# Any ideas how to proceed with interpreting this expected loss?

For this post, $$\Gamma$$ is an expectation operator, and we have two distributions $$S$$ and $$T$$. $$(x_S,y_S)\sim S$$ and $$(x_t,y_t)\sim T$$.

Define $$\Gamma^*_f = \arg \min_{\Gamma} \Gamma l(f(x_t),y_s)$$ and for now, let $$l(w,z)=(w-z)^2$$

I'm trying to interpret $$\Gamma_f^* d(y_s,y_t)$$ as it comes up in a bound I came up with recently in my research. I just don't see any natural way of doing so. Does anyone have any insights? Or at least a nice idea for a simple case to begin with? I've tried looking at linear classes of functions for $$f$$, but nothing has come of it.

• Why was this downvoted. I was asking for general ideas and strategies about how to approach this. – Glassjawed Dec 21 '18 at 21:23
• I didn't downvote, but for me it's not really clear what taking $\arg\min$ over expectation operators is defined to be. Further the indexing of your variables is very confusing. Why are there tuples if you only use one component in your problem? Also, $f$ is undefined. And in general the setting is not clear. What are the spaces of the distributions etc. – Steve Dec 22 '18 at 0:48