From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Wed Dec 03 2003 - 09:09:01 MST
Eliezer S. Yudkowsky wrote:
> Mitchell Porter wrote:
>
>> http://citeseer.nj.nec.com/bishop94training.html
> According to Bishop94 above, for example, it amounts
> to adding an extra term in the loss function, and it is possible to get
> the same benefit without some dangers by modifying the algorithm
> directly instead of training with noise.
Er, sorry, not the loss function. On closer examination it looks like a
function that represents the cost of different trained networks - it looks
like a term that approximates the prior probability of a network.
-- Eliezer S. Yudkowsky http://intelligence.org/ Research Fellow, Singularity Institute for Artificial Intelligence
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:43 MDT