Source-linked AI summary
Information Flows? A Critique of Transfer Entropies
Ryan G. James, Nix Barnett, James P. Crutchfield
TL;DR
The paper asks whether transfer entropy’s predictive influence necessarily represents information flow. Through examples of two- and three-series interactions, it shows that transfer-like entropies can misattribute flow or influence.
Problem
The paper examines whether transfer entropy’s predictive influence necessarily represents direct information flow.
Source: p. 2Method
The paper uses straightforward examples involving two and three time series to examine how transfer and causation entropies represent influence.
Source: pp. 1, 3Results
The examples show transfer entropy can overestimate information flow, while transfer and causation entropies can assign indirect or joint influence to individual sources.
Source: pp. 1–3Takeaways & Limitations
Transfer-entropy network interpretations can misidentify complex-system structure when standard networks represent relationships as dyadic and omit polyadic dependencies.
Source: pp. 1, 4, 5
Abstract
from arXiv · showhide
A central task in analyzing complex dynamics is to determine the loci of information storage and the communication topology of information flows within a system. Over the last decade and a half, diagnostics for the latter have come to be dominated by the transfer entropy. Via straightforward examples, we show that it and a derivative quantity, the causation entropy, do not, in fact, quantify the flow of information. At one and the same time they can overestimate flow or underestimate influence. We isolate why this is the case and propose several avenues to alternate measures for information flow. We also address an auxiliary consequence: The proliferation of networks as a now-common theoretical model for large-scale systems, in concert with the use of transfer-like entropies, has shoehorned dyadic relationships into our structural interpretation of the organization and behavior of complex systems. This interpretation thus fails to include the effects of polyadic dependencies. The net result is that much of the sophisticated organization of complex systems may go undetected.