User talk:MTn05b6nTO

Welcome!

Hello, MTn05b6nTO!

Welcome to Wikipedia, the free encyclopedia.

Getting Started

Tutorial
Learn everything you need to know to get started.

Introduction to contributing

The Teahouse
Ask questions and get help from experienced editors.

The Task Center
Learn what Wikipedians do and discover how to help.

Tips

Don't be afraid to edit! Just find something that can be improved and make it better. Other editors will help fix any mistakes you make.
It's normal to feel a little overwhelmed, but don't worry if you don't understand everything at first—it's fine to edit using common sense.
If an edit you make is reverted, you can discuss the issue at the article's talk page. Be civil, and don't restore the edit unless there is consensus.
Always use edit summaries to explain your changes.
When adding new content to an article, always include a citation to a reliable source.
If you wish to edit about a subject with which you are affiliated, read our conflict of interest guide and disclose your connection.
Have fun! Your presence in the Wikipedia community is welcome.

Happy editing! Cheers, 𝕁𝕄𝔽 (talk) 08:33, 16 October 2023 (UTC)[reply]

Text must faithfully summarise citation

I reverted your edit because you failed to provide a new citation that supports your proposed text. The original (now reinstated) text was an accurate reflection of the cited source (https://www.quantamagazine.org/secret-messages-can-hide-in-ai-generated-media-20230518/). Your change may be valid but it cannot continue to rely on the quantmagazine citation (since it contradicts it). If your text is to stand, it needs a new citation. You would also need to provide evidence that the difficulty has been overcome (or never really existed in the first place).

Welcome again to Wikipedia. --𝕁𝕄𝔽 (talk) 08:33, 16 October 2023 (UTC)[reply]

Hi JMF, thanks for the clarification. The already-cited source material https://arxiv.org/abs/2210.14889) states that any coupling algorithm guarantees perfect security (Theorem 1) and that there are polynomial-time minimum-entropy coupling approximation algorithms that guarantee exact coupling: "While minimum entropy coupling is an NP-hard problem, there exist O(N log N) approximation algorithms (Kocaoglu et al., 2017; Cicalese et al., 2019; Rossi, 2019) that are suboptimal (in terms of joint entropy) by no more than one bit, while retaining exact marginalization guarantees." The Quanta article may be misunderstanding the idea that these approximation algorithms lead to suboptimal information throughput, not suboptimal security. MTn05b6nTO (talk) 21:11, 17 October 2023 (UTC)[reply]

Actually, could you clarify where the Quanta article supports the reinstated text? The quoted text from the article reads to me as supporting my proposed text:

"There are limitations. Cachin pointed out that finding the true minimum entropy coupling is an NP-hard problem, which basically means that the perfect solution is too computationally expensive to be practical, getting back to that issue of efficiency.

Sokota and Schroeder de Witt acknowledge that problem: The optimal coupling would, indeed, be too complicated to compute. But to get around that bottleneck, the authors used an approximating procedure developed by Sokota and Schroeder de Witt (and based on a method introduced by Kocaoglu) that still guarantees security and reasonable efficiency."

Specifically, the part stating that the approximating procedure still guarantees security. MTn05b6nTO (talk) 21:56, 17 October 2023 (UTC)[reply]

Clearly you know more about this topic than I do [not difficult!] so it would be great if you could rewrite the section. The reason I reverted your edit was that you changed the text without providing a new citation, implying in effect that that the existing citation supported your text. (btw, I read the Quant Magazine citation as saying that the NP-hard problem does exist, so it does not seem reasonable to remove that item – how else can we explain what their work aims to circumvent?)

"Specifically, the part stating that the approximating procedure still guarantees security.", that seems to me to be a really important point that really could usefully be elaborated.

I guess you don't have time to learn how to use Wikipedia's technique for WP:citing sources so feel free to just whatever style that you use normally and I will be happy to reshape it afterwards. (I can't find any reference to https://arxiv.org/abs/2210.14889 in the article, btw. If you want to use it, the format is

de Witt, Christian Schroeder; Sokota, Samuel; Kolter, J. Zico; Foerster, Jakob; Strohmeier, Martin (11 Apr 2023). "Perfectly Secure Steganography Using Minimum Entropy Coupling". arXiv:2210.14889 [abs]. ).

I do hope that you can contribute.--𝕁𝕄𝔽 (talk) 12:26, 18 October 2023 (UTC)[reply]

I think the preprint is reference 15 in the current article.

Regarding how to proceed, I think the text I proposed previously is a relatively good one-sentence summary of [14,15]. Explaining the nuance of the complexity results and their implications might require something closer to a paragraph. Do you feel the additional space is merited? If so, I can try drafting a longer revision. MTn05b6nTO (talk) 21:06, 19 October 2023 (UTC)[reply]

@Epachamo: I'm way out of my depth here. This relates to my reversion of MTn05b6nTO's edit https://en.wikipedia.org/wiki/Special:MobileDiff/1180382636 It looks like it is a good faith edit by someone who knows the topic well and perhaps I was being too rigorous.

Could you advise? 𝕁𝕄𝔽 (talk) 22:34, 19 October 2023 (UTC)[reply]

In the meantime, maybe it makes to restore the edit I made? I think it accurately reflects [14,15], while the current text does not accurately reflect either of them. Happy to discuss details further to explain, if that would be helpful. MTn05b6nTO (talk) 21:20, 30 October 2023 (UTC)[reply]

I have restored your version. I'm satisfied that you know what you are talking about. Which of course means that you are most welcome to develop the article further. --𝕁𝕄𝔽 (talk) 22:11, 30 October 2023 (UTC)[reply]