Evolutionary Architecture Search for Chlorophyll-a Prediction in Lakes using Sentinel-2
Komurcu and Petkevicius · arXiv 2610.10496
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
NAS-1 has higher test AUC than the baseline MLP.
What we ran and what came back
The authors' released code on 10 seeds they did not use. The searched architecture (NAS-1) reached a mean test AUC of 0.818 against 0.792 for the paper's baseline network: a difference of +0.026 (interval +0.005 to +0.051, resampling the 129 test lakes), against a margin of zero. The authors' own seeds reproduced their reported figures to within 0.003. The lower end of the interval is small. In a second arm, reported separately and not part of the verdict, we gave the baseline network 300 random hyperparameter draws; NAS-1 was then ahead by +0.009 (interval -0.007 to +0.028), which is inconclusive. So this run does not show that the architecture search beats a well-tuned baseline. One dataset and one split.
What this result supports
The published gap reproduces.
What it does not support
That NAS beats a well-tuned baseline: arm B could not tell.
Record
Protocol, deposited before the run: 10.5281/zenodo.23352138
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.