---
title: 'GLACIAL: Granger Causality Paper Summary'
disqus: hackmd
---

GLACIAL: Granger Causality Paper Summary
===
By Arpan Neupane & Aaditya Panchal

General Overview
---
The paper introduces GLACIAL (Granger and Learning-based Causallty Analysis for Longitudinal studies), a method designed to address challenges of causal discovery where data is irregular and/or sparse.

Key Points
---
>Background
>-
>1. Granger Causality (GC):
>a. Widely used to detect casual relationships in time series data.
>b. Works well with densely sampled data from single systems but is not well-suited for longitudinal studies with multiple individuals and sparse observations.
>2. Challenges in Longitudinal Studies:
>a. Sparse observations: longitudinal studies often track individuals over few timepoints, making it difficult to infer accurate causal relationships.
>b. Nonlinear dynamics: relationships between variables are often nonlinear, and traditional GC methods may not effectively capture these dynamics.
>c. Missing Data: real-world longitudinal data frequently have missing values, complicating causal analysis.
>d. Direct vs Indirect Causes: GC methods traditionally do not distingusih well between direct and indirect causal effects.

GLACIAL Approach
---
>Objective: Enhance causal discovery in longitudinal studies by combining GC with machine learning techniques. GLACIAL was developed in order to handle complexities such as individual specific dynamics and missing data. This is because traditional granger causality struggles in longitudinal analysis due to sparse and irregular sampling, missing data, etc.
>
>Methods:
>-
>- Multi-task neural network: GLACIAL uses a neural network to handle nonlinear dynamics and large numbers of variables.
>- Input Feature Dropout: This technique is used to manage missing values and test causal relationships by stimulating the effect of missing data.
>- Train-test setup: Treats each individual's trajectory as an independent sample and uses average prediction on hold-out individuals to infer causal relationships.
>
:::success
>Advantages:
>-
>- Handling Missing Values: GLACIAL's use of dropout helps address missing data issues effectively.
>- Scalability: Can manage a large number of variables and nonlinear relationships better than traditional GC methods.
>- Direct vs. Indirect Causes: Includes post-processing steps to clarify the directionality of causal relationships and resolve ambiguities.
>- GLACIAL uses a train-test method to see if it can accurately predict the outcomes of some people's data. First, it leaves aside some data, in the training phase. After it is trained with the rest of the data, it tests itself, on the left out data, to see if it can predict the causal relationships in what happens.
:::