A Neural Sparse Graphical Model for Variable Selection and Time-Series Network Analysis

Author(s): Shenggang Li

Originally published on Towards AI.

A Unified Adjacency Learning and Nonlinear Forecasting Framework for High-Dimensional Data

This member-only story is on us. Upgrade to access all of Medium.

Imagine a spreadsheet with rows of timestamps and columns labeled x_1, x_2, …. Each x_n​ might represent a product’s sales, a stock’s price, or a gene’s expression level. But these variables rarely evolve in isolation — they often influence one another, sometimes with notable time lags. To handle these interactions, we need a robust Time-Series Network that models how each variable behaves in relation to the others. This paper focuses on precisely that objective.

For instance, last month’s dip in x_1​ could trigger a spike in x_2​ this month, or perhaps half these columns are simply noise that drowns out the key relationships I want to track.

My quest was to figure out how to select the most important variables and build a reliable model of how each x_m​ depends on the others over time. For example, is x_m​ mostly driven by x_1​ and x_2​ from the previous day, or does it depend on all variables from the previous week? I looked into various ideas, like Graph Neural Networks (GNNs) to capture who influences whom, structural modeling for domain-specific equations, or more exotic approaches like… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI