'Uniformization'? I am looking for a better term for what I call 'uniformification', where I change data to make it more close to uniformly distributed.
I am doing a project in which I try to make the output of a neural network uniformly distributed over the output space. Let's call the neural network $f(x | \theta)$, where $\theta$ contains the weights and other tunable parameters. What I do is to change the parameters of the network $\theta$ such that the output is closer to a uniform distribution. You could say I am making the output more uniform by modifying the function $f(x|\theta)$.
How would I call this process? 
I was thinking of the term 'uniformification', but it sounds kind of weird. What could be a better term for this? Or is there already a term for this in the literature?
PS: in no case will I use a transformation on either $x$, $\theta$ or on the output $f(x|\theta)$ in order to make the output more uniform. I am only adjusting the weights $\theta$ in order to achieve that.
Edit: According to this post on the suffixes -ise, -ate and -ify, the term should actually be 'uniformization'.
 A: It's a transformation. By analogy with other procedures, you might call it transformation to uniform scores. "Uniformification" is certainly an ugly word and I strongly recommend against it. 
But the best name depends at least partly on how you do it. 
In particular, the transformation to ranks, with averaging for ties, achieves a uniform distribution, except in so far as there are spikes and gaps as a side-effect of the ties. Unless you violate a principle that identical values mean identical transformed values, it is difficult to see how you could make a distribution more nearly uniform. 
Naturally, scaling ranks to fit in $[0, 1]$ or any other interval is just cosmetic or a matter of convenience. 
UPDATE (updated) The idea, and the terminology, of transformations within statistics are completely compatible with the ideas that transformations can be applied in sequence; that a given transformation can be a composite of others; and that a transformation can be a family, i.e. tuned by one or more parameters. An example of moderate historical importance is the so-called angular transformation, which is arcsine of square root of proportions. (Note that in this case, the proportions could, and indeed usually would, be transformations of some original data.) I say within statistics but I would assert that this idea and terminology are consistent with mathematics generally. In sum, what you are doing is a sequence of transformations; the target is a uniform distribution; it doesn't need a special or novel name, and none obviously exists at present. 
