set of samples that would lead to unsatisfactory behaviour from simple Monte Carlo. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

Bayesian quadrature fits a GP to the integrand, and thereby performs inference about the integral. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

Prediction and regression of tide height data for (a) independent and (b) multi-output GPs. (Online version in colour.). S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

(a) Comparison of active sampling of tide data using independent and (b) multi-output GPs. Note that, in the case of multi-output GPs, one sensor reading (Sotonmet) slightly leads the other readings and is hence sampled much more frequently. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

Prediction for the Nile dataset using input-scale changepoint covariance (a) and the corresponding posterior distribution for time since the changepoint (b). S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

Online (sequential) one-step predictions (a) and posterior for the location of changepoint for the Dow–Jones industrial average data using an output-scale changepoint covariance (b). S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

Predictive distribution for a quasi-periodic GP model using a mixed SE and RQ kernel, trained and conditioned on observations made with the 0.8 m Automated Patrol Telescope [24] using the Strömgren b and y filters. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society

As an example of a complex mean function, we here model data from an exoplanet transit light curve. S. Roberts et al. Phil. Trans. R. Soc. A 2013;371: ©2013 by The Royal Society