00:00Hello guys and girlies, welcome back.
00:07In the last lecture, we have seen a framework.
00:10We have seen an existing problem definition,
00:18and we have seen a machine learning solution.
00:20This is a iterative process.
00:21You have seen this process.
00:23This is the last lecture.
00:25However, point number 1,
00:29which is the problem definition.
00:32It is necessary to define that
00:35what is the problem solving?
00:37Is it machine learning problem?
00:39Is it not?
00:41It is not that I have a non-machine learning problem
00:44to solve a problem solving.
00:49For example, you have a story about a story.
00:52A farmer has a story about it.
00:54It's a story about it.
00:56It usually takes time to open a ΡΡquilεΌ penhu package into it.
00:59ItΓ£y subscribe to a box Gabo.
01:00Don't forget the more money buying this mode of place.
01:02It isIK meaning that the bot tool takes place.
01:05The label is running in one hand.
01:08The label explains the earlier question.
01:09Do you want to beΡΡΠ²Π΅Π½Π½ΡΠΉ?
01:11Now, when you have a double-numak, you have a double-dubkey.
01:16Double-numak will be a double-dubkey.
01:18So, this means that you don't always have a machine learning that you have to do.
01:26If you have to ask yourself, you have to ask yourself,
01:29you have to ask yourself, left, right, right, turn around,
01:34then you have to do shopping mall.
01:36Now, when you have a exact location available,
01:41a set of instructions available for a destination,
01:44then you have straight away code.
01:46AFL-CADD, it will be a problem.
01:48Why to bring it to a machine learning?
01:51You have to ask yourself,
01:53is it actually a machine learning problem?
01:56How will we know?
01:58We will see which machine learning types available are,
02:02and which problem will be solved.
02:04because this is very important.
02:06So, first of all, I have to remind you,
02:08that basically,
02:10we have machine learning three types.
02:12One is supervised,
02:14sorry for my bad handwriting,
02:16and one is unsupervised,
02:18and one is unsupervised,
02:21and one is transfer.
02:25I have to say transfer.
02:26I have to say transfer.
02:27I have to say transfer.
02:28I have to say transfer learning.
02:29So, I have to say transfer learning.
02:31I will say learning.
02:32This is learning,
02:33this is learning,
02:34and this is learning.
02:35This is learning.
02:36Okay.
02:37Now, we will have to go with it.
02:38So, three types of solutions exist.
02:42Reinforce learning.
02:43I will ask you to the back side of the reason.
02:46The reason is that,
02:47Reinforce learning,
02:49practical applications,
02:51in the real world,
02:52which we apply in the job,
02:54is very limited.
02:55So,
02:56we will have to keep a little back line
02:58which is exactly available for problem solving.
03:00We will focus on that.
03:01So,
03:02first of all,
03:03supervised learning.
03:04If we have a set of problems,
03:07then we need to know
03:08that supervised learning
03:10which problem will solve.
03:12If you have seen this,
03:13if you have seen the example,
03:14you should see.
03:15Turn on the list.
03:16So,
03:17there are two subjects.
03:19The first is classification.
03:23The second one,
03:24the first is regression.
03:29Classification and regression.
03:31The second one is regression.
03:32Now,
03:33I do have to tell you,
03:35The same thing.
03:36The other one is that
03:38you have to label some data
03:39or regression.
03:40We need to know
03:41Let's say, let's say, I have 10,000 of the data and 10,000 of the bikes.
03:45So, I have machine learning to make bikes and the cars and the cars.
03:49We have label data, we have to make it.
03:51Or then, I have regression to make a number predict.
03:55How to make it? For example, I have housing sales data.
03:58This data has to make it.
04:01Now, I have to predict the next point.
04:04So, simple, I have regression to make it.
04:06And then, it should be here and it should be here.
04:09Simple regression problem.
04:11You have a label data.
04:13And you have to predict.
04:15No problem.
04:16Classification issue.
04:19However, imagine that you are working on a company and you have a data.
04:25And the company says that there are numbers.
04:28And this data is what represents the data.
04:33So, let me tell you about this.
04:36So, why do you have a label data?
04:39Okay.
04:40And the other thing, you have a specific output.
04:43You don't have a specific output.
04:45So, what do you do?
04:46What do you do?
04:47You say, okay.
04:48So, I don't know.
04:49I don't know.
04:50What output is the output.
04:51Why don't I ask unsupervised learning?
04:54Because it is unsupervised.
04:55I don't know.
04:56I don't know.
04:57I don't know.
04:58I don't know.
04:59I don't know.
05:00You have to think of the data.
05:01But I think of the data.
05:02I am using them.
05:03And I don't know.
05:04You have to think of what you are doing.
05:05To understand.
05:06You are using it.
05:07This is a question.
05:08The data is not okay.
05:10-
05:12The data is real.
05:14There are the data.
05:16different customer
05:18or support
05:20or
05:21let's say
05:22kitchen
05:23like
05:24imagine
05:25Amazon
05:26customers
05:27data
05:28now
05:29a promotion
05:30in which
05:31kitchen
05:32things
05:33now
05:34Amazon
05:35send
05:36email
05:37of course
05:38not send
05:39because
05:4095%
05:41people
05:42only
05:43you can
05:46solve
05:47Amazon
05:48say
05:49data
05:50science
05:51why
05:52you can
05:53use
05:54data
05:55data
05:56and
05:58group
05:59which
06:00are
06:01in
06:02price
06:03range
06:04because
06:05they
06:07are
06:08unsupervised
06:09data
06:10and
06:11you can
06:13see
06:14for example
06:15here
06:16price
06:17and
06:18let's say
06:19wait
06:20item
06:21wait
06:22so
06:23some
06:24people
06:25here
06:26and
06:27some
06:28here
06:29of course
06:30these
06:31people
06:32fall
06:33category
06:34fall
06:35which
06:36is
06:37the
06:39category
06:40so
06:41you can
06:42simply
06:43say
06:4495%
06:45people
06:46send
06:47they
06:48don't
06:49send
06:50this
06:51problem
06:52is
06:53transfer
06:54learning
06:55now
06:56transfer
06:57learning
06:58to
06:59why
07:00do
07:01need
07:02to
07:03do
07:04you
07:05need
07:06to
07:07do
07:08the
07:09problem
07:10but
07:11the
07:12problem
07:13is
07:14that
07:15we
07:16need
07:17to
07:18do
07:19to
07:21totally agree
07:22this
07:23means
07:24this
07:25means
07:26this
07:27means
07:28that
07:29we
07:30have
07:31made
07:32a model
07:33image
07:34classifier
07:35to
07:37for example
07:38if
07:39a
07:40bike
07:41differentiate
07:42to
07:43why
07:44why
07:45why
07:46why
07:47why
07:48why
07:49why
07:50why
07:51why
07:52why
07:53why
07:54why
07:55why
07:56why
07:57why
07:58why
07:59why
08:00why
08:01why
08:02why
08:03why
08:04why
08:05why
08:10by
08:11Imagine
08:1220
08:14how
08:17why
08:18why
08:21why
08:22why
08:23what
08:24why
08:25why
08:28why
08:31LET
08:33And when I told you the types of machine learning, I told you that it was Reinforce learning. Reinforce learning of course is applied to games. Reinforce learning is a special thing that you have a set of rules.
08:47And I told you that I don't have data to give you always reward or penalize.
09:01For example, I told you that it's a game. If you're right, if you're right, you're right.
09:11So you're right. If you're wrong, you're wrong. If you're wrong, you're wrong.
09:17So you can see that you've got a minus one score.
09:20Okay. So you can see that you have a step in which I get one.
09:24And if you're wrong, you're right.
09:25So Reinforce learning will increase your steps.
09:29That's how you get one.
09:31So this method of google's alpha go, which I've given you for example.
09:35Chinese game is a world's difficult game.
09:39the world's top 3 player.
09:42Now the game is not like that there is a set of instructions
09:45that you have limited number of moves.
09:49The computer has a lot of memory,
09:53and the best moves you can feed.
09:55The best moves you can choose quickly,
09:57because the computational power is more than human.
09:59However, the Chinese game is not like that.
10:02There are so many moves,
10:04which are in this galaxy.
10:06Can you imagine?
10:08If you win, the intuition is above.
10:11If you win, the alpha has won,
10:13and the top player has won.
10:15How can you win?
10:16Basically, the reinforcement learning is a better example.
10:18Now I am going to wind up.
10:21In this lecture,
10:23there is no machine learning.
10:26For example,
10:28you have input and output,
10:31and output.
10:34I am not sure of the output,
10:49but I am not sure of the output.
11:03But I am not sure of the output,
11:04but I am not sure of the output.
11:06then what do you expect from this case?
11:09So, this case is classically unsupervised learning.
11:15You can use classification clustering,
11:18then you can give data cluster.
11:20You can say,
11:21you can say,
11:22you can say,
11:23you can say,
11:25you can say weight,
11:27you can say height,
11:29you can say height,
11:30you can say height.
11:32And then transfer learning,
11:34and then transfer learning,
11:35you can say that
11:37this is the case.
11:39You can see that
11:41I can solve the model nothing has to do.
11:43But you can now you can use it like this.
11:45So,
11:48you can use it.
11:50Good.
11:52We'll do it.
11:54And then,
11:55we can use it.
11:57We can use it from the memory.
Comments