[ExI] Fashion

[Anders Sandberg]

On 2013-05-10 14:59, spike wrote:
>
> EXCELLENT Anders me lad!  Finally!  Someone writes about fashion in a 
> language I can understand!
>

Thanks! I have put up a slightly improved version on my blog:
http://www.aleph.se/andart/archives/2013/05/a_mathematical_model_of_dressing.html

> This is ground-breaking work sir.  If fashion magazines would have 
> articles with this kind of material, I might actually read them, and 
> perhaps put together some spreadsheets.  If the models work right, I 
> could lose my geek status, by becoming a snappy dresser!  Of course it 
> would be done by formula rather than by the traditional means, so that 
> might actually strengthen my status.  It isn't clear how it would work.
>

I think one could capture a lot of simple rules as a kind of 
satisfiability problem. For example, you want your colors to match. 
Matching is not just everything having the same color, but a game where 
colors are supposed to contrast or complement each other. There is a 
fairly well developed theory for harmonious color schemes that can be 
automated (I have seen at least one SIGGRAPH paper about it), so you 
could automate calculation of that. There are similar, but more subtle 
rules about pattern and texture. So combining these rules into an 
evaluation function would allow your wardrobe software to find the 
optimal combination of clothes. The problem has complexity ~N^4 if you 
have N each of socks, pants, shirts and jackets - since a typical N is 
<< 100 a full evaluation seems entirely possible, especially since 
heuristics can quickly prune parts of the search space. It might also 
tell you how many high value combinations are enabled by a new piece of 
clothing you are considering to buy.

I think that kind of software is entirely feasible, would produce decent 
results, yet would be somewhat unsatisfactory. First, entering the 
relevant data would be a chore (if you need the software, how good would 
you be at characterizing the texture of your shirt?) - maybe image 
recognition software and webcam, or some RFID-linked product 
specification data could automate it. Also, we would want to have the 
system recognize what parts of your wardrobe is in the wash basket right 
now. Second, a lot of real style is about sending rather individual 
signals. Of course, you might have the option to add extra rules ("Not 
too strong stripes" in my case, for example), but often it is tricky to 
tell what you actually want if you are not versed in the lingo.

I still think such software would likely be a pretty helpful starting 
point. Just wish I had the time to tinker with it.


-- 
Dr Anders Sandberg
Future of Humanity Institute
Oxford Martin School
Oxford University

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20130517/0dfef4ac/attachment.html>