Don’t want to see a bunch of formula symbols, but want to know how Convolutional Neural Network does image discrimination? Share a video that I explain to my graduate student and answer questions. I hope it will be helpful to you.

At a loss

Friends often ask me which semester my Python and data science courses are in, and they want to come and sit in on them.

Sorry, I don’t have this one.

I wrote a series of data science tutorials. But it was meant to empower my own graduate students, not lecture notes.

Some of them have undergraduate degrees that have nothing to do with technology.

But information science is an interdisciplinary discipline. Especially in recent years, the integration with data science has deepened.

Looking around, other social science disciplines — journalism, psychology, sociology, political science — are using open Internet data sets to do analysis on a scale previously unimaginable.

Under such circumstances, you, a graduate student of information science, have no data science skills in a subject that already has the advantage of data analysis, so you can’t say hello to other students when you go out?

So I wrote them tutorials, as much as liberal arts students could understand them.

As it turns out, they can follow the tutorial and make the results.

However, in my Python programming problem, what do Liberal Arts students Do? “Painting the picture from the gourd” is just the first step in your introduction to data science.

You need to understand the premise and approach of the technology so that you can tackle your own research problems with the appropriate tools.

In this week’s group meeting, I listened to the translation and presentation of first-year graduate papers, and obviously felt that they were still not clear about the structure and principle of convolutional neural network.

I was surprised.

Because I’ve written at least two articles specifically for them on how to use convolutional neural networks for image processing.

What’s more, they redid the training and testing using their own data sets.

In the article, I also introduced the basic principles of the deep learning model, and listed the reference materials in detail at the end of the article for further reading.

After all this time, how come I still don’t know?

In the past, I would have scolded someone.

Because how to look at it, this is not the right attitude to learn the problem.

But with a foundation in empathy training, I suddenly understood their confusion and distress.

In the same way

They see the extended reading material, like a black hole.

The black hole absorbs their time and work, but sees no positive feedback.

Because they lack the foundation.

It doesn’t take much superior intelligence to learn deep neural networks well. But some basics are important. These foundations include:

• programming
• mathematics
• English

If you have these three foundations, you don’t need a mentor at all. Self-study a boutique MOOC like Andrew Ng’s Deep Learning on Coursera, and you’ll grow fast and enjoy it.

However, for Chinese liberal arts students, the three basic requirements listed above can be described as “three mountains”, which can weigh them down.

I didn’t learn programming, I forgot math, and I didn’t pass English.

You tell them to pick it up from scratch and make it all up?

Even if they did, it was time to graduate.

What else do you do?

Admittedly, teachers can help them streamline their learning modules.

Bad programming, that’s okay.

Don’t bother with Tensorflow’s neural network construction details, as long as you can use the simplest TuriCreate to call the transfer learning tool, a few lines of code to do the image recognition.

Not good English, no problem.

I’ll write the tutorial in Chinese for you. You just do it, you get results.

However, if I am not good at math and cannot understand the principle of neural network model, what should I do?

I used to be helpless.

Or you can treat the whole tool as a black box, just knowing the input and output, and you can make a result.

But this is the user’s attitude, not the researcher’s.

This low level of knowledge may give you the opportunity to fully practice what “garbage in, garbage out” means.

Isn’t that how many students who know nothing about statistics use SPSS?

I had an Epiphany when I thought about it.

reference

Statistics is also difficult for many liberal arts students.

How did they learn it?

It’s a limited disassembly.

Only learn to import data, click the button out of the chart, obviously not enough.

But starting with formulas for distributions and explaining how threshold Settings (like the magic 0.7) work… The man was long gone.

How to do?

I remember this book by Professor Li Lianjiang.

Teacher Li’s attitude is that the principle should be made clear, and students should not be allowed to “torture” data.

But they can’t go deep into the basic mathematical principles, so many liberal arts students can’t understand, or even quickly lose interest.

His solution is simple and practical.

Just examples and metaphors.

Using the example of an employee who comes with SPSS, he explains several chapters. From the type of data, all the way to multiple regression.

Because of the actual example, students can fully plug in, it is easy to understand.

So factor analysis, let’s do the rotation. How about this?

He used two metaphors.

One is the Three tenors, representing the three factors.

The concert of the three tenors was packed with people.

Some of the audience came for Domingo, some for Carreras, some for Pavarotti.

But the audience was all sitting together, and you couldn’t tell which audience was a fan of which singer.

How to do?

Let the tenors sing separately and against each other.

This is the second metaphor.

Once there is a counterplay, the audience’s choice of seat clearly represents attitude.

Which factor a certain question belongs to can also be distinguished by factors playing against each other (rotation).

Read “statistics”, I think speak very good.

But THEN I watched The video of Teacher Li’s class, and I learned more.

Because the information dissemination of video is richer.

The same example, because of the graphical interpretation, students can understand more thoroughly and deeply.



In particular, professor Li’s smile always makes people laugh when he talks about the need for some operation methods that are “not foreign” and “have consensus among the social science community”.

Interpretation of the

With the example of Li Lianjiang as a reference, I explained the following contents step by step in blackboard writing during the rest of the group meeting:

• The basic structure of deep neural network;
• Realization of neuron’s computing function;
• How to train deep neural network;
• How to choose the optimal model (hyperparameter adjustment);
• Basic principle of convolutional neural network;
• Implementation of transfer learning;
• Question answered.

I did not pursue the maximization of preciseness, nor did I make more requirements on the universality and practicability of the example. I just made a simplified image recognition model and compared it with the customer churn prediction model from beginning to end.

Again, I used examples and I used metaphors to try to minimize the cognitive load of listening.

In the process, I ask students to ask questions at any time. So the interaction is very close.

And when they’re done, they say, well, I’ve finally got the basics of convolutional neural networks.

Since coach Yan’s workshop recently trained visual recording behavior, I spoke for a few minutes and suddenly realized that this section could be recorded and shared with more people.

So I asked Yang Wen, sitting in the front row, to help me record the video.

The first few minutes, the first two sections of the list, are not included in the video. Rather a pity.

But it doesn’t matter, after a period of time, WHEN I prepare for the group meeting, I ask graduate students to come to the platform and repeat this paragraph as a learning effect check.

If they do well, I will record them and share them.

They don’t know my plan yet.

So when you see it, don’t tell them. SHH!

At nearly 30 minutes, the video is not short.

If you’re like them, you’ve read my book how to Lock In Lost Customers with Python and Deep Neural Networks. , How to Recognize Images with Python and Deep Neural Networks And How to Find Approximate Images with Python and Deep Neural Networks. These several articles, but for the deep neural network principle construction is still confused, suggest you read from beginning to end, there may be some harvest.

Please click on this link and watch the video.

There is a word first, because it is impromptu explanation, without any preparation. Omissions are unavoidable.

Welcome to help point out the flaws, I will explain in the future, iterative improvement.

Thanks in advance!

discuss

How did you learn how convolutional neural networks work? Is there a better way for you to understand and master the mathematical formulas encountered in your research? Welcome to leave a message, share your experience and thinking to everyone, we exchange and discuss together.

If you like, please give it a thumbs up. You can also follow and top my official account “Nkwangshuyi” on wechat.

If you’re interested in data science, check out my series of tutorial index posts entitled how to Get started in Data Science Effectively. There are more interesting problems and solutions.