Date:	12 Jun 2022
To:	"Christopher Gorham Lester" lester@hep.phy.cam.ac.uk
From:	"Hendrik Blockeel" hendrik.blockeel@cs.kuleuven.be
Subject:	#MACH-D-22-00480 - Decision
Dear Dr Lester,

With regret, I must inform you that your manuscript, "Stressed GANs snag desserts a.k.a Spotting Symmetry Violation with Symmetric Functions" cannot be accepted for publication in Machine Learning. 

Please see the editor's comments, below.

Thank you for considering Machine Learning.

Sincerely,

Hendrik Blockeel
Editor-in-Chief 
Machine Learning


Comments for the Author:

The paper proposes an interesting idea, but for this to be publishable in Machine Learning, a much more thorough investigation would be needed. Machine Learning does not only impose criteria on originality and novelty, but also on scientific depth and thoroughness. How good are the obtained results really? We see a visualization but no objective evaluation. How does the method compare to more straightforward ways of solving the problem, e.g., density estimation? (Estimate the density, then check whether for a given x, Px lies in a high-density region or not). It is not unlikely that the proposed approach works better, because it focuses on this specific problem, but we don't know for sure: this is not evaluated. Also, is there a relationship with GANs? The title does use the term, and there seems to be some similarity at a superficial level, in the sense that the adversary in a GAN is trained in a somewhat similar manner as the network proposed here (trying to distinguish two different distributions). It is stated in a footnote that the relationship is not seriously considered here, but it probably should be, if only by simply pointing out the differences. More generally, work published in Machine Learning should be firmly embedded in the state of the art, so the relationship between his work and other methods that might be used for achieving the same goal must be discussed. In terms of writing style, the paper is somewhat frivolous and not always to the point; there are a good number of irrelevant comments (e.g., the explanation of "chiral" in 2.2). I understand that the authors have tried to make this text very accessible, but conciseness and focus are also important. In this respect, there seems to be a bit of a mismatch with what Machine Learning expects. Finally, as Machine Learning focuses on fundamental contributions with general applicability, it would be good to demonstrate the potential of the proposed method for a wider range of applications.
 Overall, I would say that the paper presents an interesting idea, but the paper would need to be developed in much more detail (more discussion of how this relates to the state of the art, more thorough evaluation) before publication in Machine Learning can be considered.