YT YouTube2Text

YouTube2Text → Transcripts → GCI World 2026 September Session7 Final Assignment Part

GCI World 2026 September Session7 Final Assignment Part — Transcript

by GCI World (Matsuo-Iwasawa Lab) · 5,296 words · 978 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Thank you to Kashi San.
0:02And hi guys, my name is Shinsuke. So,
0:04let's uh get back to the more like
0:08serious part.
0:10Uh which is the about the final
0:12assignment.
0:14Okay, so let's get started. So, first uh
0:18uh let me introduce myself. My name is
0:20Shinsuke Komiya.
0:22And yeah, I'm I'm a postdoc researcher
0:25at Mazola and I'm uh the lead uh
0:28developer of this year's educational
0:29materials
0:31together with Aki. And my research
0:33interest uh lies in the intersection of
0:35dynamical systems and machine learning.
0:37And my hobby is listening to uh
0:39practically any kinds of music.
0:43And about the uh uh
0:45instructions of final assignment, uh
0:48what I'm going to be covering today is
0:51uh shown here.
0:52Which is one, the overview of final
0:54assignment. And after that, uh I'm going
0:56to be talking about uh uh the general
0:59thing about uh what does a good business
1:02proposal look like.
1:04And and which is followed by how does it
1:07apply to your uh final assignment. And
1:10after that, I'm going to be covering
1:11frequently asked questions. And after
1:14that, I'm going to be having uh a few
1:15minutes of of Q&A section.
1:19Okay, let's uh start from the overview
1:21of final assignment.
1:24So, as Aki mentioned at the beginning,
1:27the final assignment is a required to uh
1:31realize any state of completion. So,
1:34please uh please submit a final
1:36assignment. Don't forget that.
1:38Uh you need to pass uh the final
1:40assignment. Okay?
1:43And you uh as for the final assignment,
1:45you're required to make presentation
1:47slides uh to propose your business
1:49strategy using provided assets.
1:52And deadline is written here, November
1:5428th, 11:00 a.m. UTC. And please submit
1:57your presentation slides together with
1:59the novel files that you use for your
2:01analysis, and your list of references.
2:06Okay, so the
2:07next part, how to get started. So, this
2:10is pretty much the same
2:12as Aki explained at the beginning
2:15of the
2:16competition.
2:17So, what you have to do is that please
2:20go
2:21is to go to the shared folder,
2:23and please go to the 05 final
2:26assignment.
2:27And there you're going to be seeing a
2:29folder named final assignment,
2:32and together with the tutorial PDF.
2:35And first you have to download the final
2:37assignment folder to your local PC,
2:40and it should be downloaded it should be
2:42downloaded as a zip file.
2:46And please unzip the downloaded zip file
2:49maybe in your local environment, then
2:52upload the final assignment folder to
2:53your My Drive. So, I'm going to skip the
2:56real, you know,
2:58the hands-on doing this, but yeah, I
3:01think you should
3:02do this.
3:04Pretty much the same as you did in the
3:06competition.
3:07And you can
3:08always use your local environment if you
3:12don't find the online Google Colab
3:14environment uncomfortable.
3:18And okay, so if you
3:22go to the final assignment folder and
3:23unzip it and upload it to the My Drive,
3:26this pretty much looks like this, I
3:28guess.
3:30And this contains four elements, each of
3:33which
3:34is like this. So, first you're going to
3:37be seeing telecom folder,
3:39which contains all the data you'll be
3:41using.
3:42A second,
3:44some document,
3:46which is additional description of the
3:48data.
3:49And third, you're going to be seeing
3:51readme docs. This is the the very
3:54important instructions of the final
3:56assignment.
3:57I think.
3:59And after that you're going to be
4:03seeing that the tutorial IP Y and B,
4:05which is the data analysis example.
4:07So, first, maybe you should take a look
4:11at the read me docs. This This is
4:14which is the really important
4:16instruction.
4:18So, if you hit this
4:20document,
4:22you're most likely seeing this
4:26section.
4:27And the
4:29scenario and objective
4:31section, it goes like, "You're a new
4:33associate at an IT consulting firm. The
4:36sales team has just secured a POC
4:39project from a company. The company has
4:41a large data set, but has limited
4:43in-house expertise to use them for
4:45business.
4:46Your job is to run EDA and build at
4:49least one machine learning model on the
4:51data set.
4:52And from evidence of market analysis,
4:54you need to define the client core
4:56business problem. Then, you'll be create
4:58a data-driven business proposal with
5:00quantified impact."
5:02Okay, so this is basically what you
5:03should do in a final assignment. So,
5:05maybe I can end my
5:08explanation now.
5:10Yeah, which
5:12is just a joke. So, yeah. So,
5:15yeah, so if you're done with the final
5:17assignment,
5:18made all the presentation and notebook,
5:20how to submit your final assignment?
5:24So, yeah, it's as always, you do this
5:27through the Omni campus. So, please log
5:29in to the Omni campus, you're going to
5:30be finding
5:32uh the report name final assignment,
5:33okay?
5:35And I think you should be seeing this
5:37already
5:38because it's already open.
5:40Sorry. And
5:43by clicking this, you'll find these
5:46things.
5:47And you're going to be putting in a
5:49title, list of references, Jupiter
5:51notebook together with the slide, which
5:54is the most important thing.
5:56And this is accompanied by uh
6:00questionnaires. I think I set up four or
6:03five questions, I guess.
6:05And this is about the surveys of AI
6:07usage. And please answer it and honestly
6:11because we we we're not be uh, this just
6:14a pure survey and this doesn't affect
6:16your score at all.
6:18In in a good way or in a bad way.
6:22Okay, so far for the uh, bureaucratic
6:25things about the final assignment. So,
6:27I'm going to moving on to what does a
6:29good business proposal look like uh,
6:31section.
6:33So,
6:34uh, this is what you saw in your readme
6:36docs.
6:38And you're required to um, create a
6:42business proposal based on the given
6:43data
6:45and uh, identify the client's business
6:47problem and present a solution.
6:50So, at first, um, I think uh, we should
6:52cover we should go through what a good
6:55business proposal looks like uh, in the
6:57first place.
7:00And this is how the business proposal
7:04uh, usually uh, looks like. Uh, this is
7:06the a typical workflow we uh, always um,
7:10do.
7:12And this is pretty much complicated,
7:14isn't it? So, uh,
7:16and it's a bit overwhelming. But, no
7:18worries. I'm going to be explaining uh,
7:20this step by step, okay?
7:25So, although the workflow looked a
7:27little bit overwhelming and complicated
7:30um, this uh, should be similar up to
7:33three key points.
7:35Which is
7:36uh, first um, a good business proposal
7:39should specify an important issue worth
7:41solving.
7:43And second, it includes a hypothesis to
7:45solve uh, that issue.
7:48And third, this is verifiable.
7:50Okay?
7:51So, I'm going to be explaining one by
7:53one starting uh, with the specifying an
7:56important issue worth solving.
7:59So, this amounts um, to here, the very
8:02first part of this workflow.
8:07And let's imagine a guy, a client, who
8:10came up to you who comes up to up to you
8:13and saying, "Our product A is suffering
8:15from sluggish sales, although it has
8:17more functions, more settings, more
8:20fabulous, you know, functions and better
8:22performance than its competitors. What
8:24are we going to do about this?"
8:27So, what are you
8:29going to do about this?
8:31So, I'm going to take like 5 seconds, so
8:32please think about the answer.
8:37Okay, so you might answer like this,
8:40"The performance of the product is
8:42already great, so the marketing strategy
8:44seems to have a problem. Let's start
8:46from running more web advertising."
8:48Is it really good? Do you think it's
8:50really good?
8:52So,
8:53I
8:54it's kind of
8:56um
8:57naive answer, I I would say. So, it's
8:59like you're jumping So, this guy
9:02is jumping onto the very, you know, easy
9:05way to solve this problem.
9:07So, uh before that, you have to stop and
9:11think and ask yourself
9:13if you're referring to the real issue
9:17here.
9:19So, to do that, let's
9:22focus on the three things that this
9:26client is saying.
9:29So, this guy is mentioning about the
9:32sluggish sales.
9:33But, about the sluggish sales, you have
9:35to
9:36if you if you talk if you want to talk
9:38about the sluggish sales,
9:40you have to be
9:41caring about how big the market might
9:44be.
9:45And you might ask yourself, "Do other
9:48products also face sluggish sales?"
9:50So, if the market is is shrinking
9:53itself,
9:54in that case, however, it might happen
9:58that no matter how your your product is
10:02is fantastic,
10:03you might get sluggish sales.
10:06So, to know that, you have to uh some
10:09background analysis, a background
10:11search.
10:13And the client was saying about more
10:15functions.
10:16And if you talk about more functions, it
10:19it usually it often happens that um um
10:22more functions
10:24uh turned out unnecessary for the
10:26customers.
10:28So, you might want to ask, are customers
10:29happy with more uh functions?
10:32The product having more functions?
10:34Or uh isn't it worsening its uh
10:38product's operability?
10:41So, to check that, you might want to run
10:44some customer survey.
10:47And the clients was talking about better
10:49performance. So, this is a little bit
10:51like a more functions, but is really
10:52necessary? You might want to ask, is it
10:54neces- is it really necessary to have
10:56better performance?
10:58Or is this a better performance um
11:00making the price higher so that that the
11:03customers um um
11:05are uh um might hesitate buying that.
11:11So, to uh know get more information
11:13about this, um
11:19I think yeah, I'm I'm making a mistake,
11:21but um yeah, so um
11:23maybe um
11:25yeah, maybe you you can run a customer
11:27survey or or maybe um uh the another
11:30product with a less um um
11:33performance might cover up this um
11:35product.
11:38So, uh what I want to say here is that
11:41you must find and focus on the core
11:44issue
11:45uh behind uh what the client saying.
11:48Okay?
11:50And collecting background information
11:52and some quantitative analysis, like uh
11:55some very easy, like first step EDA,
11:58will help you find an issue uh worth
12:00working on.
12:01So, it's like uh this is uh this usually
12:04happens, and this um um
12:07finding the core finding the real issue
12:10um
12:11uh of
12:13in in any situation is the most
12:15difficult part.
12:17And this
12:19um
12:19applies to research. So, in in research,
12:22you have to make a good question, which
12:24is the most difficult part.
12:26And it that is that quite as as I can
12:29mention at the beginning, this is quite
12:31the same as the business setting, I
12:32guess.
12:33And um Einstein
12:37um
12:37uh in the past said that the formulation
12:40problem is often more essential than its
12:42solution, which may be merely a matter
12:44of mathematical experiment and skill.
12:46So,
12:47uh the the the the formulation of a
12:49problem is the most uh important part.
12:52And the solution uh comes just next to
12:56uh setting the right um question.
12:59So, yeah, the the this always happens
13:03that people uh always tend to, you know,
13:05more difficult technical, like
13:07mathematics technical, you know, um
13:09things technical mathematical, you know,
13:12uh machine learning models, whatever,
13:14but that's is just um um step to take
13:18to achieve the goal. And the goal is
13:20more important, of course.
13:22Okay?
13:24Okay, so uh next. So, a good business
13:26proposal um should include a hypothesis
13:29to solve the issue. So, this comes to
13:32the step two and four
13:34uh in the workflow I first showed.
13:38And once you find a good issue
13:41to uh work on, you should settle your
13:42hypothesis on that issue.
13:45So, a hypothesis, as you um you may
13:48know,
13:49is a tentative answer to your issue that
13:52seems the closest to the correct answer.
13:56Why do you have to uh build a
13:57hypothesis?
13:59Without focusing on the hypothesis, you
14:01can't take action or make the decisions.
14:04So, uh usually, if you face a client and
14:09a client having a problem,
14:11um once after, you know,
14:14even after finding a good issue, good uh
14:17starting point
14:19to uh start your workflow um over.
14:24Um that there
14:26uh almost always are um a bunch of
14:29problems you should be you should you
14:31have to solve. So, there are many many
14:33problems you have to solve if if you
14:36even if even after the focusing the
14:38issue, okay?
14:41And um
14:43I I
14:45I I think you can't
14:47you basically can't answer every issue
14:50by yourself, okay?
14:53And you you don't you even don't have to
14:55do that, okay? So, you what you have to
14:57do is that um you might be at a loss
15:00with the uh list of the problems you're
15:02going to be seeing.
15:04But in order to, you know, get the
15:07things um proceed,
15:10you don't have to answer one by one all
15:11the problems, but you have to focus
15:14on the problem you have to uh
15:17dealing be dealing with.
15:20And to do that, hypothesis uh works a
15:22lot, okay? And the lack of hypothesis uh
15:25paralyze people, especially when they're
15:27given only data. So, this happened this
15:29actually happened to me myself
15:32um until quite recently.
15:34And um I was kind of a person who like,
15:37you know, uh getting all the result of
15:39all the uh data or the experiment result
15:42whatever.
15:44Uh I uh what I would do was that I would
15:47you know, do the the do the most, you
15:50know, detailed and scrutinized
15:52experiment that to search what's
15:55happening behind this data.
15:58And um
15:59after that, I'm going to be getting
16:01nothing, okay? So, the most meticulous
16:04most like, you know,
16:05um
16:07um detailed experiments without having a
16:10hypothesis result, you know, having some
16:13um um having a like clear goal
16:16to first go to,
16:18will get you nothing, okay?
16:20So, you have first have to um uh build a
16:23question you have to answer.
16:26And you have to uh concretize your
16:28question so that you you get the uh uh
16:31uh the first tentative goal
16:34you would want to um hit, okay?
16:38And on this initial hypothesis what one
16:41important thing is that this initial
16:42hypothesis can be wrong and but that you
16:45need to uh uh
16:47run the uh like, you know, recursive
16:49step
16:51to correct the hypothesis on after um
16:54um doing some experiments or something.
16:58So, how can you make a good hypothesis?
17:02This is where your data analysis skills
17:04come in and this is where the you what
17:08you learned in this lectures come in.
17:12And um to uh make a good hypothesis it
17:15should be um
17:17uh built on the quantitative analysis
17:19and insights based on it, okay?
17:23And in in the real business settings,
17:25you can also uh collect real opinion
17:27from the people involved uh
17:29uh
17:30yeah.
17:32And about the uh
17:35uh building a hypothesis uh sorry, the
17:37EDA um you you you do in in making good
17:41hypothesis.
17:43So, actually EDA uh turns up
17:47here and here, two steps uh between uh
17:51the uh problem hypothesis creation and
17:53proposal uh before the uh proposal idea
17:56section. So, there I would say
18:00are two types of EDA.
18:02The first of which is to determine your
18:05issue and initial hypothesis.
18:08Okay, so the first uh EDA is a little
18:11bit of uh you know, um
18:13to grasp an overview and insights of
18:17what's happening uh behind the data
18:20and it can be explorative but not too
18:22haphazard.
18:23Okay? And after uh maybe finding the
18:26issue and building your hypothesis
18:29and after that you have to focus on your
18:32hypothesis and run another ED another
18:35type of EDA which is more focused and to
18:39to deepen and evaluate your hypothesis.
18:43Okay?
18:44And they should try to answer initial
18:46hypothesis as I mentioned. And this
18:50would be more
18:51concrete.
18:52Maybe with machine learning models,
18:54extra boost, random forest, whatever
18:58to
18:59answer your first hypothesis.
19:03And lastly the good business proposal
19:06should be verifiable.
19:08So this comes after proposal
19:12sorry after building hypothesis
19:15and last step and this arrow that goes
19:19from five to two. Okay?
19:24So
19:26the specify your core issue and building
19:28your hypothesis and proposal it is not
19:31the end. So you need to verify and
19:33validate your proposal
19:34over and over again. Okay?
19:37So maybe you need to answer questions as
19:40after building a hypothesis you should
19:43answer
19:44such questions as what the expected
19:47profit or the projected profit of your
19:49proposal.
19:51So
19:52you might think about a bunch of
19:54hypothesis and maybe you should take the
19:57one with the most
19:59projected profit. So that's
20:02important.
20:04So what's the limitation? So no proposal
20:07we don't expect you know perfect
20:09proposal without any flaws.
20:11Okay no proposal on the earth is
20:13perfect. So
20:14so you're you're it's it's quite okay to
20:17have you know weak points or limitations
20:20but you should be aware of that. And
20:22hopefully you should be dealing with
20:25with your verification
20:28cycle limitations
20:30as as far as you can, okay?
20:35And how do you continuously test and
20:37verify your proposal? So, how do you
20:38reflect feedback from the customers?
20:40This is another important
20:42um
20:43things to do to run the recursive step.
20:46Okay?
20:47So, um
20:49this is another kind of, you know,
20:52workflow, but things
20:54usually goes like this.
20:57And first, you find your issue, as I
20:59mentioned, and you build a hypothesis
21:02for this issue first, and do the
21:04verification. And this always should be
21:06repetitive in that
21:10looking after this result of this
21:13verification and validation, whatever.
21:15You have to renew, update your
21:18hypothesis
21:19after this, okay? And you have to this
21:22is the
21:23this is quite just the same as the
21:25competition, as Aki explained earlier.
21:29This is a recursive step you're going to
21:30be seeing. So, this is pretty much the
21:33same as the research.
21:35And this is pretty much the same as the
21:37business kind of
21:40workflow, I guess.
21:43And you might What you might get is like
21:45a meta circle in which you're going to
21:48be
21:49renewing the issue
21:52So, after running through many over and
21:54over again this cycle, you will be
21:58noticing another important issue that
22:01you have been missed or or others
22:04have missed.
22:05So, this is the very, you know, novel
22:08and great issue you find.
22:12And then
22:13yeah, this repetitive circle circle
22:17might have the outer step to to renew
22:20the issue itself.
22:22Okay.
22:24And how does it apply our
22:26finance assignment? So, let's go to the
22:31maybe I'm running out of a little bit of
22:33time, but let's go to the tutorial ipynb
22:37and
22:39which looks like
22:41this.
22:46Okay, if you go to the shared folder
22:49and if you hit the finance assignment
22:52you're going to be seeing tutorial ipynb
22:55which pretty much
22:56looks like this.
22:58And
23:00what it's doing is that
23:03um
23:06it's first downloading sorry
23:10it's downloading the data
23:12and doing some very basic EDAs.
23:15And in that there the creator is
23:20uh
23:21leading a hypothesis uh which I we're
23:24going to be looking at later.
23:28And
23:30yes, so this is a database that the
23:33database sorry the data set we're going
23:35to be using in a finance assignment is
23:37the telecom data set
23:39which is about the
23:41the the list of bunch of clients
23:44and who's turned
23:46who's
23:48continued their contract. So, trading is
23:51like you know
23:53the churning churning is the correct
23:56word. It is stop
23:58the churning is to to you know giving up
24:01the contract
24:03with the
24:04the carrier telecommunication company
24:07carrier, okay?
24:08And so what you are seeing here is that
24:12you're basically looking at the basic uh
24:16data what does what this data looks like
24:20first
24:21and you're doing in section three you're
24:23doing some a very basic
24:26EDAs
24:27where your um uh
24:31depicting the distributions of state and
24:33churned
24:35um people distribution.
24:38And um
24:40and it's closer look here. So,
24:44we're going to be looking at the uh
24:47uh
24:48the equipment age distribution.
24:50And in this um
24:53histogram,
24:55you can see that the the churned
24:57distribution is shifted to the right
25:00um compared to the state distribution,
25:03which means um
25:05the churners have longer ownership of uh
25:11the device. Okay? So, this is very
25:13natural because, you know, if you if you
25:15think your your cell phone or whatever
25:17gets old, you might want to, you know,
25:19change that. Okay?
25:22And yeah.
25:25Somebody else is uh continuous, but uh
25:29yeah, let's get back to Yeah, I'm only
25:32having a glance at the notebook, and
25:34let's get back to the upper notation.
25:36And so, yeah, how today's instructions
25:39apply to your final assignment. The most
25:41straightforward issue maybe is as you
25:43saw in the uh uh in the notebook.
25:47So, uh the the very straightforward
25:50issue is the how to increase the profit
25:53by decrease the churn rate. Okay? So,
25:55the churn rate is so was, you know, uh
25:58uh distributed and was shown in the
26:00notebook. And this might be the most
26:03straightforward and and important thing
26:05to to, you know, uh
26:07have at the first
26:09uh glance.
26:11And there can be, however, other choices
26:12of issues to approach. So, you have to
26:15um
26:16So, you you can, you know, propose your
26:17issue, another issue in your uh uh
26:21final assignment, but it has to be
26:23coupled with EDN background knowledge
26:25and and a something
26:27After EDA, you you gain some insights on
26:30what's happening, and you can uh build
26:32your hypothesis.
26:34Okay? And the hypothesis then the
26:37example notebook
26:39built was do longer equipment ages
26:42increase the generation.
26:44And you can explain this hypothesis
26:48if if it looks persuasive by adding more
26:50focused EDAs.
26:52Okay?
26:53And to make your proposal verifiable in
26:56final assignment, you can receive the
26:58real feedback of your proposal, but at
27:00least you should show what makes your
27:04proposal legitimate, the project that
27:06profit, the estimated profit of the
27:08proposal,
27:10and the way to continuously test,
27:11verify, and update the proposal. Okay?
27:14With the limitation.
27:18And the about the yeah, so
27:22the good thing is that we're going to be
27:24publishing
27:26the past outstanding student example
27:28slides
27:29in the
27:31uh uh
27:32shared folder.
27:34So although the you can't copy and paste
27:36of course, so this is the thing the data
27:38set this example is dealing with the
27:41different from the one we're using right
27:43now. So yeah, but but this pretty much
27:48would be the good example, good goal for
27:51you to get to. Okay?
27:54So frequently asked questions.
27:57So first question, am I going to fail
27:59final assignment even if I submit my
28:00report if the quality is too bad? Yes,
28:03we have our own criteria for grading
28:05final assignment. So just submitting
28:07your final assignment is not enough.
28:09Okay? So please make you know good final
28:12assignment.
28:13There should be so many submissions. Do
28:15you use AI for grading?
28:17We're afraid we can't answer anything
28:18about the concrete process of grading.
28:20So we're sorry, but we can't say
28:23we're using AI or not, but at least we
28:26can say that
28:27there are so many graders, okay? So many
28:29graders in our team. So that's uh,
28:32as long as I can say.
28:35We want to know the grading rubrics. I
28:37think I I found this question in already
28:41posted in the Q&A dashboard. And we were
28:44afraid that we're not planning to
28:46release a rubrics, okay? Sorry about
28:48that.
28:49But, in fact,
28:52the requirement section in Redmi docs
28:56tells you pretty much about how your
28:58report is going to be graded. So, we
29:00strongly recommend you to read it
29:03carefully, entirely,
29:05and follow the requirements. So,
29:07yes, all I can say is
29:10yeah, please read
29:11the uh,
29:14thing. Read what what I written in this
29:16requirement section, okay?
29:19Can I use AI for my final assignment?
29:21Yes, we actually recommend the use of
29:23AI, which accelerate the hypothesis
29:25evaluation circle
29:27cycle. Any AI usage will not affect your
29:29score, but please
29:32share your URL or output chat history in
29:35the
29:37the
29:38when when you submit your final
29:40assignment, okay?
29:42And additional notes about your AI not
29:44usage, so using AI doesn't necessarily
29:47decrease the workload, okay? So,
29:50nowadays, you know, students
29:53happy with AI because, you know,
29:56you can just
29:57maybe you like
29:59I predict that the you might be, you
30:02know, putting on all the data set, all
30:05the materials, maybe this material, this
30:07very material,
30:09and put in the cloud code or ChatGPT or
30:11whatever, and output your slide, and
30:14that's done. That that's never done,
30:15okay? And AI helps you dramatically in
30:18providing methodologies, implementing
30:20code, and visualizing. But, AI still
30:23have problems in specifying core issues
30:26with big picture or building nice
30:28hypotheses or making representation easy
30:31for human to understand, okay?
30:34So, I think these two points are pretty
30:37much answered in the previous
30:40instructions, but the last one last one
30:42is a very big problem
30:44we have to deal with. Okay, so um
30:47AI tend to lack, you know, comprehensive
30:50and broader perspective with which
30:53um
30:54which makes human beings
30:58easier to, you know, understand any kind
31:01of problem.
31:03And they they often tend to be
31:06too particular about the insignificant
31:08very insignificant very detailed very
31:11meticulous very unnecessary things. So,
31:16I don't recommend, you know,
31:18in you know, one rally
31:21you put all the materials you need in
31:23the final assignment and
31:25uh you know,
31:27press an enter key and yeah, that's
31:29done. I don't recommend that, okay? At
31:32least you have to, you know, iterate the
31:34process
31:35um I don't know about five times um at
31:38that at least, I don't know, maybe 10 or
31:41maybe 20 or 50 times
31:44to get the uh the ideal one that you
31:47that you really want.
31:49Okay? So, it's a
31:51recursive iterative stuff.
31:54Uh can you extend the deadline? So, I
31:56have my school exam. So, this is the one
31:58of the most frequently asked questions
32:01after when it comes to, you know,
32:05if if if it's near the deadline. No, we
32:08are afraid we will never make a deadline
32:10extension, okay? So, please make sure
32:12you can submit before deadline. No
32:14school schedule, technical difficulties,
32:16other situations, other explanations, no
32:19other excuses can be your excuse, okay?
32:21So, please submit your report with
32:23plenty time to spare.
32:25Can I make my final assignment in my own
32:26language? Please don't do that, okay?
32:28Please use English
32:30in your presentation.
32:33Other important notes include final
32:34assignment should be in presentation
32:36slides
32:37equals or less than 15 pages, not in
32:40document. Make it readable to those who
32:42first read your report. So, it's
32:45pretty much related to what I said in
32:47the AI usage
32:48disclaimer. But, the
32:50LLM tend to use obscure wordings, no
32:53explained jargons. Please don't do that.
32:57Please um
32:58make it human-friendly
33:01um report, okay?
33:05And put figures in your presentation
33:07slides, not just in your notebook. So,
33:09we see some reports saying that
33:12there are no figures
33:15in the presentation, but
33:17instead please look at the notebook
33:19output. So,
33:21we we actually are looking at your
33:24notebook, but the evaluation is but
33:27mostly for mostly based on the
33:30presentation slide itself. So, please
33:32put figures in the presentation, okay?
33:34Don't change your theme. Now, this is
33:36another important thing. So,
33:38we see some participants
33:41um dealing with their own problem, their
33:43own like, you know, problem in their
33:45country. So, so no matter how, you know,
33:49good
33:51these presentations are, we we have to
33:53we must fail them, okay? So, don't do
33:56that.
33:57Again, we will never make a data
33:58expansion. So, this is so important I to
34:02say twice.
34:03But, submit with plenty time to spare,
34:06okay? And so, yes, that's
34:09it from my side. And maybe I should
34:13yes, go to the Q&A.
34:17Maybe four or five questions.
34:21What are the final assignment
34:22competition requirement including topic
34:24choice, format, grading, rank is
34:25deadline, eligibility for credit? As I
34:28mentioned, the reworks on web been not
34:30open. Sorry about that. But, again,
34:33please read the requirements.
34:35Okay, requirements in your tutorial.
34:38Sorry, read me.
34:41How will the final assignment
34:42competitive evaluated? Sorry, so read
34:44the requirements section, okay?
34:48How can I submit the final assignment
34:49correctly in five minutes square
34:50competition? Submit the final Okay, so
34:53this is what I explained. So this is via
34:56the Omni campus and you can think you
34:58can find
34:59the
35:01uh
35:01page from which you can submit your
35:03final assignment.
35:05How will competition submission
35:06evaluated and ranked? So this is I think
35:09Yeah, so all these things about
35:11evaluation,
35:12we apologize, but we will not saying
35:15anything about the concrete rubrics or
35:17complete evaluation criteria. But
35:21But one thing this is really important,
35:23so I'm saying again and again, but
35:24please read your instruction. Please
35:27read your requirements, okay?
35:30One word by one word, okay?
35:32And this is so important. So this this
35:34as long as I can say is that this the
35:36requirements section is pretty much
35:38pretty much related to the evaluation
35:41rubrics. So this is
35:43as long as I can say.
35:46How will the competition be evaluated
35:47and ranked? So this is about the
35:48competition. How do you
35:54Maybe a questions a few more questions
35:58final assignment working determine the
36:00features of my rules.
36:02Issue of competition.
36:05I think they're answered by Akia, I
36:07guess. I was wondering if there are any
36:09guideline using
36:12Okay, so I'm I'm going to be answering
36:15this.
36:19I was wondering if there are any
36:21guidelines for using AI tools such as
36:23ChatGPT cloud as learning resources. For
36:26example, if you use them to clarify
36:28concepts, understand methods of
36:29discussing oops
36:31Where did it go? Discussing possible
36:33approaches to improve our models without
36:36asking them to generate code or build
36:38the model for us. How should these
36:40interactions be documented? If we use
36:42multiple conversations throughout the
36:44project, should we include all of them
36:47or only the director related to our
36:48work? I'm also interested in
36:50understanding how these guidelines help
36:52ensure fair evaluation for everyone.
36:54Distinguish
36:55between A using AI as a learning
36:57resource and having AI generate the code
36:59or develop
37:00the model on our behalf. Okay, so um
37:04yes, um I would say um
37:08yes, please share um
37:10as many AI chat any records you made uh
37:14with your AI as possible.
37:17So, uh it shouldn't be like
37:19it shouldn't be like um
37:21um
37:23yeah, you should be yeah,
37:26it has to be like on URL, I guess.
37:29to um to share your record, but um
37:32please like um include all all the
37:35conversation you have with your AI,
37:37okay? And um
37:42How should these interactions be
37:43documented? So,
37:46yes, so um honestly, you can make um
37:49your AI build or write your code by your
37:52um instead of yourself.
37:54And um
37:55but we we're focusing in in at least in
37:58the final assignment, we we're focusing
38:01on the final derivative, okay? Final PDF
38:04you're going to be making.
38:06So, um
38:09yes, so the uh whether you're you made
38:12your code by yourself or
38:14or if you let the AI do that, uh
38:17wouldn't matter as long as uh the final
38:20derivative is the same, okay? So, we're
38:23looking purely looking at the at a PDF,
38:25the slides presentation you make.
38:28And yeah, so I hope that
38:31uh answered your question.
38:34Um
38:35okay, maybe
38:37uh
38:38time is up for me. Sorry, I couldn't
38:40answer so many so many questions, but I
38:44hope these
38:46um
38:47frequently asked questions pretty much
38:49answered your
38:50questions.
38:52Okay, and
38:56back to the slides.
38:58Yes, we have attendance survey but no
39:01homework today, okay?
39:03And homework for today is I think this
39:07is pretty much mistaken but yeah, so
39:09this this is the same as
39:12always, okay?
39:15Okay, so that's it for the
39:19today. So, thank you very much everyone
39:22and I hope you guys do your best in the
39:24competition of final assignment.
39:26Bye.

About this transcript

This page contains the full transcript of GCI World 2026 September Session7 Final Assignment Part by GCI World (Matsuo-Iwasawa Lab), generated from the public captions YouTube serves with the video. The transcript has 5,296 words across 978 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.