scholarly journals Estimating feedforward and feedback effective connections from fMRI time series: Assessments of statistical methods

2019 ◽  
Vol 3 (2) ◽  
pp. 274-306 ◽  
Author(s):  
Ruben Sanchez-Romero ◽  
Joseph D. Ramsey ◽  
Kun Zhang ◽  
Madelyn R. K. Glymour ◽  
Biwei Huang ◽  
...  
Keyword(s):  
Time Series ◽  
Statistical Methods ◽  
Permutation Test ◽  
Causal Structure ◽  
Multiple Model ◽  
New Methods ◽  
Test Models ◽  
Non Gaussian ◽  
True Structure ◽  
Better Than

We test the adequacies of several proposed and two new statistical methods for recovering the causal structure of systems with feedback from synthetic BOLD time series. We compare an adaptation of the first correct method for recovering cyclic linear systems; Granger causal regression; a multivariate autoregressive model with a permutation test; the Group Iterative Multiple Model Estimation (GIMME) algorithm; the Ramsey et al. non-Gaussian methods; two non-Gaussian methods by Hyvärinen and Smith; a method due to Patel et al.; and the GlobalMIT algorithm. We introduce and also compare two new methods, Fast Adjacency Skewness (FASK) and Two-Step, both of which exploit non-Gaussian features of the BOLD signal. We give theoretical justifications for the latter two algorithms. Our test models include feedback structures with and without direct feedback (2-cycles), excitatory and inhibitory feedback, models using experimentally determined structural connectivities of macaques, and empirical human resting-state and task data. We find that averaged over all of our simulations, including those with 2-cycles, several of these methods have a better than 80% orientation precision (i.e., the probability of a directed edge is in the true structure given that a procedure estimates it to be so) and the two new methods also have better than 80% recall (probability of recovering an orientation in the true structure).

scholarly journals Causal Discovery of Feedback Networks with Functional Magnetic Resonance Imaging

2018 ◽  
Author(s):  
R Sanchez-Romero ◽  
J.D. Ramsey ◽  
K. Zhang ◽  
M. R. K Glymour ◽  
B Huang ◽  
...  
Keyword(s):  
Time Series ◽  
Causal Structure ◽  
New Methods ◽  
Test Models ◽  
Inhibitory Feedback ◽  
Non Gaussian ◽  
Recall Probability ◽  
Feedback Networks ◽  
Direct Feedback ◽  
Better Than

AbstractWe test the adequacies of several proposed and two new statistical methods for recovering the causal structure of systems with feedback that generate noisy time series closely matching real BOLD time series. We compare: an adaptation for time series of the first correct method for recovering the structure of cyclic linear systems; multivariate Granger causal regression; the GIMME algorithm; the Ramsey et al. non-Gaussian methods; two non-Gaussian methods proposed by Hyv¨arinen and Smith; a method due to Patel, et al.; and the GlobalMIT algorithm. We introduce and also compare two new methods, the Fast Adjacency Skewness (FASK) and Two-Step, which exploit non-Gaussian features of the BOLD signal in different ways. We give theoretical justifications for the latter two algorithms. Our test models include feedback structures with and without direct feedback (2-cycles), excitatory and inhibitory feedback, models using experimentally determined structural connectivities of macaques, and empirical resting state and task data. We find that averaged over all of our simulations, including those with 2-cycles, several of these methods have a better than 80% orientation precision (i.e., the probability a directed edge is in the true generating structure given that a procedure estimates it to be so) and the two new methods also have better than 80% recall (probability of recovering an orientation in the data generating model). Recovering inhibitory direct feedback loops between two regions is especially challenging.

Intervention Modeling

2017 ◽  
Author(s):  
Richard McCleary ◽  
David McDowall ◽  
Bradley J. Bartos
Keyword(s):  
Time Series ◽  
Autocorrelation Function ◽  
Moving Average ◽  
Intervention Model ◽  
Noise Component ◽  
Autoregressive Integrated Moving Average ◽  
Noise Function ◽  
Modeling Strategy ◽  
The Impact ◽  
True Structure

The general AutoRegressive Integrated Moving Average (ARIMA) model can be written as the sum of noise and exogenous components. If an exogenous impact is trivially small, the noise component can be identified with the conventional modeling strategy. If the impact is nontrivial or unknown, the sample AutoCorrelation Function (ACF) will be distorted in unknown ways. Although this problem can be solved most simply when the outcome of interest time series is long and well-behaved, these time series are unfortunately uncommon. The preferred alternative requires that the structure of the intervention is known, allowing the noise function to be identified from the residualized time series. Although few substantive theories specify the “true” structure of the intervention, most specify the dichotomous onset and duration of an impact. Chapter 5 describes this strategy for building an ARIMA intervention model and demonstrates its application to example interventions with abrupt and permanent, gradually accruing, gradually decaying, and complex impacts.

scholarly journals Adversarial Unsupervised Representation Learning for Activity Time-Series

2019 ◽  
Vol 33 ◽  
pp. 834-841 ◽  
Author(s):  
Karan Aggarwal ◽  
Shafiq Joty ◽  
Luis Fernandez-Luque ◽  
Jaideep Srivastava
Keyword(s):  
Time Series ◽  
Chronic Conditions ◽  
Empirical Evaluation ◽  
Representation Learning ◽  
Time Segment ◽  
Linear Classifiers ◽  
Daily Routines ◽  
Learning Technique ◽  
Prevention And Cure ◽  
Better Than

Sufficient physical activity and restful sleep play a major role in the prevention and cure of many chronic conditions. Being able to proactively screen and monitor such chronic conditions would be a big step forward for overall health. The rapid increase in the popularity of wearable devices pro-vides a significant new source, making it possible to track the user’s lifestyle real-time. In this paper, we propose a novel unsupervised representation learning technique called activ-ity2vecthat learns and “summarizes” the discrete-valued ac-tivity time-series. It learns the representations with three com-ponents: (i) the co-occurrence and magnitude of the activ-ity levels in a time-segment, (ii) neighboring context of the time-segment, and (iii) promoting subject-invariance with ad-versarial training. We evaluate our method on four disorder prediction tasks using linear classifiers. Empirical evaluation demonstrates that our proposed method scales and performs better than many strong baselines. The adversarial regime helps improve the generalizability of our representations by promoting subject invariant features. We also show that using the representations at the level of a day works the best since human activity is structured in terms of daily routines.

LEARNING DYNAMICS AND NONLINEAR MISSPECIFICATION IN AN ARTIFICIAL FINANCIAL MARKET

2009 ◽  
Vol 13 (5) ◽  
pp. 625-655 ◽  
Author(s):  
Christophre Georges ◽  
John C. Wallace
Keyword(s):  
Time Series ◽  
Financial Market ◽  
Consistency Condition ◽  
Nonlinear Structure ◽  
State Variable ◽  
Bounded Memory ◽  
Long Run ◽  
Finite Memory ◽  
Aggregate Dynamics ◽  
True Structure

In this paper, we explore the consequence of learning to forecast in a very simple environment. Agents have bounded memory and incorrectly believe that there is nonlinear structure underlying the aggregate time series dynamics. Under social learning with finite memory, agents may be unable to learn the true structure of the economy and rather may chase spurious trends, destabilizing the actual aggregate dynamics. We explore the degree to which agents' forecasts are drawn toward a minimal state variable learning equilibrium as well as a weaker long-run consistency condition.

scholarly journals PNA Predictability at Various Time Scales

2013 ◽  
Vol 26 (22) ◽  
pp. 9090-9114 ◽  
Author(s):  
Waqar Younas ◽  
Youmin Tang
Keyword(s):  
Time Scales ◽  
Initial Conditions ◽  
Prediction Skill ◽  
Multiple Model ◽  
Coupled Models ◽  
Potential Predictability ◽  
Effective Measure ◽  
Dispersion Component ◽  
Significant Difference ◽  
Better Than

Abstract In this study, the predictability of the Pacific–North American (PNA) pattern is evaluated on time scales from days to months using state-of-the-art dynamical multiple-model ensembles including the Canadian Historical Forecast Project (HFP2) ensemble, the Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER) ensemble, and the Ensemble-Based Predictions of Climate Changes and their Impacts (ENSEMBLES). Some interesting findings in this study include (i) multiple-model ensemble (MME) skill was better than most of the individual models; (ii) both actual prediction skill and potential predictability increased as the averaging time scale increased from days to months; (iii) there is no significant difference in actual skill between coupled and uncoupled models, in contrast with the potential predictability where coupled models performed better than uncoupled models; (iv) relative entropy (REA) is an effective measure in characterizing the potential predictability of individual prediction, whereas the mutual information (MI) is a reliable indicator of overall prediction skill; and (v) compared with conventional potential predictability measures of the signal-to-noise ratio, the MI-based measures characterized more potential predictability when the ensemble spread varied over initial conditions. Further analysis found that the signal component dominated the dispersion component in REA for PNA potential predictability from days to seasons. Also, the PNA predictability is highly related to the signal of the tropical sea surface temperature (SST), and SST–PNA correlation patterns resemble the typical ENSO structure, suggesting that ENSO is the main source of PNA seasonal predictability. The predictable component analysis (PrCA) of atmospheric variability further confirmed the above conclusion; that is, PNA is one of the most predictable patterns in the climate variability over the Northern Hemisphere, which originates mainly from the ENSO forcing.

Sign in / Sign up

Export Citation Format

Share Document