Скачать или смотреть Assimilative Causal Inference: Identifying Causes Backwards from Observed Effects

(Greek and English subtitles are available; click on the CC button to activate.)

―――― Abstract ――――
Causal inference is fundamental across scientific disciplines, yet existing methods struggle to capture instantaneous, time-evolving causal relationships in complex, high-dimensional systems. Assimilative causal inference (ACI) is a new framework that leverages data assimilation to trace causes backwards from observed effects. ACI reformulates causality as an inverse problem of uncertainty quantification rather than assessing forward influence. It uniquely identifies dynamic causal structures in real time without requiring observations of candidate causes, while scaling efficiently to high dimensions. Crucially, it facilitates mathematically rigorous measures for the forward and backward causal influence range (CIR) of a relationship; the forward CIR quantifies the temporal impact of a cause, while the backward CIR traces the onset of triggers for an observed effect, thus characterizing causal predictability and attribution of outcomes over time, respectively.

―――― Video Chapters ――――
00:00 - Introduction
00:45 - Traditional Causal Inference Methods
01:53 - The Assimilative Causal Inference (ACI) Framework
03:46 - A Brief Overview of Data Assimilation
04:39 - ACI as a Bayesian Inverse Problem
05:22 - Conditional ACI
05:41 - Causal Influence Ranges in ACI
06:17 - Conclusion

_________________________________________________________

► Acknowledgements: My academic advisor is Prof. Nan Chen - https://people.math.wisc.edu/~nchen29. I'm supported as a research assistant under his grants. Animations were made using Manim Community Edition and DaVinci Resolve.

► More Information: https://mariosandreou.short.gy/ACI

► Assimilative Causal Inference Preprint (Co-authored with Prof. Nan Chen and Prof. Erik Bollt): https://doi.org/10.48550/arXiv.2505.1...

► Causal Influence Range Preprint (Co-authored with Prof. Nan Chen): https://doi.org/10.48550/arXiv.2510.2...

► References and Sources:

● Background Music – "Vibing Over Venus", Kevin MacLeod (https://incompetech.com) | Licensed under Creative Commons: By Attribution 3.0 (http://creativecommons.org/licenses/b...)

● Detailed View of Arctic Sea Ice – NASA image (NASA Identifier: ge_07370) by Glenn Research Center, based on Landsat-7 data from the Global Land Cover Facility (https://commons.wikimedia.org/wiki/Fi...)

● Satellite 3D Model – By printable_models on free3d.com (https://free3d.com/3d-model/satellite...)

● Natural Earth Texture with Edited Clouds – Tom Patterson (https://www.shadedrelief.com)

● Hurricane Katrina Image - NASA image of Hurricane Katrina on August 28, 2005 (https://en.wikipedia.org/wiki/Hurrica...)

► Personal Website: https://mariosandreou.short.gy/Homepage

Corrections:
02:20 Strictly speaking, for our setting, the diffusion feedback matrix Σˣ in the observable process needs to be independent of the unobserved state y. This ensures a well-posed data assimilation problem where the conditional distribution of y given the data of x (posterior) contains all available information about y. This is a classical necessary assumption in nonlinear filtering (e.g., see "The Oxford Handbook of Nonlinear Filtering").