07:55:26	 Anon. ResNet101:	morning everyone
07:55:34	 Anon. Soma:	hiiii
07:56:45	 Anon. Asynchronous Update:	yes
08:09:04	 Anon. ResNet101:	that the distribution is stationary?
08:09:50	 Kushal Saharan (TA):	What do you mean by stationary in this context?
08:10:05	 Anon. ResNet101:	the distribution does not change over time
08:11:00	 Kushal Saharan (TA):	Yes, it doesn’t for the purpose of the examples you will encounter in this lecture
08:11:18	 Anon. Kaiming:	Most boring = maximum likelihood?
08:12:14	 Kushal Saharan (TA):	True if you find MLE to be boring:D
08:25:18	 Anon. Noise:	For the movie rating example, all users share a Gaussian distribution?
08:26:11	 Anon. Bias:	That is our assumption
08:27:54	 Kushal Saharan (TA):	I would suggest you pay attention to the MLE process with incomplete data. Specific distributional assumptions can change but the process remains the same
08:29:13	 Anon. Kaiming:	Are these sub-Gaussians assumed to be independently distributed?
08:30:34	 Anon. ASR:	how to estimate the number of needed k sub-Gaussians?
08:31:17	 Kushal Saharan (TA):	K is like a hyper-parameter. This is like the ‘k’ in k-means
08:31:42	 Kushal Saharan (TA):	There maybe heuristics for it but it isn’t generated by the process itself
08:32:04	 Anon. ASR:	tx
08:33:58	 Kushal Saharan (TA):	@Yiwei, no relationship is assumed between the parameters of the gaussian. Once you select a gaussian, the distribution is not dependent on anything else
08:34:43	 Anon. Kaiming:	Thank you
09:06:00	 Anon. Instance:	Assuming that we are estimating the third components of three-component vector, we will fixed the first two components. Right?
09:07:44	 Kushal Saharan (TA):	Assuming the the first two parts are complete (i.e provided in the data) they are trivially fixed
09:08:15	 Anon. Instance:	thx
09:17:29	 Anon. ResNet101:	I think this is supposed to be a select all question
09:17:53	 Anon. ResNet101:	it was presente to me a choose one
09:18:00	 Anon. ResNet101:	presented*