User talk:Stanleykywu/sandbox

Note: Since I am adding to an existing article, I underlined the parts that I have added.

1. What does the article (or section do well)?

2. What changes would you suggest overall?

3. What is the most important thing that the author could do to improve his/her contribution?

4. Did you glean anything from your classmate's work that could be applicable to your own? If so, let him/her know!

Adversarial Machine Learning Article Peer Review

Sam's peer review --Invay64 (talk) 22:54, 25 October 2021 (UTC)[reply]

1. What does the article or section do well?

You added a lot of good content to the Strategies section that gives it a lot more depth. The additions made to the history section also help give more context for some of the previous statements made in that section and connect it more to current developments.

2. What changes would you suggest?

You might want to change some of the organizational aspects of the article to help give it a better flow. At the moment, the history section seems to start with references in media which could likely be its own section. The Examples portion of this section could also probably be separated out into more of a Real Life Examples that is independent of examples in research. It could also help to introduce more sub-sections in the Attack Modalities section to clarify when one type of attack is a subset of another.

3. What is the most important thing the author could do to improve his/her contribution?

I think that the information you have is strong, but the organization is the main aspect of the article that will help it be clear. I also think more discussion of how large tech companies have started to deal with this issue (since you already mention this in the history section) would help the article.

4. What did you glean from your classmate's work?

The way that you built on top of the already existing sections in the article and added information that flows well with what was already there was well done. So far I've mostly thought about adding my own section but I will try to do more to build more on the existing article as well.

Jaden's Comments

For one, you did a great job updating the general amount of knowledge to the page - both expanding the history section so it includes more modern developments in the field and adding explanations for more strategies to give a more in depth and specific view of adversarial machine learning. I thought the additions to the strategies sections were done especially well - they are written in a way that is very understandable to someone who has background in the field which is good for a more specific topic like this. They give a good overview of each technique without going too much in depth.

The only changes I would suggest so far are organizational based ones. I feel like the divisions between the types of strategies could be made clearer. As it stand right now, it's hard to tell at a glance that Square Attack and HopSkipJumpAttack are both considered Black Box Attacks since they both have the same headings. Same for FGSM and C&W having the same heading as White Box Attacks. Also, I'm not sure the differences between the Specific attack types section and the strategy section but there may be some overlap there (specific attack types lists Membership Inference which sounds similar to the Inference strategy listed under adversarial machine learning strategies.

As mentioned above, clearing up those strategy / attack type sections with different headings or bolding and underlining could go a long way in clarifying which attacks are which.

The use of diagrams or in this case, mathematical models helps to visualize some of the things you describe through text which in turn contributes to the overall understanding of the topic and also makes it more interesting to just look at. Although I won't be able apply the equations specifically to my article, just using other methods to display information such as those diagrams, graphs, etc. will help avoid the "wall of text" problem

Jaden2xU (talk) 22:57, 25 October 2021 (UTC)[reply]

Ethan's Review I think you do a good job of getting into the specifics of how adversarial learning is done vs just explaining what it is. I also think that the technical information provided would give someone who knows about this topic a much greater understanding how to apply the concept rather than the original article.

I suggest if you can include a laymans example of what each of the math formulas given means, but I know they are somewhat complicated.

I think finding a middle ground between a hardcore technical style and a explain-like-I'm-5 style would be good, as for me its a bit too complex to understand the technical parts.

I think I can definitely incorporate a deep dive into the technical sections, like you do, and not defaulting to a very "easy to read" article. I also need to include images! Currently I have none. Eggsbendyboy (talk) 23:06, 25 October 2021 (UTC)[reply]