Gaussian BROJA Partial Information Decomposition in Closed Form
Sheelo, Golan and Pakman · arXiv 2610.13464
The outcome is the verdict on the one claim below, under the protocol we fixed in advance. It is not a verdict on the paper as a whole.
Claim tested
Closed-form Gaussian PID is orders of magnitude faster than the iterative solver, with matching atoms.
What we ran and what came back
The authors' released code and bundled baselines, at source dimension 512 on five seeds the authors did not use, on a 4-core cloud CPU. Eigen-PID took about 2.6 seconds per run and Thin-PID between 635 and 4,626 seconds: a geometric-mean speedup of 489 times (interval 164 to 1,453; bound 100). The interval is wide because Thin-PID's runtime varied sevenfold across seeds on this machine; the authors' own median of 210 times lies inside it. The decomposition agreed with Thin-PID to within 0.0001 bits on every seed (tolerance 0.001). An independent calculation of the mutual-information terms matched Eigen-PID to 1e-8 bits. An unmodified upstream copy of Thin-PID ran at the same speed as the bundled one. One dimension, CPU only, default settings.
What this result supports
Closed-form Gaussian PID is a usable fast primitive at these sizes.
What it does not support
Non-Gaussian data; the speedup depends on the solver compared against.
Record
Protocol, deposited before the run: 10.5281/zenodo.23338453
The authors were sent this result. They have not reviewed this listing.
How the protocol works is on the method page. If you are an author of this paper and something here is wrong, write to customers@aegismind.app and we will correct it on this page.