# What to do about obsolete, or out-of-date questions, specifically this one about neural networks?

Some questions have answers which are obsolete, or outdated. One prominent example is How to choose the number of hidden layers and nodes in a feedforward neural network? This question has an answer with an incredible (for stats.SE) 605 upvotes. The question is among one of the most up-voted questions on our site.

However, the time that this question was asked (2010), was just before neural networks had their recent renaissance. It would not be until 2012 that the AlexNet paper (Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton, "ImageNet Classification with Deep Convolutional Neural Networks") used CNNs for the ImageNet task and vastly out-performed their competitors, sparking a resurgence of interest in neural networks generally, convolutional neural networks specifically, and the technique of multi-layered networks along with it.

In other words, the current state of the art has advanced well beyond the suggestions in most of the answers to this question. The question itself, I believe, is implicitly asking about simple feed-forward neural networks because CNNs, RNNs, residual networks and other exotic architectures had not yet experienced this explosion in attention. Related neural network techniques, such as , hadn't even been invented yet. (Mikolov et al published in 2013.)

In light of the deficiencies in the answers to this question, and perhaps the limitations of the question itself (due to its place in time), what, if anything, should be done to direct readers to more recent, and more relevant, answers to the question?

These are two options that have occurred to me. There are probably more.

1. One option is to simply use the SE machinery as it is to write a new answer and hope that intrepid readers will make it all the way down to the bottom of the list to find recent information. Or that they'll sort by recent... and scroll past the Accepted answer pinned to the top.
2. Ask a new question which is deliberately distinct. "How do I choose the right number of layers and neurons in light of the many recent advancements in neural network architectures since 2011?" This question, lacking the upvotes and visibility that comes from 10ish years of accumulated hyperlinks, might be harder to find but would report more recent information. This question, because it deliberately asks about what's changed in the intervening time, cannot be closed as a duplicate of the older question.
• In either case we could perhaps, subject to agreement here on Meta, prepend or append a note to the question explaining the circumstances. Jan 30 at 16:39
• Sometimes CMs will give Mods access to special notices. The World Building SE has a custom post notice to draw attention to the fact that some questions request answers based on hard science.
– Sycorax Mod
Jan 30 at 16:42
• That sounds useful. We'd need a post here on Meta tagged with feature-request to discuss it - alert CMs. Jan 30 at 17:45
• +1 because I find this an important problem, but I have little idea how to solve it. The problem is strongly related to the simplisitic voting system that is based on simply acquiring a lot of votes. The voting is not a rating but only indirectly measures quality by popularity. This is not sustainable for very old questions where new answers are not gonna catch up 600 votes which has little to do with popularity or quality. Possibly the problem is not that big. It is maybe only a few questions with highly voted but problematic answers.... Jan 30 at 23:05
• If you ask a new question which is essentially the same and answer it then surely the old one can be closed as a duplicate of the new one? Or do I not understand the rules about duplicates? Jan 31 at 16:02
• @mdewey Well, that's a bit subtle. Some questions have answers which are essentially unchanged over time -- the answer states a proof, and math hasn't changed, so the proof is still valid. But NNs are an active area of research, so a person could argue that old answers are interesting for historical reasons, or that old answers were valid at the time they were written. This is one of the areas where I want to solicit feedback -- what does "duplicate" mean in this setting? SO faces a similar, but distinct, problem with software versioning (new software obsoletes answers).
– Sycorax Mod
Jan 31 at 16:05
• I think making the old a duplicate of the new might be possible here, & might be a good solution. Feb 1 at 12:22
• SO's Outdated Answers Project seems relevant, but it's not clear if/when the resulting features will be available to the SE network at large (or if those features will even be generalizable). The announcement includes a section about how the discovery/testing efforts will be limited solely to SO for the time being.
– tdy
Mar 26 at 4:53
• Currently they're A/B testing some vote-decaying functions for a trending sort option that would theoretically help to surface updated answers to old questions.
– tdy
Mar 26 at 4:54

If the question allows it, then a new question can be made and a reference can be added to the old question to point to the new question. In the case of questions that make no sense at all because of developments in technology they could (and maybe should) be locked (and left undeleted for historical reasons).

• What exactly is the connection between the headline and the rest of the answer? Jan 31 at 10:50
• @RichardHardy I should improve that headline a bit more. I wrote it down too swiftly. It relates a bit to the idea of managing the questions which already happens by generating canonical questions. An alternative is that several questions could be sort of 'bundled' by making references between them. The new question becomes the 'main' question and the older question could either be edited with a reference to the new question, or it could be even marked as a duplicate. Jan 31 at 12:19

Personally I prefer your option 1 relying on existing SE machinery. As a user in search of any SE answer, you should look at both the date of the question and answer. A 10 year old question/answer may be out of date, so the onus is on the user to find the answer that is most relevant. A newer answer that is better/more relevant can make its way toward the top and positive comments on the quality of the new answer can guide users to its accuracy.

I've responded to old popular questions that had high-rated original accepted answers, and I've seen my new answer slowly march its way toward the top over years. The new/better answer may never be accepted ('tis the case with all good answers), but a combination of submission dates, votes, and comments should be sufficient to guide users to the most timely answer.

I also don't have an issue with your solution 2, so long as sufficient time has passed between original question and new question (AT LEAST 5 years perhaps?).

Eventually the activity of making new questions and answers decreases (a lot of questions have simply already been asked, some even several times) and the activity should be replaced by curating the answers and questions.

### Community Wiki

This can be done in a Wikipedia style. Instead of many multiple questions and answers, which creates a chaotic database where only a layer of votes creates an overview, we could also strive for having less posts, which are edited by multiple people instead of each writing their own separate two cents of contribution.

It is a bit confusing on the platform here.

• On the one hand there is an individualistic style of writing posts where individuals each write their own posts and the contributions are being mixed in some sort of ranking of posts by means of votes. Editing by others is only done to a limited extent. There is some sort of ownership of the post because votes are being 'collected' by the person that wrote the original post. Not only are the posts being ranked but also the people that write those posts.

• On the other hand there is in principle no objection to editing another person's post. Anybody can edit a question and/or answer and you do not even need an account for this. In some cases this is explicitly promoted by turning posts into wikis.

Possibly we could turn older posts with many votes into community wikis. The voting, to measure popularity, has become deprecated for such questions and answers. Improvements should be made to the existing questions and answers, rather than adding a new answer.

A major problem with this approach is that many old posts are of high quality and moving towards a community style of answers may actually reduce the overall quality. In addition it will remove the individual style of answers and turn it into writing-by-committee with little variation. Stackexchange is not a Wikipedia.

• I think this brings in a related but distinct issue, which is how much credit you get for improving existing threads. Short answer is None (once your reputation is high enough to edit freely), apart from perhaps bringing badges in its wake. Implication: If you care about getting reputation for work of this kind, you should devote your efforts to asking and especially answering new questions. I can't sense that SE cares about any of this as much as some of us do. Its business model seems to hinge on continually attracting new users. Jan 31 at 17:12