One of the persistent deficiencies of our Stack is low voting participation. This is the area we most need to improve.

Our Q&A is an informal peer review, that ultimately looks to crowdsource information vetting, ideally by knowledgeable participants. We get a lot of good questions and answers, but the sample size is small in regard to votes.

Yesterday we got a question that received 23 upvotes, and 28 answer votes, in a single day:

Are neural networks prone to catastrophic forgetting?

I'm interested in people's thoughts--theories on how the question attracted this level of attention, the nature of question, etc.

Also, how we can use this type of question to support what seems to be the "meat and potatoes" of the stack, which is more technically specific questions and the math.

i.e. There are only so many unique questions you can ask about theory, philosophy, social impacts, and so forth, but the technical stuff is a boundless wellspring. The technical Q&As also provide the immediate, tangible utility that built the original Stack.


1 Answer 1


First of all, I think this question has attracted a lot of people because it is about neural networks, which is a "hot topic" nowadays, and an issue that neural networks face. Given that a lot of people use and like neural networks and were not aware of this issue, people are probably interested in knowing about this problem and how to solve it.

Furthermore, I think the original title of this question, A flaw with nerual networks?, would not have attracted so many people. I realized that the author of such post, to some extent, was asking about the catastrophic forgetting (or inference) of neural networks, so I changed the title to the current one (which is quite descriptive), which I think I has contributed to the current popularity of the question, given that it contains the known (and maybe mnemonic) technical expression "catastrophic forgetting". In general, I have been trying to edit questions, so that to improve their titles and make them more descriptive, which I think can potentially attract more people. The title A flaw with nerual networks? is not very descriptive, because neural networks might have many flaws. So, I encourage every user to edit questions to make their titles more descriptive of the actual problem.

The given answers are also not very technical or long, so they are accessible or understandable by anyone in the field. Hence we should also strive for simplicity, when possible!

There are other questions that I think should have received a lot more attention and upvotes (for example, Where can I find the proof of the universal approximation theorem?), but I don't know how exactly Stack Exchange tags a question as a "hot" (and maybe this is just my opinion).


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .