CMU Advanced NLP 2024 (1) Introduction to NLP/transcript.vtt

WEBVTT

00:00:01.280 --> 00:00:06.759
so the class today is uh introduction to

00:00:04.680 --> 00:00:09.480
natural language processing and I'll be

00:00:06.759 --> 00:00:11.200
talking a little bit about you know what

00:00:09.480 --> 00:00:14.719
is natural language processing why we're

00:00:11.200 --> 00:00:16.720
motivated to do it and also some of the

00:00:14.719 --> 00:00:18.039
difficulties that we encounter and I'll

00:00:16.720 --> 00:00:19.880
at the end I'll also be talking about

00:00:18.039 --> 00:00:22.519
class Logistics so you can ask any

00:00:19.880 --> 00:00:25.439
Logistics questions at that

00:00:22.519 --> 00:00:27.720
time so if we talk about what is NLP

00:00:25.439 --> 00:00:29.320
anyway uh does anyone have any opinions

00:00:27.720 --> 00:00:31.439
about the definition of what natural

00:00:29.320 --> 00:00:33.239
language process would be oh one other

00:00:31.439 --> 00:00:35.680
thing I should mention is I am recording

00:00:33.239 --> 00:00:38.600
the class uh I put the class on YouTube

00:00:35.680 --> 00:00:40.520
uh afterwards I will not take pictures

00:00:38.600 --> 00:00:41.920
or video of any of you uh but if you

00:00:40.520 --> 00:00:44.719
talk your voice might come in the

00:00:41.920 --> 00:00:47.440
background so just uh be aware of that

00:00:44.719 --> 00:00:49.000
um usually not it's a directional mic so

00:00:47.440 --> 00:00:51.559
I try to repeat the questions after

00:00:49.000 --> 00:00:54.079
everybody um but uh for the people who

00:00:51.559 --> 00:00:57.680
are recordings uh listening to the

00:00:54.079 --> 00:00:59.320
recordings um so anyway what is NLP

00:00:57.680 --> 00:01:03.120
anyway does anybody have any ideas about

00:00:59.320 --> 00:01:03.120
the definition of what NLP might

00:01:06.119 --> 00:01:09.119
be

00:01:15.439 --> 00:01:21.759
yes okay um it so the answer was it

00:01:19.240 --> 00:01:25.759
helps machines understand language

00:01:21.759 --> 00:01:27.920
better uh so to facilitate human human

00:01:25.759 --> 00:01:31.159
and human machine interactions I think

00:01:27.920 --> 00:01:32.759
that's very good um it's

00:01:31.159 --> 00:01:36.520
uh similar to what I have written on my

00:01:32.759 --> 00:01:38.040
slide here uh but natur in addition to

00:01:36.520 --> 00:01:41.280
natural language understanding there's

00:01:38.040 --> 00:01:46.000
one major other segment of NLP uh does

00:01:41.280 --> 00:01:46.000
anyone uh have an idea what that might

00:01:48.719 --> 00:01:53.079
be we often have a dichotomy between two

00:01:51.399 --> 00:01:55.240
major segments natural language

00:01:53.079 --> 00:01:57.520
understanding and natural language

00:01:55.240 --> 00:01:59.439
generation yeah exactly so I I would say

00:01:57.520 --> 00:02:03.119
that's almost perfect if you had said

00:01:59.439 --> 00:02:06.640
understand and generate so very good um

00:02:03.119 --> 00:02:08.560
so I I say natural technology to handle

00:02:06.640 --> 00:02:11.400
human language usually text using

00:02:08.560 --> 00:02:13.200
computers uh to Aid human machine

00:02:11.400 --> 00:02:15.480
communication and this can include

00:02:13.200 --> 00:02:17.879
things like question answering dialogue

00:02:15.480 --> 00:02:20.840
or generation of code that can be

00:02:17.879 --> 00:02:23.239
executed with uh

00:02:20.840 --> 00:02:25.080
computers it can also Aid human human

00:02:23.239 --> 00:02:27.440
communication and this can include

00:02:25.080 --> 00:02:30.440
things like machine translation or spell

00:02:27.440 --> 00:02:32.640
checking or assisted writing

00:02:30.440 --> 00:02:34.560
and then a final uh segment that people

00:02:32.640 --> 00:02:37.400
might think about a little bit less is

00:02:34.560 --> 00:02:39.400
analyzing and understanding a language

00:02:37.400 --> 00:02:42.400
and this includes things like syntactic

00:02:39.400 --> 00:02:44.959
analysis text classification entity

00:02:42.400 --> 00:02:47.400
recognition and linking and these can be

00:02:44.959 --> 00:02:49.159
used for uh various reasons not

00:02:47.400 --> 00:02:51.000
necessarily for direct human machine

00:02:49.159 --> 00:02:52.720
communication but also for like

00:02:51.000 --> 00:02:54.400
aggregating information across large

00:02:52.720 --> 00:02:55.760
things for scientific studies and other

00:02:54.400 --> 00:02:57.519
things like that I'll give a few

00:02:55.760 --> 00:03:00.920
examples of

00:02:57.519 --> 00:03:04.040
this um we now use an many times a day

00:03:00.920 --> 00:03:06.480
sometimes without even knowing it so uh

00:03:04.040 --> 00:03:09.400
whenever you're typing a doc in Google

00:03:06.480 --> 00:03:11.599
Docs there's you know spell checking and

00:03:09.400 --> 00:03:13.959
grammar checking going on behind it's

00:03:11.599 --> 00:03:15.920
gotten frighten frighteningly good

00:03:13.959 --> 00:03:18.280
recently that where it checks like most

00:03:15.920 --> 00:03:20.720
of my mistakes and rarely Flags things

00:03:18.280 --> 00:03:22.799
that are not mistakes so obviously they

00:03:20.720 --> 00:03:25.080
have powerful models running behind that

00:03:22.799 --> 00:03:25.080
uh

00:03:25.640 --> 00:03:33.080
so and it can do things like answer

00:03:28.720 --> 00:03:34.599
questions uh so I asked chat GPT who is

00:03:33.080 --> 00:03:37.000
the current president of Carnegie melan

00:03:34.599 --> 00:03:38.920
University and chat GPT said I did a

00:03:37.000 --> 00:03:40.920
quick search for more information here

00:03:38.920 --> 00:03:43.439
is what I found uh the current president

00:03:40.920 --> 00:03:47.120
of car Mel University is faram Janan he

00:03:43.439 --> 00:03:50.040
has been serving since July 1 etc etc so

00:03:47.120 --> 00:03:50.040
as far as I can tell that's

00:03:50.400 --> 00:03:56.319
correct um at the same time I asked how

00:03:53.799 --> 00:04:00.280
many layers are included in the GP 3.5

00:03:56.319 --> 00:04:02.360
turbo architecture and it said to me

00:04:00.280 --> 00:04:05.400
GPT 3.5 turbo which is an optimized

00:04:02.360 --> 00:04:07.239
version of GPT 3.5 for faster responses

00:04:05.400 --> 00:04:08.959
doesn't have a specific layer art

00:04:07.239 --> 00:04:11.720
structure like the traditional gpt3

00:04:08.959 --> 00:04:13.560
models um and I don't know if this is

00:04:11.720 --> 00:04:16.600
true or not but I'm pretty sure it's not

00:04:13.560 --> 00:04:18.840
true I'm pretty sure that you know GPT

00:04:16.600 --> 00:04:20.560
is a model that's much like other models

00:04:18.840 --> 00:04:21.560
uh so it basically just made up the spec

00:04:20.560 --> 00:04:22.880
because it didn't have any information

00:04:21.560 --> 00:04:26.000
on the Internet or couldn't talk about

00:04:22.880 --> 00:04:26.000
it so

00:04:26.120 --> 00:04:33.479
um another thing is uh NLP can translate

00:04:29.639 --> 00:04:37.759
text pretty well so I ran um Google

00:04:33.479 --> 00:04:39.560
translate uh on Japanese uh this example

00:04:37.759 --> 00:04:41.639
is a little bit old it's from uh you

00:04:39.560 --> 00:04:44.639
know a few years ago about Co but I I

00:04:41.639 --> 00:04:46.240
retranslated it a few days ago and it

00:04:44.639 --> 00:04:47.680
comes up pretty good uh you can

00:04:46.240 --> 00:04:49.639
basically understand what's going on

00:04:47.680 --> 00:04:53.520
here it's not perfect but you can

00:04:49.639 --> 00:04:56.400
understand the uh the general uh

00:04:53.520 --> 00:04:58.560
gist at the same time uh if I put in a

00:04:56.400 --> 00:05:02.280
relatively low resource language this is

00:04:58.560 --> 00:05:05.759
Kurdish um it has a number of problems

00:05:02.280 --> 00:05:08.199
when you try to understand it and just

00:05:05.759 --> 00:05:12.400
to give an example this is talking about

00:05:08.199 --> 00:05:14.320
uh some uh paleontology Discovery it

00:05:12.400 --> 00:05:15.800
called this person a fossil scientist

00:05:14.320 --> 00:05:17.440
instead of the kind of obvious English

00:05:15.800 --> 00:05:20.120
term

00:05:17.440 --> 00:05:23.520
paleontologist um and it's talking about

00:05:20.120 --> 00:05:25.240
three different uh T-Rex species uh how

00:05:23.520 --> 00:05:27.039
T-Rex should actually be split into

00:05:25.240 --> 00:05:29.639
three species where T-Rex says king of

00:05:27.039 --> 00:05:31.560
ferocious lizards emperator says emperor

00:05:29.639 --> 00:05:33.720
of Savaged lizards and then T Regina

00:05:31.560 --> 00:05:35.120
means clean of ferocious snail I'm

00:05:33.720 --> 00:05:37.240
pretty sure that's not snail I'm pretty

00:05:35.120 --> 00:05:41.080
sure that's lizard so uh you can see

00:05:37.240 --> 00:05:41.080
that this is not uh this is not perfect

00:05:41.280 --> 00:05:46.680
either some people might be thinking why

00:05:43.960 --> 00:05:48.400
Google translate and why not GPD well it

00:05:46.680 --> 00:05:49.960
turns out um according to one of the

00:05:48.400 --> 00:05:51.759
recent studies we've done GPD is even

00:05:49.960 --> 00:05:55.479
worse at these slow resource languages

00:05:51.759 --> 00:05:58.120
so I use the best thing that's out

00:05:55.479 --> 00:06:00.440
there um another thing is language

00:05:58.120 --> 00:06:02.039
analysis can Aid scientific ific inquiry

00:06:00.440 --> 00:06:03.600
so this is an example that I've been

00:06:02.039 --> 00:06:06.120
using for a long time it's actually from

00:06:03.600 --> 00:06:09.160
Martin sap another faculty member here

00:06:06.120 --> 00:06:12.440
uh but I have been using it since uh

00:06:09.160 --> 00:06:14.160
like before he joined and it uh this is

00:06:12.440 --> 00:06:16.039
an example from computational social

00:06:14.160 --> 00:06:18.599
science uh answering questions about

00:06:16.039 --> 00:06:20.240
Society given observational data and

00:06:18.599 --> 00:06:22.280
their question was do movie scripts

00:06:20.240 --> 00:06:24.599
portray female or male characters with

00:06:22.280 --> 00:06:27.520
more power or agency in movie script

00:06:24.599 --> 00:06:30.120
films so it's asking kind of a so

00:06:27.520 --> 00:06:32.160
societal question by using NLP

00:06:30.120 --> 00:06:35.360
technology and the way they did it is

00:06:32.160 --> 00:06:36.880
they basically analyzed text trying to

00:06:35.360 --> 00:06:43.080
find

00:06:36.880 --> 00:06:45.280
uh the uh agents and patients in a a

00:06:43.080 --> 00:06:46.479
particular text which are the the things

00:06:45.280 --> 00:06:49.280
that are doing things and the things

00:06:46.479 --> 00:06:52.639
that things are being done to and you

00:06:49.280 --> 00:06:54.440
can see that essentially male characters

00:06:52.639 --> 00:06:56.560
in these movie scripts were given more

00:06:54.440 --> 00:06:58.080
power in agency and female characters

00:06:56.560 --> 00:06:59.960
were given less power in agency and they

00:06:58.080 --> 00:07:02.680
were able to do this because they had

00:06:59.960 --> 00:07:04.840
NLP technology that analyzed and

00:07:02.680 --> 00:07:08.960
extracted useful data and made turned it

00:07:04.840 --> 00:07:11.520
into a very easy form to do kind of

00:07:08.960 --> 00:07:15.840
analysis of the variety that they want

00:07:11.520 --> 00:07:17.400
so um I think that's a major use case of

00:07:15.840 --> 00:07:19.400
NLP technology that does language

00:07:17.400 --> 00:07:20.919
analysis nowadays turn it into a form

00:07:19.400 --> 00:07:23.960
that allows you to very quickly do

00:07:20.919 --> 00:07:27.440
aggregate queries and other things like

00:07:23.960 --> 00:07:30.479
this um but at the same time uh language

00:07:27.440 --> 00:07:33.520
analysis tools fail at very basic tasks

00:07:30.479 --> 00:07:36.000
so these are

00:07:33.520 --> 00:07:38.199
some things that I ran through a named

00:07:36.000 --> 00:07:41.080
entity recognizer and these were kind of

00:07:38.199 --> 00:07:43.160
very nice named entity recognizers uh

00:07:41.080 --> 00:07:46.240
that a lot of people were using for

00:07:43.160 --> 00:07:48.039
example Stanford core NLP and Spacey and

00:07:46.240 --> 00:07:50.319
both of them I just threw in the first

00:07:48.039 --> 00:07:53.120
thing that I found on the New York Times

00:07:50.319 --> 00:07:55.199
at the time and it basically made at

00:07:53.120 --> 00:07:58.319
least one mistake in the first sentence

00:07:55.199 --> 00:08:00.840
and here it recognizes Baton Rouge as an

00:07:58.319 --> 00:08:04.720
organization and here it recognized

00:08:00.840 --> 00:08:07.000
hurricane EA as an organization so um

00:08:04.720 --> 00:08:08.879
like even uh these things that we expect

00:08:07.000 --> 00:08:10.360
should work pretty well make pretty

00:08:08.879 --> 00:08:13.360
Solly

00:08:10.360 --> 00:08:16.199
mistakes so in the class uh basically

00:08:13.360 --> 00:08:18.479
what I want to cover is uh what goes

00:08:16.199 --> 00:08:20.360
into building uh state-of-the-art NLP

00:08:18.479 --> 00:08:24.000
systems that work really well on a wide

00:08:20.360 --> 00:08:26.240
variety of tasks um where do current

00:08:24.000 --> 00:08:28.840
systems

00:08:26.240 --> 00:08:30.479
fail and how can we make appropriate

00:08:28.840 --> 00:08:35.000
improvements and Achieve whatever we

00:08:30.479 --> 00:08:37.719
want to do with nalp and this set of

00:08:35.000 --> 00:08:39.360
questions that I'm asking here is

00:08:37.719 --> 00:08:40.919
exactly the same as the set of questions

00:08:39.360 --> 00:08:43.519
that I was asking two years ago before

00:08:40.919 --> 00:08:45.480
chat GPT uh I still think they're

00:08:43.519 --> 00:08:46.920
important questions but I think the

00:08:45.480 --> 00:08:48.399
answers to these questions is very

00:08:46.920 --> 00:08:50.040
different and because of that we're

00:08:48.399 --> 00:08:52.120
updating the class materials to try to

00:08:50.040 --> 00:08:54.399
cover you know the answers to these

00:08:52.120 --> 00:08:56.000
questions and uh in kind of the era of

00:08:54.399 --> 00:08:58.200
large language models and other things

00:08:56.000 --> 00:08:59.720
like

00:08:58.200 --> 00:09:02.079
that

00:08:59.720 --> 00:09:03.360
so that's all I have for the intro maybe

00:09:02.079 --> 00:09:06.640
maybe pretty straightforward are there

00:09:03.360 --> 00:09:08.480
any questions or comments so far if not

00:09:06.640 --> 00:09:14.399
I'll I'll just go

00:09:08.480 --> 00:09:17.160
on okay great so I want to uh first go

00:09:14.399 --> 00:09:19.480
into a very high Lev overview of NLP

00:09:17.160 --> 00:09:20.839
system building and most of the stuff

00:09:19.480 --> 00:09:22.399
that I want to do today is to set the

00:09:20.839 --> 00:09:24.320
stage for what I'm going to be talking

00:09:22.399 --> 00:09:25.040
about in more detail uh over the rest of

00:09:24.320 --> 00:09:29.200
the

00:09:25.040 --> 00:09:31.720
class and we could think of NLP syst

00:09:29.200 --> 00:09:34.040
systems through this kind of General

00:09:31.720 --> 00:09:36.560
framework where we want to create a

00:09:34.040 --> 00:09:40.600
function to map an input X into an

00:09:36.560 --> 00:09:44.440
output y uh where X and or Y involve

00:09:40.600 --> 00:09:47.000
language and uh do some people have

00:09:44.440 --> 00:09:50.120
favorite NLP tasks or NLP tasks that you

00:09:47.000 --> 00:09:52.399
want to uh want to be handling in some

00:09:50.120 --> 00:09:57.000
way or maybe what what do you think are

00:09:52.399 --> 00:09:57.000
the most popular and important NLP tasks

00:09:58.120 --> 00:10:03.200
nowadays

00:10:00.800 --> 00:10:06.120
okay so translation is maybe easy what's

00:10:03.200 --> 00:10:06.120
the input and output of

00:10:11.440 --> 00:10:15.720
translation okay yeah so uh in

00:10:13.800 --> 00:10:17.959
Translation inputs text in one language

00:10:15.720 --> 00:10:21.760
output is text in another language and

00:10:17.959 --> 00:10:21.760
then what what is a good

00:10:27.680 --> 00:10:32.160
translation yeah corre or or the same is

00:10:30.320 --> 00:10:35.839
the input basically yes um it also

00:10:32.160 --> 00:10:37.760
should be fluent but I agree any other

00:10:35.839 --> 00:10:39.839
things generation the reason why I said

00:10:37.760 --> 00:10:41.519
it's tough is it's pretty broad um and

00:10:39.839 --> 00:10:43.360
it's not like we could be doing

00:10:41.519 --> 00:10:46.360
generation with lots of different inputs

00:10:43.360 --> 00:10:51.440
but um yeah any any other things maybe a

00:10:46.360 --> 00:10:51.440
little bit different yeah like

00:10:51.480 --> 00:10:55.959
scenario a scenario and a multiple

00:10:54.000 --> 00:10:58.200
choice question about the scenario and

00:10:55.959 --> 00:10:59.680
so what would the scenario in the

00:10:58.200 --> 00:11:01.760
multiple choice question are probably

00:10:59.680 --> 00:11:04.040
the input and then the output

00:11:01.760 --> 00:11:06.480
is an answer to the multiple choice

00:11:04.040 --> 00:11:07.920
question um and then there it's kind of

00:11:06.480 --> 00:11:12.279
obvious like what is good it's the

00:11:07.920 --> 00:11:14.880
correct answer sure um interestingly I

00:11:12.279 --> 00:11:17.440
think a lot of llm evaluation is done on

00:11:14.880 --> 00:11:21.160
these multiple choice questions but I'm

00:11:17.440 --> 00:11:22.320
yet to encounter an actual application

00:11:21.160 --> 00:11:24.880
that cares about multiple choice

00:11:22.320 --> 00:11:26.880
question answering so uh there's kind of

00:11:24.880 --> 00:11:30.959
a funny disconnect there but uh yeah I

00:11:26.880 --> 00:11:33.519
saw hand that think about V search comp

00:11:30.959 --> 00:11:36.360
yeah Vector search uh that's very good

00:11:33.519 --> 00:11:36.360
so the input

00:11:37.120 --> 00:11:45.000
is can con it into or understanding and

00:11:42.560 --> 00:11:45.000
it to

00:11:47.360 --> 00:11:53.760
another okay yeah so I'd say the input

00:11:49.880 --> 00:11:56.160
there is a query and a document base um

00:11:53.760 --> 00:11:57.959
and then the output is maybe an index

00:11:56.160 --> 00:11:59.800
into the document or or something else

00:11:57.959 --> 00:12:01.279
like that sure um and then something

00:11:59.800 --> 00:12:05.040
that's good here here's a good question

00:12:01.279 --> 00:12:05.040
what what's a good result from

00:12:06.560 --> 00:12:10.200
that what's a good

00:12:10.839 --> 00:12:19.279
output be sort of simar the major

00:12:15.560 --> 00:12:21.680
problem there I see is how you def SAR

00:12:19.279 --> 00:12:26.199
and how you

00:12:21.680 --> 00:12:29.760
a always like you understand

00:12:26.199 --> 00:12:33.000
whether is actually

00:12:29.760 --> 00:12:35.079
yeah exactly so that um just to repeat

00:12:33.000 --> 00:12:36.880
it's like uh we need to have a

00:12:35.079 --> 00:12:38.399
similarity a good similarity metric we

00:12:36.880 --> 00:12:40.120
need to have a good threshold where we

00:12:38.399 --> 00:12:41.760
get like the ones we want and we don't

00:12:40.120 --> 00:12:43.240
get the ones we don't want we're going

00:12:41.760 --> 00:12:44.959
to talk more about that in the retrieval

00:12:43.240 --> 00:12:48.440
lecture exactly how we evaluate and

00:12:44.959 --> 00:12:49.920
stuff but um yeah good so this is a good

00:12:48.440 --> 00:12:53.279
uh here are some good examples I have

00:12:49.920 --> 00:12:55.519
some examples of my own um the first one

00:12:53.279 --> 00:12:58.360
is uh kind of the very generic one maybe

00:12:55.519 --> 00:13:00.800
kind of like generation here but text in

00:12:58.360 --> 00:13:02.959
continuing text uh so this is language

00:13:00.800 --> 00:13:04.160
modeling so you have a text and then you

00:13:02.959 --> 00:13:05.440
have the continuation you want to

00:13:04.160 --> 00:13:07.680
predict the

00:13:05.440 --> 00:13:10.480
continuation um text and text in another

00:13:07.680 --> 00:13:13.040
language is translation uh text in a

00:13:10.480 --> 00:13:15.800
label could be text classification uh

00:13:13.040 --> 00:13:17.760
text in linguistic structure or uh some

00:13:15.800 --> 00:13:21.360
s kind of entities or something like

00:13:17.760 --> 00:13:22.680
that could be uh language analysis or um

00:13:21.360 --> 00:13:24.839
information

00:13:22.680 --> 00:13:29.440
extraction uh we could also have image

00:13:24.839 --> 00:13:31.320
and text uh which is image captioning um

00:13:29.440 --> 00:13:33.560
or speech and text which is speech

00:13:31.320 --> 00:13:35.240
recognition and I take the very broad

00:13:33.560 --> 00:13:38.000
view of natural language processing

00:13:35.240 --> 00:13:39.519
which is if it's any variety of language

00:13:38.000 --> 00:13:41.519
uh if you're handling language in some

00:13:39.519 --> 00:13:42.800
way it's natural language processing it

00:13:41.519 --> 00:13:45.880
doesn't necessarily have to be text

00:13:42.800 --> 00:13:47.480
input text output um so that's relevant

00:13:45.880 --> 00:13:50.199
for the projects that you're thinking

00:13:47.480 --> 00:13:52.160
about too at the end of this course so

00:13:50.199 --> 00:13:55.519
the the most common FAQ for this course

00:13:52.160 --> 00:13:57.839
is does my project count and if you're

00:13:55.519 --> 00:13:59.360
uncertain you should ask but usually

00:13:57.839 --> 00:14:01.040
like if it has some sort of language

00:13:59.360 --> 00:14:05.079
involved then I'll usually say yes it

00:14:01.040 --> 00:14:07.920
does kind so um if it's like uh code to

00:14:05.079 --> 00:14:09.680
code there that's not code is not

00:14:07.920 --> 00:14:11.480
natural language it is language but it's

00:14:09.680 --> 00:14:13.000
not natural language so that might be

00:14:11.480 --> 00:14:15.320
borderline we might have to discuss

00:14:13.000 --> 00:14:15.320
about

00:14:15.759 --> 00:14:21.800
that cool um so next I'd like to talk

00:14:18.880 --> 00:14:25.240
about methods for creating NLP systems

00:14:21.800 --> 00:14:27.839
um and there's a lot of different ways

00:14:25.240 --> 00:14:29.720
to create MLP systems all of these are

00:14:27.839 --> 00:14:32.880
alive and well in

00:14:29.720 --> 00:14:35.759
2024 uh the first one is Rule uh

00:14:32.880 --> 00:14:37.959
rule-based system creation and so the

00:14:35.759 --> 00:14:40.399
way this works is like let's say you

00:14:37.959 --> 00:14:42.480
want to build a text classifier you just

00:14:40.399 --> 00:14:46.560
write the simple python function that

00:14:42.480 --> 00:14:48.639
classifies things into uh sports or

00:14:46.560 --> 00:14:50.240
other and the way it classifies it into

00:14:48.639 --> 00:14:52.959
sports or other is it checks whether

00:14:50.240 --> 00:14:55.160
baseball soccer football and Tennis are

00:14:52.959 --> 00:14:59.399
included in the document and classifies

00:14:55.160 --> 00:15:01.959
it into uh Sports if so uh other if not

00:14:59.399 --> 00:15:05.279
so has anyone written something like

00:15:01.959 --> 00:15:09.720
this maybe not a text classifier but um

00:15:05.279 --> 00:15:11.880
you know to identify entities or uh

00:15:09.720 --> 00:15:14.279
split words

00:15:11.880 --> 00:15:16.680
or something like

00:15:14.279 --> 00:15:18.399
that has anybody not ever written

00:15:16.680 --> 00:15:22.800
anything like

00:15:18.399 --> 00:15:24.639
this yeah that's what I thought so um

00:15:22.800 --> 00:15:26.079
rule-based systems are very convenient

00:15:24.639 --> 00:15:28.920
when you don't really care about how

00:15:26.079 --> 00:15:30.759
good your system is um or you're doing

00:15:28.920 --> 00:15:32.360
that's really really simple and like

00:15:30.759 --> 00:15:35.600
it'll be perfect even if you do the very

00:15:32.360 --> 00:15:37.079
simple thing and so I I think it's worth

00:15:35.600 --> 00:15:39.959
talking a little bit about them and I'll

00:15:37.079 --> 00:15:43.319
talk a little bit about that uh this

00:15:39.959 --> 00:15:45.680
time the second thing which like very

00:15:43.319 --> 00:15:47.680
rapidly over the course of maybe three

00:15:45.680 --> 00:15:50.279
years or so has become actually maybe

00:15:47.680 --> 00:15:52.720
the dominant Paradigm in NLP is

00:15:50.279 --> 00:15:56.360
prompting uh in prompting a language

00:15:52.720 --> 00:15:58.560
model and the way this works is uh you

00:15:56.360 --> 00:16:00.720
ask a language model if the following

00:15:58.560 --> 00:16:03.079
sent is about sports reply Sports

00:16:00.720 --> 00:16:06.120
otherwise reply other and you feed it to

00:16:03.079 --> 00:16:08.480
your favorite LM uh usually that's GPT

00:16:06.120 --> 00:16:11.399
something or other uh sometimes it's an

00:16:08.480 --> 00:16:14.440
open source model of some variety and

00:16:11.399 --> 00:16:17.759
then uh it will give you the

00:16:14.440 --> 00:16:20.639
answer and then finally uh fine-tuning

00:16:17.759 --> 00:16:22.240
uh so you take some paired data and you

00:16:20.639 --> 00:16:23.600
do machine learning from paired data

00:16:22.240 --> 00:16:25.680
where you have something like I love to

00:16:23.600 --> 00:16:27.440
play baseball uh the stock price is

00:16:25.680 --> 00:16:29.519
going up he got a hatrick yesterday he

00:16:27.440 --> 00:16:32.759
is wearing tennis shoes and you assign

00:16:29.519 --> 00:16:35.319
all these uh labels to them training a

00:16:32.759 --> 00:16:38.160
model and you can even start out with a

00:16:35.319 --> 00:16:41.480
prompting based model and fine-tune a a

00:16:38.160 --> 00:16:41.480
language model

00:16:42.920 --> 00:16:49.399
also so one major consideration when

00:16:47.519 --> 00:16:52.000
you're Building Systems like this is the

00:16:49.399 --> 00:16:56.440
data requirements for building such a

00:16:52.000 --> 00:16:59.319
system and for rules or prompting where

00:16:56.440 --> 00:17:02.240
it's just based on intuition really no

00:16:59.319 --> 00:17:04.640
data is needed whatsoever it you don't

00:17:02.240 --> 00:17:08.240
need a single example and you can start

00:17:04.640 --> 00:17:11.000
writing rules or like just just to give

00:17:08.240 --> 00:17:12.640
an example the rules and prompts I wrote

00:17:11.000 --> 00:17:14.679
here I didn't look at any examples and I

00:17:12.640 --> 00:17:17.240
just wrote them uh so this is something

00:17:14.679 --> 00:17:20.000
that you could start out

00:17:17.240 --> 00:17:21.559
with uh the problem is you also have no

00:17:20.000 --> 00:17:24.720
idea how well it works if you don't have

00:17:21.559 --> 00:17:26.760
any data whatsoever right so um you'll

00:17:24.720 --> 00:17:30.400
you might be in trouble if you think

00:17:26.760 --> 00:17:30.400
something should be working

00:17:30.919 --> 00:17:34.440
so normally the next thing that people

00:17:32.919 --> 00:17:36.880
move to nowadays when they're building

00:17:34.440 --> 00:17:39.559
practical systems is rules are prompting

00:17:36.880 --> 00:17:41.240
based on spot checks so that basically

00:17:39.559 --> 00:17:42.919
means that you start out with a

00:17:41.240 --> 00:17:45.840
rule-based system or a prompting based

00:17:42.919 --> 00:17:47.240
system and then you go in and you run it

00:17:45.840 --> 00:17:48.720
on some data that you're interested in

00:17:47.240 --> 00:17:50.799
you just kind of qualitatively look at

00:17:48.720 --> 00:17:52.160
the data and say oh it's messing up here

00:17:50.799 --> 00:17:53.440
then you go in and fix your prompt a

00:17:52.160 --> 00:17:54.919
little bit or you go in and fix your

00:17:53.440 --> 00:17:57.320
rules a little bit or something like

00:17:54.919 --> 00:18:00.400
that so uh this is kind of the second

00:17:57.320 --> 00:18:00.400
level of difficulty

00:18:01.400 --> 00:18:04.640
so the third level of difficulty would

00:18:03.159 --> 00:18:07.400
be something like rules are prompting

00:18:04.640 --> 00:18:09.039
with rigorous evaluation and so here you

00:18:07.400 --> 00:18:12.840
would create a development set with

00:18:09.039 --> 00:18:14.840
inputs and outputs uh so you uh create

00:18:12.840 --> 00:18:17.039
maybe 200 to 2,000

00:18:14.840 --> 00:18:20.080
examples um

00:18:17.039 --> 00:18:21.720
and then evaluate your actual accuracy

00:18:20.080 --> 00:18:23.880
so you need an evaluation metric you

00:18:21.720 --> 00:18:26.120
need other things like this this is the

00:18:23.880 --> 00:18:28.400
next level of difficulty but if you're

00:18:26.120 --> 00:18:30.240
going to be a serious you know NLP

00:18:28.400 --> 00:18:33.000
engineer or something like this you

00:18:30.240 --> 00:18:34.720
definitely will be doing this a lot I

00:18:33.000 --> 00:18:37.760
feel and

00:18:34.720 --> 00:18:40.360
then so that here now you start needing

00:18:37.760 --> 00:18:41.960
a depth set and a test set and then

00:18:40.360 --> 00:18:46.280
finally fine-tuning you need an

00:18:41.960 --> 00:18:48.480
additional training set um and uh this

00:18:46.280 --> 00:18:52.240
will generally be a lot bigger than 200

00:18:48.480 --> 00:18:56.080
to 2,000 examples and generally the rule

00:18:52.240 --> 00:18:56.080
is that every time you

00:18:57.320 --> 00:19:01.080
double

00:18:59.520 --> 00:19:02.400
every time you double your training set

00:19:01.080 --> 00:19:07.480
size you get about a constant

00:19:02.400 --> 00:19:07.480
Improvement so if you start

00:19:07.799 --> 00:19:15.080
out if you start out down here with

00:19:12.240 --> 00:19:17.039
um zero shot accuracy with a language

00:19:15.080 --> 00:19:21.559
model you you create a small printing

00:19:17.039 --> 00:19:21.559
set and you get you know a pretty big

00:19:22.000 --> 00:19:29.120
increase and then every time you double

00:19:26.320 --> 00:19:30.799
it it increases by constant fact it's

00:19:29.120 --> 00:19:32.480
kind of like just in general in machine

00:19:30.799 --> 00:19:37.360
learning this is a trend that we tend to

00:19:32.480 --> 00:19:40.679
see so um So based on this

00:19:37.360 --> 00:19:41.880
uh there's kind of like you get a big

00:19:40.679 --> 00:19:44.200
gain from having a little bit of

00:19:41.880 --> 00:19:45.760
training data but the gains very quickly

00:19:44.200 --> 00:19:48.919
drop off and you start spending a lot of

00:19:45.760 --> 00:19:48.919
time annotating

00:19:51.000 --> 00:19:55.880
an so um yeah this is the the general

00:19:54.760 --> 00:19:58.280
overview of the different types of

00:19:55.880 --> 00:20:00.000
system building uh any any question

00:19:58.280 --> 00:20:01.559
questions about this or comments or

00:20:00.000 --> 00:20:04.000
things like

00:20:01.559 --> 00:20:05.840
this I think one thing that's changed

00:20:04.000 --> 00:20:08.159
really drastically from the last time I

00:20:05.840 --> 00:20:09.600
taught this class is the fact that

00:20:08.159 --> 00:20:11.000
number one and number two are the things

00:20:09.600 --> 00:20:13.799
that people are actually doing in

00:20:11.000 --> 00:20:15.360
practice uh which was you know people

00:20:13.799 --> 00:20:16.679
who actually care about systems are

00:20:15.360 --> 00:20:18.880
doing number one and number two is the

00:20:16.679 --> 00:20:20.440
main thing it used to be that if you

00:20:18.880 --> 00:20:22.679
were actually serious about building a

00:20:20.440 --> 00:20:24.320
system uh you really needed to do the

00:20:22.679 --> 00:20:27.080
funing and now it's kind of like more

00:20:24.320 --> 00:20:27.080
optional

00:20:27.159 --> 00:20:30.159
so

00:20:44.039 --> 00:20:50.960
yeah

00:20:46.320 --> 00:20:53.960
so it's it's definitely an empirical

00:20:50.960 --> 00:20:53.960
observation

00:20:54.720 --> 00:21:01.080
um in terms of the theoretical

00:20:57.640 --> 00:21:03.120
background I am not I can't immediately

00:21:01.080 --> 00:21:05.840
point to a

00:21:03.120 --> 00:21:10.039
particular paper that does that but I

00:21:05.840 --> 00:21:12.720
think if you think about

00:21:10.039 --> 00:21:14.720
the I I think I have seen that they do

00:21:12.720 --> 00:21:17.039
exist in the past but I I can't think of

00:21:14.720 --> 00:21:19.000
it right now I can try to uh try to come

00:21:17.039 --> 00:21:23.720
up with an example of

00:21:19.000 --> 00:21:23.720
that so yeah I I should take

00:21:26.799 --> 00:21:31.960
notes or someone wants to share one on

00:21:29.360 --> 00:21:33.360
Piaza uh if you have any ideas and want

00:21:31.960 --> 00:21:34.520
to share on Patza I'm sure that would be

00:21:33.360 --> 00:21:35.640
great it'd be great to have a discussion

00:21:34.520 --> 00:21:39.320
on

00:21:35.640 --> 00:21:44.960
Patza um Pi

00:21:39.320 --> 00:21:46.880
one cool okay so next I want to try to

00:21:44.960 --> 00:21:48.200
make a rule-based system and I'm going

00:21:46.880 --> 00:21:49.360
to make a rule-based system for

00:21:48.200 --> 00:21:51.799
sentiment

00:21:49.360 --> 00:21:53.480
analysis uh and this is a bad idea I

00:21:51.799 --> 00:21:55.400
would not encourage you to ever do this

00:21:53.480 --> 00:21:57.440
in real life but I want to do it here to

00:21:55.400 --> 00:21:59.640
show you why it's a bad idea and like

00:21:57.440 --> 00:22:01.200
what are some of the hard problems that

00:21:59.640 --> 00:22:03.960
you encounter when trying to create a

00:22:01.200 --> 00:22:06.600
system based on rules

00:22:03.960 --> 00:22:08.080
and then we'll move into building a

00:22:06.600 --> 00:22:12.360
machine learning base system after we

00:22:08.080 --> 00:22:15.400
finish this so if we look at the example

00:22:12.360 --> 00:22:18.559
test this is review sentiment analysis

00:22:15.400 --> 00:22:21.799
it's one of the most valuable uh tasks

00:22:18.559 --> 00:22:24.039
uh that people do in NLP nowadays

00:22:21.799 --> 00:22:26.400
because it allows people to know how

00:22:24.039 --> 00:22:29.200
customers are thinking about products uh

00:22:26.400 --> 00:22:30.799
improve their you know their product

00:22:29.200 --> 00:22:32.919
development and other things like that

00:22:30.799 --> 00:22:34.799
may monitor people's you know

00:22:32.919 --> 00:22:36.760
satisfaction with their social media

00:22:34.799 --> 00:22:39.200
service other things like this so

00:22:36.760 --> 00:22:42.720
basically the way it works is um you

00:22:39.200 --> 00:22:44.400
have uh outputs or you have sentences

00:22:42.720 --> 00:22:46.720
inputs like I hate this movie I love

00:22:44.400 --> 00:22:48.520
this movie I saw this movie and this

00:22:46.720 --> 00:22:50.600
gets mapped into positive neutral or

00:22:48.520 --> 00:22:53.120
negative so I hate this movie would be

00:22:50.600 --> 00:22:55.480
negative I love this movie positive and

00:22:53.120 --> 00:22:59.039
I saw this movie is

00:22:55.480 --> 00:23:01.200
neutral so um

00:22:59.039 --> 00:23:05.200
that that's the task input tax output

00:23:01.200 --> 00:23:08.880
labels uh Kary uh sentence

00:23:05.200 --> 00:23:11.679
label and in order to do this uh we

00:23:08.880 --> 00:23:13.120
would like to build a model um and we're

00:23:11.679 --> 00:23:16.159
going to build the model in a rule based

00:23:13.120 --> 00:23:19.000
way but it we'll still call it a model

00:23:16.159 --> 00:23:21.600
and the way it works is we do feature

00:23:19.000 --> 00:23:23.159
extraction um so we extract the Salient

00:23:21.600 --> 00:23:25.279
features for making the decision about

00:23:23.159 --> 00:23:27.320
what to Output next we do score

00:23:25.279 --> 00:23:29.880
calculation calculate a score for one or

00:23:27.320 --> 00:23:32.320
more possib ities and we have a decision

00:23:29.880 --> 00:23:33.520
function so we choose one of those

00:23:32.320 --> 00:23:37.679
several

00:23:33.520 --> 00:23:40.120
possibilities and so for feature

00:23:37.679 --> 00:23:42.200
extraction uh formally what this looks

00:23:40.120 --> 00:23:44.240
like is we have some function and it

00:23:42.200 --> 00:23:48.039
extracts a feature

00:23:44.240 --> 00:23:51.159
Vector for score calculation um we

00:23:48.039 --> 00:23:54.240
calculate the scores based on either a

00:23:51.159 --> 00:23:56.279
binary classification uh where we have a

00:23:54.240 --> 00:23:58.279
a weight vector and we take the dot

00:23:56.279 --> 00:24:00.120
product with our feature vector or we

00:23:58.279 --> 00:24:02.480
have multi class classification where we

00:24:00.120 --> 00:24:04.520
have a weight Matrix and we take the

00:24:02.480 --> 00:24:08.640
product with uh the vector and that

00:24:04.520 --> 00:24:08.640
gives us you know squares over multiple

00:24:08.919 --> 00:24:14.840
classes and then we have a decision uh

00:24:11.600 --> 00:24:17.520
rule so this decision rule tells us what

00:24:14.840 --> 00:24:20.080
the output is going to be um does anyone

00:24:17.520 --> 00:24:22.200
know what a typical decision rule is

00:24:20.080 --> 00:24:24.520
maybe maybe so obvious that you don't

00:24:22.200 --> 00:24:28.760
think about it often

00:24:24.520 --> 00:24:31.000
but uh a threshold um so like for would

00:24:28.760 --> 00:24:34.440
that be for binary a single binary

00:24:31.000 --> 00:24:37.000
scaler score or a multiple

00:24:34.440 --> 00:24:38.520
class binary yeah so and then you would

00:24:37.000 --> 00:24:39.960
pick a threshold and if it's over the

00:24:38.520 --> 00:24:42.919
threshold

00:24:39.960 --> 00:24:45.760
you say yes and if it's under the

00:24:42.919 --> 00:24:50.279
threshold you say no um another option

00:24:45.760 --> 00:24:51.679
would be um you have a threshold and you

00:24:50.279 --> 00:24:56.080
say

00:24:51.679 --> 00:24:56.080
yes no

00:24:56.200 --> 00:25:00.559
obain so you know you don't give an

00:24:58.360 --> 00:25:02.520
answer and depending on how you're

00:25:00.559 --> 00:25:03.720
evaluated what what is a good classifier

00:25:02.520 --> 00:25:07.799
you might want to abstain some of the

00:25:03.720 --> 00:25:10.960
time also um for multiclass what what's

00:25:07.799 --> 00:25:10.960
a standard decision role for

00:25:11.120 --> 00:25:16.720
multiclass argmax yeah exactly so um

00:25:14.279 --> 00:25:19.520
basically you you find the index that

00:25:16.720 --> 00:25:22.000
has the highest score in you output

00:25:19.520 --> 00:25:24.480
it we're going to be talking about other

00:25:22.000 --> 00:25:26.559
decision rules also um like

00:25:24.480 --> 00:25:29.480
self-consistency and minimum based risk

00:25:26.559 --> 00:25:30.760
later uh for text generation so you can

00:25:29.480 --> 00:25:33.000
just keep that in mind and then we'll

00:25:30.760 --> 00:25:36.279
forget about it for like several

00:25:33.000 --> 00:25:39.559
classes um so for sentiment

00:25:36.279 --> 00:25:42.159
class um I have a Cod

00:25:39.559 --> 00:25:45.159
walk

00:25:42.159 --> 00:25:45.159
here

00:25:46.240 --> 00:25:54.320
and this is pretty simple um but if

00:25:50.320 --> 00:25:58.559
you're bored uh of the class and would

00:25:54.320 --> 00:26:01.000
like to um try out yourself you can

00:25:58.559 --> 00:26:04.480
Challenge and try to get a better score

00:26:01.000 --> 00:26:06.120
than I do um over the next few minutes

00:26:04.480 --> 00:26:06.880
but we have this rule based classifier

00:26:06.120 --> 00:26:10.240
in

00:26:06.880 --> 00:26:12.640
here and I will open it up in my vs

00:26:10.240 --> 00:26:15.360
code

00:26:12.640 --> 00:26:18.360
to try to create a rule-based classifier

00:26:15.360 --> 00:26:18.360
and basically the way this

00:26:22.799 --> 00:26:29.960
works is

00:26:25.159 --> 00:26:29.960
that we have a feature

00:26:31.720 --> 00:26:37.720
extraction we have feature extraction we

00:26:34.120 --> 00:26:40.679
have scoring and we have um a decision

00:26:37.720 --> 00:26:43.480
rle so here for our feature extraction I

00:26:40.679 --> 00:26:44.720
have created a list of good words and a

00:26:43.480 --> 00:26:46.720
list of bad

00:26:44.720 --> 00:26:48.960
words

00:26:46.720 --> 00:26:51.320
and what we do is we just count the

00:26:48.960 --> 00:26:53.000
number of good words that appeared and

00:26:51.320 --> 00:26:55.320
count the number of bad words that

00:26:53.000 --> 00:26:57.880
appeared then we also have a bias

00:26:55.320 --> 00:27:01.159
feature so the bias feature is a feature

00:26:57.880 --> 00:27:03.679
that's always one and so what that

00:27:01.159 --> 00:27:06.799
results in is we have a dimension three

00:27:03.679 --> 00:27:08.880
feature Vector um where this is like the

00:27:06.799 --> 00:27:11.320
number of good words this is the number

00:27:08.880 --> 00:27:15.320
of bad words and then you have the

00:27:11.320 --> 00:27:17.760
bias and then I also Define the feature

00:27:15.320 --> 00:27:20.039
weights that so for every good word we

00:27:17.760 --> 00:27:22.200
add one to our score for every bad word

00:27:20.039 --> 00:27:25.559
we add uh we subtract one from our score

00:27:22.200 --> 00:27:29.399
and for the BIOS we absor and so we then

00:27:25.559 --> 00:27:30.480
take the dot product between

00:27:29.399 --> 00:27:34.360
these

00:27:30.480 --> 00:27:36.919
two and we get minus

00:27:34.360 --> 00:27:37.640
0.5 and that gives us uh that gives us

00:27:36.919 --> 00:27:41.000
the

00:27:37.640 --> 00:27:46.000
squore so let's run

00:27:41.000 --> 00:27:50.320
that um and I read in some

00:27:46.000 --> 00:27:52.600
data and what this data looks like is

00:27:50.320 --> 00:27:55.000
basically we have a

00:27:52.600 --> 00:27:57.559
review um which says the rock is

00:27:55.000 --> 00:27:59.480
destined to be the 21st Century's new

00:27:57.559 --> 00:28:01.240
Conan and that he's going to make a

00:27:59.480 --> 00:28:03.600
splash even greater than Arnold

00:28:01.240 --> 00:28:07.000
Schwarzenegger jeanclaude vanam or

00:28:03.600 --> 00:28:09.519
Steven Seagal um so this seems pretty

00:28:07.000 --> 00:28:10.840
positive right I like that's a pretty

00:28:09.519 --> 00:28:13.200
high order to be better than Arnold

00:28:10.840 --> 00:28:16.080
Schwarzenegger or John Claude vanam uh

00:28:13.200 --> 00:28:19.519
if you're familiar with action movies um

00:28:16.080 --> 00:28:22.840
and so of course this gets a positive

00:28:19.519 --> 00:28:24.120
label and so uh we have run classifier

00:28:22.840 --> 00:28:25.240
actually maybe I should call this

00:28:24.120 --> 00:28:27.600
decision rule because this is

00:28:25.240 --> 00:28:29.120
essentially our decision Rule and here

00:28:27.600 --> 00:28:32.600
basically do the thing that I mentioned

00:28:29.120 --> 00:28:35.440
here the yes no obstain or in this case

00:28:32.600 --> 00:28:38.360
positive negative neutral so if the

00:28:35.440 --> 00:28:40.159
score is greater than zero we uh return

00:28:38.360 --> 00:28:42.480
one if the score is less than zero we

00:28:40.159 --> 00:28:44.679
return negative one which is negative

00:28:42.480 --> 00:28:47.240
and otherwise we returns

00:28:44.679 --> 00:28:48.760
zero um we have an accuracy calculation

00:28:47.240 --> 00:28:51.519
function just calculating the outputs

00:28:48.760 --> 00:28:55.840
are good and

00:28:51.519 --> 00:28:57.440
um this is uh the overall label count in

00:28:55.840 --> 00:28:59.919
the in the output so we can see there

00:28:57.440 --> 00:29:03.120
slightly more positives than there are

00:28:59.919 --> 00:29:06.080
negatives and then we can run this and

00:29:03.120 --> 00:29:10.200
we get a a score of

00:29:06.080 --> 00:29:14.760
43 and so one one thing that I have

00:29:10.200 --> 00:29:19.279
found um is I I do a lot of kind

00:29:14.760 --> 00:29:21.240
of research on how to make NLP systems

00:29:19.279 --> 00:29:23.600
better and one of the things I found

00:29:21.240 --> 00:29:26.679
really invaluable

00:29:23.600 --> 00:29:27.840
is if you're in a situation where you

00:29:26.679 --> 00:29:29.720
have a

00:29:27.840 --> 00:29:31.760
set task and you just want to make the

00:29:29.720 --> 00:29:33.760
system better on the set task doing

00:29:31.760 --> 00:29:35.159
comprehensive error analysis and

00:29:33.760 --> 00:29:37.320
understanding where your system is

00:29:35.159 --> 00:29:39.880
failing is one of the best ways to do

00:29:37.320 --> 00:29:42.200
that and I would like to do a very

00:29:39.880 --> 00:29:43.640
rudimentary version of this here and

00:29:42.200 --> 00:29:46.519
what I'm doing essentially is I'm just

00:29:43.640 --> 00:29:47.480
randomly picking uh several examples

00:29:46.519 --> 00:29:49.320
that were

00:29:47.480 --> 00:29:52.000
correct

00:29:49.320 --> 00:29:54.840
um and so like let let's look at the

00:29:52.000 --> 00:29:58.200
examples here um here the true label is

00:29:54.840 --> 00:30:00.760
zero um in this predicted one um it may

00:29:58.200 --> 00:30:03.440
not be as cutting as Woody or as true as

00:30:00.760 --> 00:30:05.039
back in the Glory Days of uh weekend and

00:30:03.440 --> 00:30:07.440
two or three things that I know about

00:30:05.039 --> 00:30:09.640
her but who else engaged in film Mak

00:30:07.440 --> 00:30:12.679
today is so cognizant of the cultural

00:30:09.640 --> 00:30:14.480
and moral issues involved in the process

00:30:12.679 --> 00:30:17.600
so what words in here are a good

00:30:14.480 --> 00:30:20.840
indication that this is a neutral

00:30:17.600 --> 00:30:20.840
sentence any

00:30:23.760 --> 00:30:28.399
ideas little bit tough

00:30:26.240 --> 00:30:30.919
huh starting to think maybe we should be

00:30:28.399 --> 00:30:30.919
using machine

00:30:31.480 --> 00:30:37.440
learning

00:30:34.080 --> 00:30:40.320
um even by the intentionally low

00:30:37.440 --> 00:30:41.559
standards of fratboy humor sority boys

00:30:40.320 --> 00:30:43.840
is a

00:30:41.559 --> 00:30:46.080
Bowser I think frat boy is maybe

00:30:43.840 --> 00:30:47.360
negative sentiment if you're familiar

00:30:46.080 --> 00:30:50.360
with

00:30:47.360 --> 00:30:51.960
us us I don't have any negative

00:30:50.360 --> 00:30:54.519
sentiment but the people who say it that

00:30:51.960 --> 00:30:55.960
way have negative senent maybe so if we

00:30:54.519 --> 00:31:01.080
wanted to go in and do that we could

00:30:55.960 --> 00:31:01.080
maybe I won't save this but

00:31:01.519 --> 00:31:08.919
uh

00:31:04.240 --> 00:31:11.840
um oh whoops I'll go back and fix it uh

00:31:08.919 --> 00:31:14.840
crass crass is pretty obviously negative

00:31:11.840 --> 00:31:14.840
right so I can add

00:31:17.039 --> 00:31:21.080
crass actually let me just add

00:31:21.760 --> 00:31:29.159
CR and then um I'll go back and have our

00:31:26.559 --> 00:31:29.159
train accurate

00:31:32.159 --> 00:31:36.240
wa maybe maybe I need to run the whole

00:31:33.960 --> 00:31:36.240
thing

00:31:36.960 --> 00:31:39.960
again

00:31:40.960 --> 00:31:45.880
and that budg the training accuracy a

00:31:43.679 --> 00:31:50.360
little um the dev test accuracy not very

00:31:45.880 --> 00:31:53.919
much so I could go through and do this

00:31:50.360 --> 00:31:53.919
um let me add

00:31:54.000 --> 00:31:58.320
unengaging so I could go through and do

00:31:56.000 --> 00:32:01.720
this all day and you probably be very

00:31:58.320 --> 00:32:01.720
bored on

00:32:04.240 --> 00:32:08.360
engage but I won't do that uh because we

00:32:06.919 --> 00:32:10.679
have much more important things to be

00:32:08.360 --> 00:32:14.679
doing

00:32:10.679 --> 00:32:16.440
um and uh so anyway we um we could go

00:32:14.679 --> 00:32:18.919
through and design all the features here

00:32:16.440 --> 00:32:21.279
but like why is this complicated like

00:32:18.919 --> 00:32:22.600
the the reason why it was complicated

00:32:21.279 --> 00:32:25.840
became pretty

00:32:22.600 --> 00:32:27.840
clear from the uh from the very

00:32:25.840 --> 00:32:29.639
beginning uh the very first example I

00:32:27.840 --> 00:32:32.200
showed you which was that was a really

00:32:29.639 --> 00:32:34.720
complicated sentence like all of us

00:32:32.200 --> 00:32:36.240
could see that it wasn't like really

00:32:34.720 --> 00:32:38.679
strongly positive it wasn't really

00:32:36.240 --> 00:32:40.519
strongly negative it was kind of like in

00:32:38.679 --> 00:32:42.919
the middle but it was in the middle and

00:32:40.519 --> 00:32:44.600
it said it in a very long way uh you

00:32:42.919 --> 00:32:46.120
know not using any clearly positive

00:32:44.600 --> 00:32:47.639
sentiment words not using any clearly

00:32:46.120 --> 00:32:49.760
negative sentiment

00:32:47.639 --> 00:32:53.760
words

00:32:49.760 --> 00:32:56.519
um so yeah basically I I

00:32:53.760 --> 00:33:00.559
improved um but what are the difficult

00:32:56.519 --> 00:33:03.720
cases uh that we saw here so the first

00:33:00.559 --> 00:33:07.639
one is low frequency

00:33:03.720 --> 00:33:09.760
words so um here's an example the action

00:33:07.639 --> 00:33:11.519
switches between past and present but

00:33:09.760 --> 00:33:13.120
the material link is too tenuous to

00:33:11.519 --> 00:33:16.840
Anchor the emotional connections at

00:33:13.120 --> 00:33:19.519
purport to span a 125 year divide so

00:33:16.840 --> 00:33:21.080
this is negative um tenuous is kind of a

00:33:19.519 --> 00:33:22.799
negative word purport is kind of a

00:33:21.080 --> 00:33:24.760
negative word but it doesn't appear very

00:33:22.799 --> 00:33:26.159
frequently so I would need to spend all

00:33:24.760 --> 00:33:29.720
my time looking for these words and

00:33:26.159 --> 00:33:32.480
trying to them in um here's yet another

00:33:29.720 --> 00:33:34.240
horse franchise mucking up its storyline

00:33:32.480 --> 00:33:36.639
with glitches casual fans could correct

00:33:34.240 --> 00:33:40.159
in their sleep negative

00:33:36.639 --> 00:33:42.600
again um so the solutions here are keep

00:33:40.159 --> 00:33:46.880
working until we get all of them which

00:33:42.600 --> 00:33:49.159
is maybe not super fun um or incorporate

00:33:46.880 --> 00:33:51.639
external resources such as sentiment

00:33:49.159 --> 00:33:52.880
dictionaries that people created uh we

00:33:51.639 --> 00:33:55.960
could do that but that's a lot of

00:33:52.880 --> 00:33:57.480
engineering effort to make something

00:33:55.960 --> 00:34:00.639
work

00:33:57.480 --> 00:34:03.720
um another one is conjugation so we saw

00:34:00.639 --> 00:34:06.600
unengaging I guess that's an example of

00:34:03.720 --> 00:34:08.359
conjugation uh some other ones are

00:34:06.600 --> 00:34:10.520
operatic sprawling picture that's

00:34:08.359 --> 00:34:12.040
entertainingly acted magnificently shot

00:34:10.520 --> 00:34:15.480
and gripping enough to sustain most of

00:34:12.040 --> 00:34:17.399
its 170 minute length so here we have

00:34:15.480 --> 00:34:19.079
magnificently so even if I added

00:34:17.399 --> 00:34:20.480
magnificent this wouldn't have been

00:34:19.079 --> 00:34:23.800
clocked

00:34:20.480 --> 00:34:26.599
right um it's basically an overlong

00:34:23.800 --> 00:34:28.839
episode of tales from the cryp so that's

00:34:26.599 --> 00:34:31.480
maybe another

00:34:28.839 --> 00:34:33.040
example um so some things that we could

00:34:31.480 --> 00:34:35.320
do or what we would have done before the

00:34:33.040 --> 00:34:37.720
modern Paradigm of machine learning is

00:34:35.320 --> 00:34:40.079
we would run some sort of normalizer

00:34:37.720 --> 00:34:42.800
like a stemmer or other things like this

00:34:40.079 --> 00:34:45.240
in order to convert this into uh the

00:34:42.800 --> 00:34:48.599
root wordss that we already have seen

00:34:45.240 --> 00:34:52.040
somewhere in our data or have already

00:34:48.599 --> 00:34:54.040
handed so that requires um conjugation

00:34:52.040 --> 00:34:55.879
analysis or morphological analysis as we

00:34:54.040 --> 00:34:57.400
say it in

00:34:55.879 --> 00:35:00.680
technicals

00:34:57.400 --> 00:35:03.960
negation this is a tricky one so this

00:35:00.680 --> 00:35:06.760
one's not nearly as Dreadful as expected

00:35:03.960 --> 00:35:08.800
so Dreadful is a pretty bad word right

00:35:06.760 --> 00:35:13.000
but not nearly as Dreadful as expected

00:35:08.800 --> 00:35:14.440
is like a solidly neutral um you know or

00:35:13.000 --> 00:35:16.359
maybe even

00:35:14.440 --> 00:35:18.920
positive I would I would say that's

00:35:16.359 --> 00:35:20.640
neutral but you know uh neutral or

00:35:18.920 --> 00:35:23.800
positive it's definitely not

00:35:20.640 --> 00:35:26.359
negative um serving s doesn't serve up a

00:35:23.800 --> 00:35:29.480
whole lot of laughs so laughs is

00:35:26.359 --> 00:35:31.880
obviously positive but not serving UPS

00:35:29.480 --> 00:35:34.440
is obviously

00:35:31.880 --> 00:35:36.839
negative so if negation modifies the

00:35:34.440 --> 00:35:38.240
word disregard it now we would probably

00:35:36.839 --> 00:35:41.440
need to do some sort of syntactic

00:35:38.240 --> 00:35:45.599
analysis or semantic analysis of

00:35:41.440 --> 00:35:47.520
some metaphor an analogy so puts a human

00:35:45.599 --> 00:35:50.640
face on a land most westerners are

00:35:47.520 --> 00:35:52.880
unfamiliar though uh this is

00:35:50.640 --> 00:35:54.960
positive green might want to hang on to

00:35:52.880 --> 00:35:58.800
that ski mask as robbery may be the only

00:35:54.960 --> 00:35:58.800
way to pay for this next project

00:35:58.839 --> 00:36:03.640
so this this is saying that the movie

00:36:01.960 --> 00:36:05.560
was so bad that the director will have

00:36:03.640 --> 00:36:08.359
to rob people in order to get money for

00:36:05.560 --> 00:36:11.000
the next project so that's kind of bad I

00:36:08.359 --> 00:36:12.880
guess um has all the depth of a waiting

00:36:11.000 --> 00:36:14.520
pool this is kind of my favorite one

00:36:12.880 --> 00:36:15.880
because it's really short and sweet but

00:36:14.520 --> 00:36:18.800
you know you need to know how deep a

00:36:15.880 --> 00:36:21.440
waiting pool is um so that's

00:36:18.800 --> 00:36:22.960
negative so the solution here I don't

00:36:21.440 --> 00:36:24.680
really even know how to handle this with

00:36:22.960 --> 00:36:26.880
a rule based system I have no idea how

00:36:24.680 --> 00:36:30.040
we would possibly do this yeah machine

00:36:26.880 --> 00:36:32.400
learning based models seem to be pretty

00:36:30.040 --> 00:36:37.000
adaptive okay and then I start doing

00:36:32.400 --> 00:36:37.000
these ones um anyone have a good

00:36:38.160 --> 00:36:46.800
idea any any other friends who know

00:36:42.520 --> 00:36:50.040
Japanese no okay um so yeah that's

00:36:46.800 --> 00:36:52.839
positive um that one's negative uh and

00:36:50.040 --> 00:36:54.920
the solution here is learn Japanese I

00:36:52.839 --> 00:36:56.800
guess or whatever other language you

00:36:54.920 --> 00:37:00.040
want to process so like obviously

00:36:56.800 --> 00:37:03.720
rule-based systems don't scale very

00:37:00.040 --> 00:37:05.119
well so um we've moved but like rule

00:37:03.720 --> 00:37:06.319
based systems don't scale very well

00:37:05.119 --> 00:37:08.160
we're not going to be using them for

00:37:06.319 --> 00:37:11.400
most of the things we do in this class

00:37:08.160 --> 00:37:14.240
but I do think it's sometimes useful to

00:37:11.400 --> 00:37:15.640
try to create one for your task maybe

00:37:14.240 --> 00:37:16.680
right at the very beginning of a project

00:37:15.640 --> 00:37:18.560
because it gives you an idea about

00:37:16.680 --> 00:37:21.160
what's really hard about the task in

00:37:18.560 --> 00:37:22.480
some cases so um yeah I wouldn't

00:37:21.160 --> 00:37:25.599
entirely discount them I'm not

00:37:22.480 --> 00:37:27.400
introducing them for no reason

00:37:25.599 --> 00:37:29.880
whatsoever

00:37:27.400 --> 00:37:34.160
so next is machine learning based anal

00:37:29.880 --> 00:37:35.400
and machine learning uh in general uh I

00:37:34.160 --> 00:37:36.640
here actually when I say machine

00:37:35.400 --> 00:37:38.160
learning I'm going to be talking about

00:37:36.640 --> 00:37:39.560
the traditional fine-tuning approach

00:37:38.160 --> 00:37:43.520
where we have a training set Dev set

00:37:39.560 --> 00:37:46.359
test set and so we take our training set

00:37:43.520 --> 00:37:49.680
we run some learning algorithm over it

00:37:46.359 --> 00:37:52.319
we have a learned feature extractor F A

00:37:49.680 --> 00:37:55.839
possibly learned feature extractor F

00:37:52.319 --> 00:37:57.880
possibly learned scoring function W and

00:37:55.839 --> 00:38:00.800
uh then we apply our inference algorithm

00:37:57.880 --> 00:38:02.839
our decision Rule and make decisions

00:38:00.800 --> 00:38:04.200
when I say possibly learned actually the

00:38:02.839 --> 00:38:06.119
first example I'm going to give of a

00:38:04.200 --> 00:38:07.760
machine learning based technique is uh

00:38:06.119 --> 00:38:10.079
doesn't have a learned feature extractor

00:38:07.760 --> 00:38:12.800
but most things that we use nowadays do

00:38:10.079 --> 00:38:12.800
have learned feature

00:38:13.200 --> 00:38:18.040
extractors so our first attempt is going

00:38:15.640 --> 00:38:21.760
to be a bag of words model uh and the

00:38:18.040 --> 00:38:27.119
way a bag of wordss model works is uh

00:38:21.760 --> 00:38:30.160
essentially we start out by looking up a

00:38:27.119 --> 00:38:33.240
Vector where one element in the vector

00:38:30.160 --> 00:38:36.240
is uh is one and all the other elements

00:38:33.240 --> 00:38:38.040
in the vector are zero and so if the

00:38:36.240 --> 00:38:40.319
word is different the position in the

00:38:38.040 --> 00:38:42.839
vector that's one will be different we

00:38:40.319 --> 00:38:46.280
add all of these together and this gives

00:38:42.839 --> 00:38:48.200
us a vector where each element is the

00:38:46.280 --> 00:38:50.359
frequency of that word in the vector and

00:38:48.200 --> 00:38:52.520
then we multiply that by weights and we

00:38:50.359 --> 00:38:55.520
get a

00:38:52.520 --> 00:38:57.160
score and um here as I said this is not

00:38:55.520 --> 00:39:00.359
a learned feature

00:38:57.160 --> 00:39:02.079
uh Vector this is basically uh sorry not

00:39:00.359 --> 00:39:04.359
a learn feature extractor this is

00:39:02.079 --> 00:39:06.200
basically a fixed feature extractor but

00:39:04.359 --> 00:39:09.839
the weights themselves are

00:39:06.200 --> 00:39:11.640
learned um so my my question is I

00:39:09.839 --> 00:39:14.599
mentioned a whole lot of problems before

00:39:11.640 --> 00:39:17.480
I mentioned infrequent words I mentioned

00:39:14.599 --> 00:39:20.760
conjugation I mentioned uh different

00:39:17.480 --> 00:39:22.880
languages I mentioned syntax and

00:39:20.760 --> 00:39:24.599
metaphor so which of these do we think

00:39:22.880 --> 00:39:25.440
would be fixed by this sort of learning

00:39:24.599 --> 00:39:27.400
based

00:39:25.440 --> 00:39:29.640
approach

00:39:27.400 --> 00:39:29.640
any

00:39:29.920 --> 00:39:35.200
ideas maybe not fixed maybe made

00:39:32.520 --> 00:39:35.200
significantly

00:39:36.880 --> 00:39:41.560
better any Brave uh brave

00:39:44.880 --> 00:39:48.440
people maybe maybe

00:39:53.720 --> 00:39:58.400
negation okay so maybe doesn't when it

00:39:55.760 --> 00:39:58.400
have a negative qu

00:40:02.960 --> 00:40:07.560
yeah yeah so for the conjugation if we

00:40:05.520 --> 00:40:09.200
had the conjugations of the stems mapped

00:40:07.560 --> 00:40:11.119
in the same position that might fix a

00:40:09.200 --> 00:40:12.920
conjugation problem but I would say if

00:40:11.119 --> 00:40:15.200
you don't do that then this kind of

00:40:12.920 --> 00:40:18.160
fixes conjugation a little bit but maybe

00:40:15.200 --> 00:40:21.319
not not really yeah kind of fix

00:40:18.160 --> 00:40:24.079
conjugation because like they're using

00:40:21.319 --> 00:40:26.760
the same there

00:40:24.079 --> 00:40:28.400
probably different variations so we

00:40:26.760 --> 00:40:31.359
learn how to

00:40:28.400 --> 00:40:33.400
classify surrounding

00:40:31.359 --> 00:40:35.000
structure yeah if it's a big enough

00:40:33.400 --> 00:40:36.760
training set you might have covered the

00:40:35.000 --> 00:40:37.880
various conjugations but if you haven't

00:40:36.760 --> 00:40:43.000
and you don't have any rule-based

00:40:37.880 --> 00:40:43.000
processing it it might still be problems

00:40:45.400 --> 00:40:50.359
yeah yeah so in frequent words if you

00:40:48.280 --> 00:40:52.560
have a large enough training set yeah

00:40:50.359 --> 00:40:54.599
you'll be able to fix it to some extent

00:40:52.560 --> 00:40:56.480
so none of the problems are entirely

00:40:54.599 --> 00:40:57.880
fixed but a lot of them are made better

00:40:56.480 --> 00:40:58.960
different languages is also made better

00:40:57.880 --> 00:41:00.119
if you have training data in that

00:40:58.960 --> 00:41:04.599
language but if you don't then you're

00:41:00.119 --> 00:41:06.240
out of BL so um so now what I'd like to

00:41:04.599 --> 00:41:10.800
do is I'd look to like to look at what

00:41:06.240 --> 00:41:15.079
our vectors represent so basically um in

00:41:10.800 --> 00:41:16.880
uh in binary classification each word um

00:41:15.079 --> 00:41:19.119
sorry so the vectors themselves

00:41:16.880 --> 00:41:21.880
represent the counts of the words here

00:41:19.119 --> 00:41:25.319
I'm talking about what the weight uh

00:41:21.880 --> 00:41:28.520
vectors or matrices correspond to and

00:41:25.319 --> 00:41:31.640
the weight uh Vector here will be

00:41:28.520 --> 00:41:33.680
positive if the word it tends to be

00:41:31.640 --> 00:41:36.680
positive if in a binary classification

00:41:33.680 --> 00:41:38.400
case in a multiclass classification case

00:41:36.680 --> 00:41:42.480
we'll actually have a matrix that looks

00:41:38.400 --> 00:41:45.480
like this where um each column or row uh

00:41:42.480 --> 00:41:47.079
corresponds to the word and each row or

00:41:45.480 --> 00:41:49.319
column corresponds to a label and it

00:41:47.079 --> 00:41:51.960
will be higher if that row tends to uh

00:41:49.319 --> 00:41:54.800
correlate with that uh that word tends

00:41:51.960 --> 00:41:56.920
to correlate that little

00:41:54.800 --> 00:41:59.240
bit so

00:41:56.920 --> 00:42:04.079
this um training of the bag of words

00:41:59.240 --> 00:42:07.720
model is can be done uh so simply that

00:42:04.079 --> 00:42:10.200
we uh can put it in a single slide so

00:42:07.720 --> 00:42:11.599
basically here uh what we do is we start

00:42:10.200 --> 00:42:14.760
out with the feature

00:42:11.599 --> 00:42:18.880
weights and for each example in our data

00:42:14.760 --> 00:42:20.800
set we extract features um the exact way

00:42:18.880 --> 00:42:23.920
I'm extracting features is basically

00:42:20.800 --> 00:42:25.720
splitting uh splitting the words using

00:42:23.920 --> 00:42:28.000
the python split function and then uh

00:42:25.720 --> 00:42:31.319
Counting number of times each word

00:42:28.000 --> 00:42:33.160
exists uh we then run the classifier so

00:42:31.319 --> 00:42:36.280
actually running the classifier is

00:42:33.160 --> 00:42:38.200
exactly the same as what we did for the

00:42:36.280 --> 00:42:42.640
uh the rule based system it's just that

00:42:38.200 --> 00:42:47.359
we have feature vectors instead and

00:42:42.640 --> 00:42:51.559
then if the predicted value is

00:42:47.359 --> 00:42:55.160
not value then for each of the

00:42:51.559 --> 00:42:56.680
features uh in the feature space we

00:42:55.160 --> 00:43:02.200
upweight

00:42:56.680 --> 00:43:03.599
the um we upweight The Weight by the

00:43:02.200 --> 00:43:06.000
vector

00:43:03.599 --> 00:43:09.920
size by or by the amount of the vector

00:43:06.000 --> 00:43:13.240
if Y is positive and we downweight the

00:43:09.920 --> 00:43:16.240
vector uh by the size of the vector if Y

00:43:13.240 --> 00:43:18.520
is negative so this is really really

00:43:16.240 --> 00:43:20.559
simple it's uh probably the simplest

00:43:18.520 --> 00:43:25.079
possible algorithm for training one of

00:43:20.559 --> 00:43:27.559
these models um but I have an

00:43:25.079 --> 00:43:30.040
example in this that you can also take a

00:43:27.559 --> 00:43:31.960
look at here's a trained bag of words

00:43:30.040 --> 00:43:33.680
classifier and we could step through

00:43:31.960 --> 00:43:34.960
this is on exactly the same data set as

00:43:33.680 --> 00:43:37.240
I did before we're training on the

00:43:34.960 --> 00:43:42.359
training set

00:43:37.240 --> 00:43:43.640
um and uh evaluating on the dev set um I

00:43:42.359 --> 00:43:45.880
also have some extra stuff like I'm

00:43:43.640 --> 00:43:47.079
Shuffling the order of the data IDs

00:43:45.880 --> 00:43:49.440
which is really important if you're

00:43:47.079 --> 00:43:53.160
doing this sort of incremental algorithm

00:43:49.440 --> 00:43:54.960
uh because uh what if what if your

00:43:53.160 --> 00:43:57.400
creating data set was ordered in this

00:43:54.960 --> 00:44:00.040
way where you have all of the positive

00:43:57.400 --> 00:44:00.040
labels on

00:44:00.359 --> 00:44:04.520
top and then you have all of the

00:44:02.280 --> 00:44:06.680
negative labels on the

00:44:04.520 --> 00:44:08.200
bottom if you do something like this it

00:44:06.680 --> 00:44:10.200
would see only negative labels at the

00:44:08.200 --> 00:44:11.800
end of training and you might have

00:44:10.200 --> 00:44:14.400
problems because your model would only

00:44:11.800 --> 00:44:17.440
predict negatives so we also Shuffle

00:44:14.400 --> 00:44:20.319
data um and then step through we run the

00:44:17.440 --> 00:44:22.559
classifier and I'm going to run uh five

00:44:20.319 --> 00:44:23.640
epochs of training through the data set

00:44:22.559 --> 00:44:27.160
uh very

00:44:23.640 --> 00:44:29.599
fast and calculate our accuracy

00:44:27.160 --> 00:44:33.280
and this got 75% accuracy on the

00:44:29.599 --> 00:44:36.160
training data set and uh 56% accuracy on

00:44:33.280 --> 00:44:40.000
the Deb data set so uh if you remember

00:44:36.160 --> 00:44:41.520
our rule-based classifier had 42 uh 42

00:44:40.000 --> 00:44:43.880
accuracy and now our training based

00:44:41.520 --> 00:44:45.760
classifier has 56 accuracy but it's

00:44:43.880 --> 00:44:49.359
overfitting heavily to the training side

00:44:45.760 --> 00:44:50.880
so um basically this is a pretty strong

00:44:49.359 --> 00:44:53.480
advertisement for why we should be using

00:44:50.880 --> 00:44:54.960
machine learning you know I the amount

00:44:53.480 --> 00:44:57.800
of code that we had for this machine

00:44:54.960 --> 00:44:59.720
learning model is basically very similar

00:44:57.800 --> 00:45:02.680
um it's not using any external libraries

00:44:59.720 --> 00:45:02.680
but we're getting better at

00:45:03.599 --> 00:45:08.800
this

00:45:05.800 --> 00:45:08.800
cool

00:45:09.559 --> 00:45:16.000
so cool any any questions

00:45:13.520 --> 00:45:18.240
here and so I'm going to talk about the

00:45:16.000 --> 00:45:20.760
connection to between this algorithm and

00:45:18.240 --> 00:45:22.839
neural networks in the next class um

00:45:20.760 --> 00:45:24.200
because this actually is using a very

00:45:22.839 --> 00:45:26.319
similar training algorithm to what we

00:45:24.200 --> 00:45:27.480
use in neural networks with some uh

00:45:26.319 --> 00:45:30.079
particular

00:45:27.480 --> 00:45:32.839
assumptions cool um so what's missing in

00:45:30.079 --> 00:45:34.800
bag of words um still handling of

00:45:32.839 --> 00:45:36.880
conjugation or compound words is not

00:45:34.800 --> 00:45:39.160
perfect it we can do it to some extent

00:45:36.880 --> 00:45:41.079
to the point where we can uh memorize

00:45:39.160 --> 00:45:44.079
things so I love this movie I love this

00:45:41.079 --> 00:45:46.920
movie another thing is handling word Ser

00:45:44.079 --> 00:45:49.240
uh similarities so I love this movie and

00:45:46.920 --> 00:45:50.720
I adore this movie uh these basically

00:45:49.240 --> 00:45:52.119
mean the same thing as humans we know

00:45:50.720 --> 00:45:54.200
they mean the same thing so we should be

00:45:52.119 --> 00:45:56.079
able to take advantage of that fact to

00:45:54.200 --> 00:45:57.839
learn better models but we're not doing

00:45:56.079 --> 00:46:02.760
that in this model at the moment because

00:45:57.839 --> 00:46:05.440
each unit is uh treated as a atomic unit

00:46:02.760 --> 00:46:08.040
and there's no idea of

00:46:05.440 --> 00:46:11.040
similarity also handling of combination

00:46:08.040 --> 00:46:12.760
features so um I love this movie and I

00:46:11.040 --> 00:46:14.920
don't love this movie I hate this movie

00:46:12.760 --> 00:46:17.079
and I don't hate this movie actually

00:46:14.920 --> 00:46:20.400
this is a little bit tricky because

00:46:17.079 --> 00:46:23.240
negative words are slightly indicative

00:46:20.400 --> 00:46:25.280
of it being negative but actually what

00:46:23.240 --> 00:46:28.119
they do is they negate the other things

00:46:25.280 --> 00:46:28.119
that you're saying in the

00:46:28.240 --> 00:46:36.559
sentence

00:46:30.720 --> 00:46:40.480
so um like love is positive hate is

00:46:36.559 --> 00:46:40.480
negative but like don't

00:46:50.359 --> 00:46:56.079
love it's actually kind of like this

00:46:52.839 --> 00:46:59.359
right like um Love is very positive POS

00:46:56.079 --> 00:47:01.760
hate is very negative but don't love is

00:46:59.359 --> 00:47:04.680
like slightly less positive than don't

00:47:01.760 --> 00:47:06.160
hate right so um It's actually kind of

00:47:04.680 --> 00:47:07.559
tricky because you need to combine them

00:47:06.160 --> 00:47:10.720
together and figure out what's going on

00:47:07.559 --> 00:47:12.280
based on that another example that a lot

00:47:10.720 --> 00:47:14.160
of people might not think of immediately

00:47:12.280 --> 00:47:17.880
but is super super common in sentiment

00:47:14.160 --> 00:47:20.160
analysis or any other thing is butt so

00:47:17.880 --> 00:47:22.599
basically what but does is it throws

00:47:20.160 --> 00:47:24.160
away all the stuff that you said before

00:47:22.599 --> 00:47:26.119
um and you can just pay attention to the

00:47:24.160 --> 00:47:29.000
stuff that you saw beforehand so like we

00:47:26.119 --> 00:47:30.440
could even add this to our um like if

00:47:29.000 --> 00:47:31.760
you want to add this to your rule based

00:47:30.440 --> 00:47:33.240
classifier you can do that you just

00:47:31.760 --> 00:47:34.640
search for butt and delete everything

00:47:33.240 --> 00:47:37.240
before it and see if that inputs your

00:47:34.640 --> 00:47:39.240
accuracy might be might be a fun very

00:47:37.240 --> 00:47:43.480
quick thing

00:47:39.240 --> 00:47:44.880
to cool so the better solution which is

00:47:43.480 --> 00:47:46.800
what we're going to talk about for every

00:47:44.880 --> 00:47:49.480
other class other than uh other than

00:47:46.800 --> 00:47:52.160
this one is neural network models and

00:47:49.480 --> 00:47:55.800
basically uh what they do is they do a

00:47:52.160 --> 00:47:59.400
lookup of uh dense word embeddings so

00:47:55.800 --> 00:48:02.520
instead of looking up uh individual uh

00:47:59.400 --> 00:48:04.640
sparse uh vectors individual one hot

00:48:02.520 --> 00:48:06.920
vectors they look up dense word

00:48:04.640 --> 00:48:09.680
embeddings and then throw them into some

00:48:06.920 --> 00:48:11.880
complicated function to extract features

00:48:09.680 --> 00:48:16.359
and based on the features uh multiply by

00:48:11.880 --> 00:48:18.280
weights and get a score um and if you're

00:48:16.359 --> 00:48:20.359
doing text classification in the

00:48:18.280 --> 00:48:22.520
traditional way this is normally what

00:48:20.359 --> 00:48:23.760
you do um if you're doing text

00:48:22.520 --> 00:48:25.960
classification with something like

00:48:23.760 --> 00:48:27.280
prompting you're still actually doing

00:48:25.960 --> 00:48:29.960
this because you're calculating the

00:48:27.280 --> 00:48:32.960
score of the next word to predict and

00:48:29.960 --> 00:48:34.720
that's done in exactly the same way so

00:48:32.960 --> 00:48:37.760
uh even if you're using a large language

00:48:34.720 --> 00:48:39.359
model like GPT this is still probably

00:48:37.760 --> 00:48:41.800
happening under the hood unless open the

00:48:39.359 --> 00:48:43.400
eye invented something that very

00:48:41.800 --> 00:48:45.559
different in Alien than anything else

00:48:43.400 --> 00:48:48.440
that we know of but I I'm guessing that

00:48:45.559 --> 00:48:48.440
that propably hasn't

00:48:48.480 --> 00:48:52.880
happen um one nice thing about neural

00:48:50.880 --> 00:48:54.480
networks is neural networks

00:48:52.880 --> 00:48:57.559
theoretically are powerful enough to

00:48:54.480 --> 00:49:00.000
solve any task if you make them uh deep

00:48:57.559 --> 00:49:01.160
enough or wide enough uh like if you

00:49:00.000 --> 00:49:04.520
make them wide enough and then if you

00:49:01.160 --> 00:49:06.799
make them deep it also helps further so

00:49:04.520 --> 00:49:08.079
anytime somebody says well you can't

00:49:06.799 --> 00:49:11.119
just solve that problem with neural

00:49:08.079 --> 00:49:13.240
networks you know that they're lying

00:49:11.119 --> 00:49:15.720
basically because they theoretically can

00:49:13.240 --> 00:49:17.359
solve every problem uh but you have you

00:49:15.720 --> 00:49:19.799
have issues of data you have issues of

00:49:17.359 --> 00:49:23.079
other things like that so you know they

00:49:19.799 --> 00:49:23.079
don't just necessarily work

00:49:23.119 --> 00:49:28.040
outs cool um so the final thing I'd like

00:49:26.400 --> 00:49:29.319
to talk about is the road map going

00:49:28.040 --> 00:49:31.319
forward some of the things I'm going to

00:49:29.319 --> 00:49:32.799
cover in the class and some of the

00:49:31.319 --> 00:49:35.200
logistics

00:49:32.799 --> 00:49:36.799
issues so um the first thing I'm going

00:49:35.200 --> 00:49:38.240
to talk about in the class is language

00:49:36.799 --> 00:49:40.559
modeling fun

00:49:38.240 --> 00:49:42.720
fundamentals and uh so this could

00:49:40.559 --> 00:49:44.240
include language models uh that just

00:49:42.720 --> 00:49:46.559
predict the next words it could include

00:49:44.240 --> 00:49:50.559
language models that predict the output

00:49:46.559 --> 00:49:51.599
given the uh the input or the prompt um

00:49:50.559 --> 00:49:54.559
I'm going to be talking about

00:49:51.599 --> 00:49:56.520
representing words uh how how we get

00:49:54.559 --> 00:49:59.319
word representation subword models other

00:49:56.520 --> 00:50:01.440
things like that uh then go kind of

00:49:59.319 --> 00:50:04.200
deeper into language modeling uh how do

00:50:01.440 --> 00:50:07.799
we do it how do we evaluate it other

00:50:04.200 --> 00:50:10.920
things um sequence encoding uh and this

00:50:07.799 --> 00:50:13.240
is going to cover things like uh

00:50:10.920 --> 00:50:16.280
Transformers uh self attention modals

00:50:13.240 --> 00:50:18.559
but also very quickly cnns and rnns

00:50:16.280 --> 00:50:20.880
which are useful in some

00:50:18.559 --> 00:50:22.200
cases um and then we're going to

00:50:20.880 --> 00:50:24.040
specifically go very deep into the

00:50:22.200 --> 00:50:25.960
Transformer architecture and also talk a

00:50:24.040 --> 00:50:27.280
little bit about some of the modern uh

00:50:25.960 --> 00:50:30.240
improvements to the Transformer

00:50:27.280 --> 00:50:31.839
architecture so the Transformer we're

00:50:30.240 --> 00:50:33.839
using nowadays is very different than

00:50:31.839 --> 00:50:36.200
the Transformer that was invented in

00:50:33.839 --> 00:50:37.240
2017 uh so we're going to talk well I

00:50:36.200 --> 00:50:38.760
wouldn't say very different but

00:50:37.240 --> 00:50:41.359
different enough that it's important so

00:50:38.760 --> 00:50:43.280
we're going to talk about some of those

00:50:41.359 --> 00:50:45.079
things second thing I'd like to talk

00:50:43.280 --> 00:50:47.000
about is training and inference methods

00:50:45.079 --> 00:50:48.839
so this includes uh generation

00:50:47.000 --> 00:50:52.119
algorithms uh so we're going to have a

00:50:48.839 --> 00:50:55.520
whole class on how we generate text uh

00:50:52.119 --> 00:50:58.319
in different ways uh prompting how uh we

00:50:55.520 --> 00:50:59.720
can prompt things I hear uh world class

00:50:58.319 --> 00:51:01.799
prompt engineers make a lot of money

00:50:59.720 --> 00:51:05.480
nowadays so uh you'll want to pay

00:51:01.799 --> 00:51:08.760
attention to that one um and instruction

00:51:05.480 --> 00:51:11.520
tuning uh so how do we train models to

00:51:08.760 --> 00:51:13.720
handle a lot of different tasks and

00:51:11.520 --> 00:51:15.839
reinforcement learning so how do we uh

00:51:13.720 --> 00:51:18.520
you know like actually generate outputs

00:51:15.839 --> 00:51:19.839
uh kind of Judge them and then learn

00:51:18.520 --> 00:51:22.599
from

00:51:19.839 --> 00:51:25.880
there also experimental design and

00:51:22.599 --> 00:51:28.079
evaluation so experimental design uh so

00:51:25.880 --> 00:51:30.480
how do we design an experiment well uh

00:51:28.079 --> 00:51:32.000
so that it backs up what we want to be

00:51:30.480 --> 00:51:34.559
uh our conclusions that we want to be

00:51:32.000 --> 00:51:37.000
backing up how do we do human annotation

00:51:34.559 --> 00:51:38.880
of data in a reliable way this is

00:51:37.000 --> 00:51:41.160
getting harder and harder as models get

00:51:38.880 --> 00:51:43.359
better and better because uh getting

00:51:41.160 --> 00:51:45.000
humans who don't care very much about

00:51:43.359 --> 00:51:48.559
The annotation task they might do worse

00:51:45.000 --> 00:51:51.119
than gp4 so um you need to be careful of

00:51:48.559 --> 00:51:52.240
that also debugging and interpretation

00:51:51.119 --> 00:51:53.960
technique so what are some of the

00:51:52.240 --> 00:51:55.160
automatic techniques that you can do to

00:51:53.960 --> 00:51:57.720
quickly figure out what's going wrong

00:51:55.160 --> 00:52:00.040
with your models and improve

00:51:57.720 --> 00:52:01.599
them and uh bias and fairness

00:52:00.040 --> 00:52:04.200
considerations so it's really really

00:52:01.599 --> 00:52:05.799
important nowadays uh that models are

00:52:04.200 --> 00:52:07.880
being deployed to real people in the

00:52:05.799 --> 00:52:09.880
real world and like actually causing

00:52:07.880 --> 00:52:11.760
harm to people in some cases that we

00:52:09.880 --> 00:52:15.160
need to be worried about

00:52:11.760 --> 00:52:17.000
that Advanced Training in architectures

00:52:15.160 --> 00:52:19.280
so we're going to talk about distill

00:52:17.000 --> 00:52:21.400
distillation and quantization how can we

00:52:19.280 --> 00:52:23.520
make small language models uh that

00:52:21.400 --> 00:52:24.880
actually still work well like not large

00:52:23.520 --> 00:52:27.559
you can run them on your phone you can

00:52:24.880 --> 00:52:29.920
run them on your local

00:52:27.559 --> 00:52:31.640
laptop um ensembling and mixtures of

00:52:29.920 --> 00:52:33.480
experts how can we combine together

00:52:31.640 --> 00:52:34.760
multiple models in order to create

00:52:33.480 --> 00:52:35.880
models that are better than the sum of

00:52:34.760 --> 00:52:38.799
their

00:52:35.880 --> 00:52:40.720
parts and um retrieval and retrieval

00:52:38.799 --> 00:52:43.920
augmented

00:52:40.720 --> 00:52:45.480
generation long sequence models uh so

00:52:43.920 --> 00:52:49.920
how do we handle long

00:52:45.480 --> 00:52:52.240
outputs um and uh we're going to talk

00:52:49.920 --> 00:52:55.760
about applications to complex reasoning

00:52:52.240 --> 00:52:57.760
tasks code generation language agents

00:52:55.760 --> 00:52:59.920
and knowledge-based QA and information

00:52:57.760 --> 00:53:04.160
extraction I picked

00:52:59.920 --> 00:53:06.760
these because they seem to be maybe the

00:53:04.160 --> 00:53:09.880
most important at least in research

00:53:06.760 --> 00:53:11.440
nowadays and also they cover uh the

00:53:09.880 --> 00:53:13.640
things that when I talk to people in

00:53:11.440 --> 00:53:15.280
Industry are kind of most interested in

00:53:13.640 --> 00:53:17.559
so hopefully it'll be useful regardless

00:53:15.280 --> 00:53:19.799
of uh whether you plan on doing research

00:53:17.559 --> 00:53:22.839
or or plan on doing industry related

00:53:19.799 --> 00:53:24.160
things uh by by the way the two things

00:53:22.839 --> 00:53:25.920
that when I talk to people in Industry

00:53:24.160 --> 00:53:29.599
they're most interested in are Rag and

00:53:25.920 --> 00:53:31.079
code generation at the moment for now um

00:53:29.599 --> 00:53:32.319
so those are ones that you'll want to

00:53:31.079 --> 00:53:34.680
pay attention

00:53:32.319 --> 00:53:36.599
to and then finally we have a few

00:53:34.680 --> 00:53:40.079
lectures on Linguistics and

00:53:36.599 --> 00:53:42.720
multilinguality um I love Linguistics

00:53:40.079 --> 00:53:44.839
but uh to be honest at the moment most

00:53:42.720 --> 00:53:47.760
of our Cutting Edge models don't

00:53:44.839 --> 00:53:49.240
explicitly use linguistic structure um

00:53:47.760 --> 00:53:50.799
but I still think it's useful to know

00:53:49.240 --> 00:53:52.760
about it especially if you're working on

00:53:50.799 --> 00:53:54.880
multilingual things especially if you're

00:53:52.760 --> 00:53:57.040
interested in very robust generalization

00:53:54.880 --> 00:53:58.920
to new models so we're going to talk a

00:53:57.040 --> 00:54:02.599
little bit about that and also

00:53:58.920 --> 00:54:06.079
multilingual LP I'm going to have

00:54:02.599 --> 00:54:09.119
fure so also if you have any suggestions

00:54:06.079 --> 00:54:11.400
um we have two guest lecture slots still

00:54:09.119 --> 00:54:12.799
open uh that I'm trying to fill so if

00:54:11.400 --> 00:54:15.440
you have any things that you really want

00:54:12.799 --> 00:54:16.440
to hear about um I could either add them

00:54:15.440 --> 00:54:19.319
to the

00:54:16.440 --> 00:54:21.079
existing you know content or I could

00:54:19.319 --> 00:54:23.240
invite a guest lecturer who's working on

00:54:21.079 --> 00:54:24.079
that topic so you know please feel free

00:54:23.240 --> 00:54:26.760
to tell

00:54:24.079 --> 00:54:29.160
me um then the class format and

00:54:26.760 --> 00:54:32.280
structure uh the class

00:54:29.160 --> 00:54:34.000
content my goal is to learn in detail

00:54:32.280 --> 00:54:36.640
about building NLP systems from a

00:54:34.000 --> 00:54:40.520
research perspective so this is a 700

00:54:36.640 --> 00:54:43.599
level course so it's aiming to be for

00:54:40.520 --> 00:54:46.960
people who really want to try new and

00:54:43.599 --> 00:54:49.280
Innovative things in uh kind of natural

00:54:46.960 --> 00:54:51.359
language processing it's not going to

00:54:49.280 --> 00:54:52.760
focus solely on reimplementing things

00:54:51.359 --> 00:54:54.319
that have been done before including in

00:54:52.760 --> 00:54:55.280
the project I'm going to be expecting

00:54:54.319 --> 00:54:58.480
everybody to do something something

00:54:55.280 --> 00:54:59.920
that's kind of new whether it's coming

00:54:58.480 --> 00:55:01.359
up with a new method or applying

00:54:59.920 --> 00:55:03.559
existing methods to a place where they

00:55:01.359 --> 00:55:05.079
haven't been used before or building out

00:55:03.559 --> 00:55:06.640
things for a new language or something

00:55:05.079 --> 00:55:08.359
like that so that's kind of one of the

00:55:06.640 --> 00:55:11.480
major goals of this

00:55:08.359 --> 00:55:13.000
class um learn basic and advanced topics

00:55:11.480 --> 00:55:15.559
in machine learning approaches to NLP

00:55:13.000 --> 00:55:18.359
and language models learn some basic

00:55:15.559 --> 00:55:21.480
linguistic knowledge useful in NLP uh

00:55:18.359 --> 00:55:23.200
see case studies of NLP applications and

00:55:21.480 --> 00:55:25.680
learn how to identify unique problems

00:55:23.200 --> 00:55:29.039
for each um one thing i' like to point

00:55:25.680 --> 00:55:31.160
out is I'm not going to cover every NLP

00:55:29.039 --> 00:55:32.920
application ever because that would be

00:55:31.160 --> 00:55:35.520
absolutely impossible NLP is being used

00:55:32.920 --> 00:55:37.079
in so many different areas nowadays but

00:55:35.520 --> 00:55:38.960
what I want people to pay attention to

00:55:37.079 --> 00:55:41.280
like even if you're not super interested

00:55:38.960 --> 00:55:42.400
in code generation for example what you

00:55:41.280 --> 00:55:44.200
can do is you can look at code

00:55:42.400 --> 00:55:46.160
generation look at how people identify

00:55:44.200 --> 00:55:47.680
problems look at the methods that people

00:55:46.160 --> 00:55:50.880
have proposed to solve those unique

00:55:47.680 --> 00:55:53.039
problems and then kind of map that try

00:55:50.880 --> 00:55:54.799
to do some generalization onto your own

00:55:53.039 --> 00:55:57.799
problems of Interest so uh that's kind

00:55:54.799 --> 00:56:00.280
of the goal of the NLP

00:55:57.799 --> 00:56:02.440
applications finally uh learning how to

00:56:00.280 --> 00:56:05.160
debug when and where NLP systems fail

00:56:02.440 --> 00:56:08.200
and build improvements based on this so

00:56:05.160 --> 00:56:10.200
um ever since I was a graduate student

00:56:08.200 --> 00:56:12.720
this has been like one of the really

00:56:10.200 --> 00:56:15.920
important things that I feel like I've

00:56:12.720 --> 00:56:17.440
done well or done better than some other

00:56:15.920 --> 00:56:19.280
people and I I feel like it's a really

00:56:17.440 --> 00:56:21.119
good way to like even if you're only

00:56:19.280 --> 00:56:22.680
interested in improving accuracy knowing

00:56:21.119 --> 00:56:25.039
why your system's failing still is the

00:56:22.680 --> 00:56:27.599
best way to do that I so I'm going to

00:56:25.039 --> 00:56:30.559
put a lot of emphasis on

00:56:27.599 --> 00:56:32.559
that in terms of the class format um

00:56:30.559 --> 00:56:36.280
before class for some classes there are

00:56:32.559 --> 00:56:37.880
recommended reading uh this can be

00:56:36.280 --> 00:56:39.559
helpful to read I'm never going to

00:56:37.880 --> 00:56:41.119
expect you to definitely have read it

00:56:39.559 --> 00:56:42.480
before the class but I would suggest

00:56:41.119 --> 00:56:45.160
that maybe you'll get more out of the

00:56:42.480 --> 00:56:47.319
class if you do that um during class

00:56:45.160 --> 00:56:48.079
we'll have the lecture um in discussion

00:56:47.319 --> 00:56:50.559
with

00:56:48.079 --> 00:56:52.359
everybody um sometimes we'll have a code

00:56:50.559 --> 00:56:55.839
or data walk

00:56:52.359 --> 00:56:58.760
um actually this is a a little bit old I

00:56:55.839 --> 00:57:01.880
I have this slide we're this year we're

00:56:58.760 --> 00:57:04.160
going to be adding more uh code and data

00:57:01.880 --> 00:57:07.400
walks during office hours and the way it

00:57:04.160 --> 00:57:09.400
will work is one of the Tas we have

00:57:07.400 --> 00:57:11.160
seven Tas who I'm going to introduce

00:57:09.400 --> 00:57:15.000
very soon but one of the Tas will be

00:57:11.160 --> 00:57:16.839
doing this kind of recitation where you

00:57:15.000 --> 00:57:18.200
um where we go over a library so if

00:57:16.839 --> 00:57:19.480
you're not familiar with the library and

00:57:18.200 --> 00:57:21.960
you want to be more familiar with the

00:57:19.480 --> 00:57:23.720
library you can join this and uh then

00:57:21.960 --> 00:57:25.400
we'll be able to do this and this will

00:57:23.720 --> 00:57:28.240
cover things like

00:57:25.400 --> 00:57:31.039
um pie torch and sentence piece uh we're

00:57:28.240 --> 00:57:33.280
going to start out with hugging face um

00:57:31.039 --> 00:57:36.559
inference stuff like

00:57:33.280 --> 00:57:41.520
VM uh debugging software like

00:57:36.559 --> 00:57:41.520
Xeno um what were the other

00:57:41.960 --> 00:57:47.200
ones oh the open AI API and light llm

00:57:45.680 --> 00:57:50.520
other stuff like that so we we have lots

00:57:47.200 --> 00:57:53.599
of them planned we'll uh uh we'll update

00:57:50.520 --> 00:57:54.839
that um and then after class after

00:57:53.599 --> 00:57:58.079
almost every class we'll have a question

00:57:54.839 --> 00:58:00.079
quiz um and the quiz is intended to just

00:57:58.079 --> 00:58:02.000
you know make sure that you uh paid

00:58:00.079 --> 00:58:04.480
attention to the material and are able

00:58:02.000 --> 00:58:07.520
to answer questions about it we will aim

00:58:04.480 --> 00:58:09.559
to release it on the day of the course

00:58:07.520 --> 00:58:11.599
the day of the actual lecture and it

00:58:09.559 --> 00:58:14.559
will be due at the end of the following

00:58:11.599 --> 00:58:15.960
day of the lecture so um it will be

00:58:14.559 --> 00:58:18.920
three questions it probably shouldn't

00:58:15.960 --> 00:58:20.680
take a whole lot of time but um uh yeah

00:58:18.920 --> 00:58:23.400
so we'll H

00:58:20.680 --> 00:58:26.319
that in terms of assignments assignment

00:58:23.400 --> 00:58:28.640
one is going to be build your own llama

00:58:26.319 --> 00:58:30.200
and so what this is going to look like

00:58:28.640 --> 00:58:32.680
is we're going to give you a partial

00:58:30.200 --> 00:58:34.319
implementation of llama which is kind of

00:58:32.680 --> 00:58:37.960
the most popular open source language

00:58:34.319 --> 00:58:40.160
model nowadays and ask you to fill in um

00:58:37.960 --> 00:58:42.839
ask you to fill in the parts we're going

00:58:40.160 --> 00:58:45.920
to train a very small version of llama

00:58:42.839 --> 00:58:47.319
on a small data set and get it to work

00:58:45.920 --> 00:58:48.880
and the reason why it's very small is

00:58:47.319 --> 00:58:50.480
because the smallest actual version of

00:58:48.880 --> 00:58:53.039
llama is 7 billion

00:58:50.480 --> 00:58:55.359
parameters um and that might be a little

00:58:53.039 --> 00:58:58.400
bit difficult to train with

00:58:55.359 --> 00:59:00.680
resources um for assignment two we're

00:58:58.400 --> 00:59:04.559
going to try to do an NLP task from

00:59:00.680 --> 00:59:06.920
scratch and so the way this will work is

00:59:04.559 --> 00:59:08.520
we're going to give you an assignment

00:59:06.920 --> 00:59:10.880
which we're not going to give you an

00:59:08.520 --> 00:59:13.400
actual data set and instead we're going

00:59:10.880 --> 00:59:15.760
to ask you to uh perform data creation

00:59:13.400 --> 00:59:19.359
modeling and evaluation for a specified

00:59:15.760 --> 00:59:20.640
task and so we're going to tell you uh

00:59:19.359 --> 00:59:22.599
what to do but we're not going to tell

00:59:20.640 --> 00:59:26.400
you exactly how to do it but we're going

00:59:22.599 --> 00:59:29.680
to try to give as conrete directions as

00:59:26.400 --> 00:59:32.359
we can um

00:59:29.680 --> 00:59:34.160
yeah will you be given a parameter limit

00:59:32.359 --> 00:59:36.559
on the model so that's a good question

00:59:34.160 --> 00:59:39.119
or like a expense limit or something

00:59:36.559 --> 00:59:40.440
like that um I maybe actually I should

00:59:39.119 --> 00:59:44.240
take a break from the assignments and

00:59:40.440 --> 00:59:46.520
talk about compute so right now um for

00:59:44.240 --> 00:59:49.319
assignment one we're planning on having

00:59:46.520 --> 00:59:51.599
this be able to be done either on a Mac

00:59:49.319 --> 00:59:53.520
laptop with an M1 or M2 processor which

00:59:51.599 --> 00:59:57.079
I think a lot of people have or Google

00:59:53.520 --> 00:59:59.839
collab um so it should be like

00:59:57.079 --> 01:00:02.160
sufficient to use free computational

00:59:59.839 --> 01:00:03.640
resources that you have for number two

01:00:02.160 --> 01:00:06.079
we'll think about that I think that's

01:00:03.640 --> 01:00:08.280
important we do have Google cloud

01:00:06.079 --> 01:00:11.520
credits for $50 for everybody and I'm

01:00:08.280 --> 01:00:13.440
working to get AWS credits for more um

01:00:11.520 --> 01:00:18.160
but the cloud providers nowadays are

01:00:13.440 --> 01:00:19.680
being very stingy so um so it's uh been

01:00:18.160 --> 01:00:22.160
a little bit of a fight to get uh

01:00:19.680 --> 01:00:23.680
credits but I I it is very important so

01:00:22.160 --> 01:00:28.480
I'm going to try to get as as many as we

01:00:23.680 --> 01:00:31.119
can um and so yeah I I think basically

01:00:28.480 --> 01:00:32.280
uh there will be some sort of like limit

01:00:31.119 --> 01:00:34.480
on the amount of things you can

01:00:32.280 --> 01:00:36.240
practically do and so because of that

01:00:34.480 --> 01:00:39.920
I'm hoping that people will rely very

01:00:36.240 --> 01:00:43.359
heavily on pre-trained models um or uh

01:00:39.920 --> 01:00:46.079
yeah pre-trained models

01:00:43.359 --> 01:00:49.599
and yeah so that that's the the short

01:00:46.079 --> 01:00:52.799
story B um the second thing uh the

01:00:49.599 --> 01:00:54.720
assignment three is to do a survey of

01:00:52.799 --> 01:00:57.920
some sort of state-ofthe-art research

01:00:54.720 --> 01:01:00.760
resarch and do a reimplementation of

01:00:57.920 --> 01:01:02.000
this and in doing this again you will

01:01:00.760 --> 01:01:03.440
have to think about something that's

01:01:02.000 --> 01:01:06.359
feasible within computational

01:01:03.440 --> 01:01:08.680
constraints um and so you can discuss

01:01:06.359 --> 01:01:11.839
with your Tas about uh about the best

01:01:08.680 --> 01:01:13.920
way to do this um and then the final

01:01:11.839 --> 01:01:15.400
project is to perform a unique project

01:01:13.920 --> 01:01:17.559
that either improves on the state-of-the

01:01:15.400 --> 01:01:21.000
art with respect to whatever you would

01:01:17.559 --> 01:01:23.440
like to improve with this could be uh

01:01:21.000 --> 01:01:25.280
accuracy for sure this could be

01:01:23.440 --> 01:01:27.760
efficiency

01:01:25.280 --> 01:01:29.599
it could be some sense of

01:01:27.760 --> 01:01:31.520
interpretability but if it's going to be

01:01:29.599 --> 01:01:33.599
something like interpretability you'll

01:01:31.520 --> 01:01:35.440
have to discuss with us what that means

01:01:33.599 --> 01:01:37.240
like how we measure that how we can like

01:01:35.440 --> 01:01:40.839
actually say that you did a good job

01:01:37.240 --> 01:01:42.839
with improving that um another thing

01:01:40.839 --> 01:01:44.680
that you can do is take whatever you

01:01:42.839 --> 01:01:47.280
implemented for assignment 3 and apply

01:01:44.680 --> 01:01:49.039
it to a new task or apply it to a new

01:01:47.280 --> 01:01:50.760
language that has never been examined

01:01:49.039 --> 01:01:53.119
before so these are also acceptable

01:01:50.760 --> 01:01:54.240
final projects but basically the idea is

01:01:53.119 --> 01:01:55.559
for the final project you need to do

01:01:54.240 --> 01:01:57.480
something something new that hasn't been

01:01:55.559 --> 01:01:59.880
done before and create new knowledge

01:01:57.480 --> 01:02:04.520
with the respect

01:01:59.880 --> 01:02:07.640
toy um so for this the instructor is me

01:02:04.520 --> 01:02:09.920
um I'm uh looking forward to you know

01:02:07.640 --> 01:02:13.599
discussing and working with all of you

01:02:09.920 --> 01:02:16.119
um for TAS we have seven Tas uh two of

01:02:13.599 --> 01:02:18.319
them are in transit so they're not here

01:02:16.119 --> 01:02:22.279
today um the other ones uh Tas would you

01:02:18.319 --> 01:02:22.279
mind coming up uh to introduce

01:02:23.359 --> 01:02:26.359
yourself

01:02:28.400 --> 01:02:32.839
so um yeah nhir and akshai couldn't be

01:02:31.599 --> 01:02:34.039
here today because they're traveling

01:02:32.839 --> 01:02:37.119
I'll introduce them later because

01:02:34.039 --> 01:02:37.119
they're coming uh next

01:02:40.359 --> 01:02:46.480
time cool and what I'd like everybody to

01:02:43.000 --> 01:02:48.680
do is say um like you know what your

01:02:46.480 --> 01:02:53.079
name is uh what

01:02:48.680 --> 01:02:55.799
your like maybe what you're interested

01:02:53.079 --> 01:02:57.319
in um and the reason the goal of this is

01:02:55.799 --> 01:02:59.200
number one for everybody to know who you

01:02:57.319 --> 01:03:00.720
are and number two for everybody to know

01:02:59.200 --> 01:03:03.440
who the best person to talk to is if

01:03:00.720 --> 01:03:03.440
they're interested in

01:03:04.200 --> 01:03:09.079
particular hi uh I'm

01:03:07.000 --> 01:03:15.520
Aila second

01:03:09.079 --> 01:03:15.520
year I work on language and social

01:03:16.200 --> 01:03:24.559
and I'm I'm a second this year PhD

01:03:21.160 --> 01:03:26.799
student Grand and Shar with you I search

01:03:24.559 --> 01:03:28.480
is like started in the border of MP and

01:03:26.799 --> 01:03:31.000
computer interaction with a lot of work

01:03:28.480 --> 01:03:32.640
on automating parts of the developer

01:03:31.000 --> 01:03:35.319
experience to make it easier for anyone

01:03:32.640 --> 01:03:35.319
to

01:03:39.090 --> 01:03:42.179
[Music]

01:03:47.520 --> 01:03:53.279
orif

01:03:50.079 --> 01:03:54.680
everyone first

01:03:53.279 --> 01:03:57.119
year

01:03:54.680 --> 01:04:00.119
[Music]

01:03:57.119 --> 01:04:03.559
I don't like updating primar models I

01:04:00.119 --> 01:04:03.559
hope to not update Prim

01:04:14.599 --> 01:04:19.400
modelm yeah thanks a lot everyone and

01:04:17.200 --> 01:04:19.400
yeah

01:04:20.839 --> 01:04:29.400
than and so we will um we'll have people

01:04:25.640 --> 01:04:30.799
uh kind of have office hours uh every ta

01:04:29.400 --> 01:04:32.880
has office hours at a regular time

01:04:30.799 --> 01:04:34.480
during the week uh please feel free to

01:04:32.880 --> 01:04:38.400
come to their office hours or my office

01:04:34.480 --> 01:04:41.960
hours um I think they are visha are they

01:04:38.400 --> 01:04:43.880
posted on the site or okay yeah they

01:04:41.960 --> 01:04:47.240
they either are or will be posted on the

01:04:43.880 --> 01:04:49.720
site very soon um and come by to talk

01:04:47.240 --> 01:04:51.480
about anything uh if there's nobody in

01:04:49.720 --> 01:04:53.079
my office hours I'm happy to talk about

01:04:51.480 --> 01:04:54.599
things that are unrelated but if there's

01:04:53.079 --> 01:04:58.039
lots of people waiting outside or I

01:04:54.599 --> 01:05:00.319
might limit it to uh like um just things

01:04:58.039 --> 01:05:02.480
about the class so cool and we have

01:05:00.319 --> 01:05:04.760
Patza we'll be checking that regularly

01:05:02.480 --> 01:05:06.839
uh striving to get you an answer in 24

01:05:04.760 --> 01:05:12.240
hours on weekdays over weekends we might

01:05:06.839 --> 01:05:16.000
not so um yeah so that's all for today

01:05:12.240 --> 01:05:16.000
are there any questions