The Singularity

nbro

01:56

Here is a post that needs to be updated to use MathJax (or Latex). If someone wants to update it, this would be appreciated.

5

Q: TRPO/PPO importance sampling term in loss function

In the Trust-Region Policy Optimisation (TRPO) algorithm (and subsequently in PPO also), I do not understand the motivation behind replacing the log probability term from standard policy gradients with the importance sampling term of the policy output probability over the old policy output pro...

John Doucette

02:10

@nbro Done. Thanks for finding that, and for raising all those close votes. There was only one I disagreed with, which I think could be a good example of a "programming" question that I think is on-topic (ai.stackexchange.com/questions/16557/…). Although the poster includes code, and the code is using a modern library, the problem the asker is having is really rooted in an AI-related misunderstanding.

I think we can help with that kind of question because we have the expertise to know that the asker's problem is algorithmic, even though they choose to post a Python description of their model, rather than a textual one.

nbro

@JohnDoucette Then I would suggest reformulating the question such that the user does not simply ask "What's wrong with this code?"

Anyway, thanks for updating that post. There would also be the need to update the answers, but do not feel forced to do it only because I'm saying it. Anyway, if I am not wrong, there should be a tool that automatically converts pictures with formulas to Latex code.

John Doucette

@nbro Hmm. In fact, the user asks "What's wrong with my design?", and describes an observed behavior that is unexpected, not a bug in the code. I think that's the key distinction: they are asking about the design/model, and the code is just a handy way to express their model formally for us to look at.

That sounds like a good tool. I do not know of one

I can do the answers too. I did not think to look at them.

nbro

I don't see a big difference between "What's wrong with my design?" and "What's wrong with this code?". Maybe I've not yet caught it, but the asker is not telling anywhere what the expected output should be, apart from saying that the network is not learning.

However, I'm not sure whether this type of question is suitable or not for this website. Wouldn't this just be the ideal question for Data Science SE?

John Doucette

02:26

I guess the way I think about it is as follows. Suppose that the asker had said "I've implemented a neural network 4 layers deep, with sigmoid activation functions. I used a trusted library, so I'm certain that the implementation is exactly as I describe. When I train the network, it fails to learn any patterns from my dataset. What might be causing this?"

That phrasing seems on-topic to me.

And the difference here is that they showed us the code, rather than hiding it.

and perhaps weren't as eloquent in explaining what they wanted

nbro

Well, in that case, I would suggest you edit the post and add your formulation of the question.

John Doucette

There do indeed appear to be tools for automatic conversion to Latex (e.g. snapcraft.io/install/mathpix-snipping-tool/mint). Thanks for suggesting that.

nbro

In that case, I think it may be on-topic. However, in general, I will be flagging questions as off-topic, if they ask just "What is wrong with this code?", because, in those cases, the users want us to debug the source code, while not fully describing the problem. We should demand a full description of the problem. The code should be there only to help to solve the problem, as you correctly say.

John Doucette

You make a good point, and I have rewritten the post. I think part of the issue with these kinds of posts is that the posters often include code because they do not quite know what is wrong, and seek to provide more information. Maybe a good practice would be to flag such posts for moderator attention. Perhaps code is a smell here.

nbro

Additionally, we should also demand a MWE (i.e. code that you can copy, paste and execute), whenever they provide source code.

I will update my on-page suggestion with these details later

John Doucette

02:37

That's a good idea. MWEs are always helpful things to have.

nbro

@JohnDoucette Well, apparently, in this case (if you look at the history of the post), the asker didn't originally provide the source code. The source code was provided by the asker because another user requested it

John Doucette

Hmm. That's an interesting wrinkle. I think the original question was not answerable, because the poster didn't know what was wrong. When they posted their code, it was easy for me to see that the critical issue was their choice of activation function, and I could then re-write the question.

nbro

I think it is fine to request the source code (if it helps to solve the actual theoretical/conceptual problem), we just need to ask for an MWE.

@JohnDoucette What do you think about the new paragraph related to implementation questions?

> Implementation questions in the context of understanding the theoretical topics are on-topic. For example, if a theoretical topic is described by a certain mathematical formula and you want to understand how a certain implementation is related to the formula, then your question is on-topic.

> As a rule of thumb, if you can describe your problem without the source code and if you think that a solution to your problem can be given without the source code, then your question is on-topic. The source code can be provided to further clarify the issue, but you should provide a Minimal, Reproducible Example.

John Doucette

03:11

@nbro I like that a lot actually. I think that should be our policy.

nbro

14:14

You should flag the following question as a duplicate.

8

Q: Will quantum computing have any kind of effect on the development of AI?

Recently, according to some reports Google achieved something called 'Quantum Supremacy'. Whether its true or not remains to be seen. But my question is does Quantum Computers or the principle they work on has any effect on AI. We know that according to Shor's Algorithm Quantum Computing can red...

training hardware hardware-evaluation quantum-computing

nbro

16:30

@DuttaA Jaden has just removed some comments I've flagged, so he's still doing some work, but he's not been very active anyway! Maybe we need more active moderators!!

DuttaA

17:07

@nbro apparently when I checked Jaden's profile earlier it showed 5 day's ago last seen. Although I don't mind such a long break but I think since we are going through so many changes might as well try 2 new mods.

nbro

17:26

@DuttaA Yes, I agree with you. We definitely need more active moderators!

DukeZhou

21:02

@duttaA @nbro The interest check meta has sufficient candidates that I'm confident I can get them to expand the election to two moderators slots.

I'll put in the request and let you know when we have a decision.

nbro

Yes, that would be a good idea!

DukeZhou

21:29

@nbro I was thinking on the idea of DS ⊂ AI, and remembered Automated Theorem Proving/Automated Reasoning

nbro

@DukeZhou What do you want to say exactly?

DukeZhou

Just that automated theorem solving is not statistical--an area of contemporary AI outside of data science.

nbro

Ha, yes, there are non-statistical approaches/solutions in AI, like all symbolic stuff

DukeZhou

TO ALL ACTIVE MEMBERS OF THE COMMUNITY: it's been pointed out we need to revise our help page. Please review this meta post as we will likely be making alterations!

2

Q: What should the on-topic page look like?

Similarly to What should the AI.SE Site Description be? and after the discussions On-topic and off-topic pages need to be clarified and Who decides and writes the on-topic and off-topic pages?, I think it is time to vote for a clearer and updated version of the on-topic page, which users (but esp...

discussion scope site-promotion

2

A: What should the on-topic page look like?

What topics can I ask about here? If you have a question about theoretical, philosophical, historical, social and certain technical and academic aspects of AI, then you are probably in the right place to ask your question! Notes Before posting, please, look around to see if your question has...

1

A: What should the on-topic page look like?

I adjusted @nbro's answer to remove the parts I thought were too restrictive. AI is a broad field, and the whitelist of "on-topic" areas omits a huge number of topics which are certainly within AI (consider, for contrast, the topics that are present at AAAI this year alone, all of which are activ...

nbro

23:35

The community is working better now! We're no more afraid of closing questions that need to be closed.

Transcript for