[FoRK] Capitalism, Contracts and Cronyism
sdw at lig.net
Wed Aug 27 12:55:41 PDT 2008
Jeff Bone wrote:
> On Aug 27, 2008, at 2:14 PM, Stephen Williams wrote:
>> Even with a neural net or other machine learning system, you should
>> be able to feed it ranges of inputs to derive approximate formulas
>> that explain its behavior. While that may take a long time to
>> iterate through all variables, it is possible.
> We've had this conversation before, but you're just not correct about
> this, your statement is too broad. It may be true in certain limited
> cases, but it's not true in general. This is a topic I happen to live
> with day to day, so I know whereof I speak on this.
> The map is not the territory. And the mapping between the map and the
> territory may be complex enough that it cannot be sensibly expressed
> in any finite terms less complicated than the map itself, the
> territory, or both.
True. It may not be. And finding a mathematical representation of the
full model may be NP-Complete. Obviously, a neural net trained on the
trajectory of a bouncing ball has an output that can be quantified into
a formula easily. At least some of what is learned should have similar
meta characteristics, if such exist in the model.
More information about the FoRK