22:44:32	  Anon. Baum:	School of Athenes
22:44:40	  Anon. Bigelow:	vitcan
22:44:46	  Anon. Centre:	school of athens
22:44:50	  Anon. Centre:	by raphael
22:46:41	  Haoxuan Zhu (TA):	Guys, if you want, you can hold space bar to temporarily unmute yourselves to ask/answer questions
22:47:03	  Anon. Phillips:	Ring the bell, dog salivates
22:53:52	  Anon. Bartlett:	100 billion
22:53:56	  Anon. Bellefield:	80 billion
22:53:58	  Anon. Murray:	15 billion
22:53:59	  Anon. Friendship:	1 trillion
22:54:15	  Anon. Bartlett:	Trillion
22:56:02	  Anon. Forward:	x86 or ar!m
22:56:07	  Anon. Phillips:	Von neumann
22:56:09	  Anon. Murray:	von neuman
22:56:13	  Anon. Beacon:	von neuman
22:56:14	  Anon. Ellsworth:	Von Neumann
22:56:19	  Anon. Fifth:	von neumann
22:56:25	  Anon. Baum:	Program in memory
22:59:40	  Nour Ali (TA):	30 seconds left
23:06:01	  Anon. Bellefonte:	Is inhibitory synapse analogous to bias?
23:07:45	  Jinhyung David Park (TA):	Not necessarily - don't think there is a direct correlation
23:08:31	  Anon. Bellefonte:	got it.
23:15:12	  Anon. Beacon:	the weight always increases
23:15:16	  Anon. Wightman:	It doesn’t learn when x doesn’t excite y
23:15:16	  Anon. Hobart:	No way to decrease intensity
23:17:19	  Nour Ali (TA):	30 seconds left
23:28:25	  Anon. S. Highland:	the invention of perceptron is attributed to hebb or Rosenblatt?
23:32:19	  Anon. Smithfield:	and
23:32:21	  Anon. Liberty:	AND
23:32:21	  Anon. Wilkins:	and]
23:32:22	  Anon. Penn:	and
23:32:22	  Anon. Tech:	and
23:32:22	  Anon. Northumberland:	and
23:32:23	  Anon. Myrtle:	and
23:32:27	  Anon. Bigelow:	and
23:32:32	  Anon. Darlington:	xor
23:32:41	  Anon. Beechwood:	not x2
23:32:44	  Anon. Walnut:	Not x1
23:32:44	  Anon. Northumberland:	Not x1
23:32:46	  Anon. Frew:	not x1
23:32:46	  Anon. Forbes:	Not x1
23:33:44	  Haoxuan Zhu (TA):	@xyz I think the answer is Rosenblatt. But his model uses (is inspired by) the findings of Hebb
23:38:26	  Anon. Walnut:	What is MLP?
23:38:36	  Anon. Morewood:	Multi-layer perceptrons
23:38:38	  Anon. Wilkins:	In case of overlapping networks, wouldn't treating them as separate and then using OR be computationally redundant?
23:40:24	  Nour Ali (TA):	30 seconds left
23:42:42	  Jinhyung David Park (TA):	@abc maybe in some cases it can be more computationally efficient to separate the two parts, but might have to just think about it on a case-by-case basis
23:43:54	  Anon. Murdoch:	Does more layers of perceptron imply a better model or is there a limit after which the performance doesn’t improve?
23:44:26	  Anon. N.Craig:	If we want continuous, why don't we just use sigmoid instead of threshold?
23:46:08	  Jinhyung David Park (TA):	@efg That's a good question! We'll talk about this more in later lectures, and you'll experience it in your homework. Some general intuition may be that yes, more layers can theoretically capture more information, but sometimes that makes optimization harder leading to worse performance (but also sometimes easier leading to better learning and generalization). This is still an ongoing direction of research
23:47:00	  Anon. Grandview:	@klm My understanding is that sigmoid would only give us between -1 and 1, whereas here we are trying to regress on more general values
23:47:48	  Anon. Shady:	Could someone explain the answer to the poll again? Thank you!
23:47:51	  Jinhyung David Park (TA):	@fgh So "continuous" in the previous example wasn't just saying that the output itself is continuous, but rather whether an MLP can model a continuous function (a specific value for each input). The threshold activation there can be thought of as "hard coding" or "memorizing" a specific output for each input, not necessarily having much to do with whether the output of each perceptron is continuous or not
23:47:58	  Anon. Northumberland:	Can you please post the question in the last poll in chat?
23:48:20	  Jinhyung David Park (TA):	We'll upload all the polls & answers online after the lecture
23:48:46	  Jinhyung David Park (TA):	(We'll also upload the chat, I believe)
23:50:39	  Anon. Ivy:	May I ask that whether students from other sections (section B) could attend the lecture synchronously sometimes instead of watching the video for attendance grade?
23:50:45	  Anon. Bellefonte:	The answer to last poll is one because the weight is -2. Is it correct?
23:51:07	  Anon. Bellefonte:	Ignore
23:51:14	  Anon. S. Aiken:	thank you!
23:51:18	  Jinhyung David Park (TA):	@asd yes
23:51:29	  Jinhyung David Park (TA):	The poll is on the piazza lecture 1 discussion thread!