00:47:20	Anze Slosar:	But any Taylor expansions has a radius of convergence. How can we know that the expansion is valid?
00:49:02	Jan:	Is it actually a taylor expansion, or just a polynomial? That's not the same thing.
00:53:15	Van Dam, Hubertus:	Just a comment: I also think that f is typically a simple function with sensible derivatives (in some cases even a finite number of non-zero derivatives).
00:55:56	Milind Diwan:	Could you quickly review the 3 conditions again
01:07:00	Van Dam, Hubertus:	When you consider “cat” pictures then you look at only binary classifiers. If you want to classify pictures of “cat”, “dog”, “horse”, “raccoon”, etc. then you have two options:
01:07:15	Van Dam, Hubertus:	1. A multivalue classifier
01:07:30	Van Dam, Hubertus:	2. Multiple binary classifiers.
01:07:40	Van Dam, Hubertus:	Can you comment on that?
01:14:30	mccorkle:	Are you making assumptions about threshold functions in these nodes (ie that they are sigmoid and well-behaved?)
01:45:30	mccorkle:	The early paper(s) on multi-layer perceptrons nicely show what the nodes are doing -  basically cutting the parameter space into half-plains, returning 1 if the vector position is above the cutoff and 0 if lower
01:51:30	mccorkle:	A great intro book is “Perceptrons” by Marvin Minksy and Seymour Papert (1969)
02:02:25	Milind Diwan:	Thanks George, Daniel. Thanks for this glimpse of analysis.
02:03:17	Van Dam, Hubertus:	Thanks Daniel.