[DLFL22] Rephrase \beta description in week 7

[pscollins]

"More fluctate" is somewhat difficult to parse.

[Atcold]

I have no idea what "fluctuate" means…

[pscollins]

Would something like "greater variance" be more precise? Or even just "less smooth"?

I think the original, "more fluctuate model," is a bit difficult to follow. The direct fix, "more fluctuating model," I find hard to interpret -- I'm not clear if it means that the ripples in the surface have a higher amplitude or a higher frequency. My understanding is that it's really talking about frequency -- more ripples, but the maximum height of any ripple is not necessarily larger -- which is why I suggested "a model that fluctuates more."

[Atcold]

Since what the student wrote about this section makes no sense, I've gone back to Yann's lecture recording.

The caption reads as follow:

0:30:53.400,0:30:55.400 The β parameter is kind of arbitrary

0:30:55.690,0:31:00.450 It's the way you calibrate your probabilities as a function of your energies, so

0:31:01.030,0:31:06.660 The larger the β, the more sort of binary your probability will be for a given energy function

0:31:08.710,0:31:12.270 If β is very very large, it is basically just the

0:31:13.180,0:31:14.860 the E(x,y)

0:31:14.860,0:31:20.640 for the y that produces the lowest energy that will have high probability and everything else will have very low probability and

0:31:20.980,0:31:26.159 For small β for a small β then you get kind of a smoother distribution

0:31:26.680,0:31:28.390 Okay

0:31:28.390,0:31:32.099 Β in physics terms is akin to an inverse temperature

0:31:33.310,0:31:36.900 Okay, so the β goes to infinity is equal to zero temperature

[pscollins]

How about this? I changed it to talk about argmax explicitly.

[Atcold]

Thanks!