Name: 機械学習を学ぶための最高のリソース？そもそもMLって何？MLエンジニアとのQ&A (Best Resources for Learning Machine Learning? What Is ML Anyway? Q&A with ML Engineers!)
Uploaded: 2021-01-14T10:24:17.000Z
Duration: 12 min 1 s
Description: VoiceTubeの動画で発音を聞きながら英語表現を覚えよう！学べる英語：

過去形

So the other day I hosted a meet up right here in Toronto.

On in this meet up, we had a Q and A system with machine learning.

It's a company that implements machine lining solutions for other companies, including a bank on.

I thought this Q and A session was really informative, but it was pretty long.

On the audio we got from the event wasn't the best, I would say.

So I decided to give you a summarized version of the Q and A session in this video.

But just in case you wanna watch the entire raw footage of this cure me, I'm gonna put a link to the in the description on If you want to sign up for future Meetups in Toronto and maybe elsewhere, I'm gonna put a link to sign up for that into this kid from below two.

And a session was with machine learning engineers Joe, Mark, Helen Pippen, Danny and such on.

The first question I asked was, What is machine learning exactly?

This is actually something I already talked about on this channel My video about what can you do with Python?

But let me summarize their explanation here.

Eso Machine learning is used when you can't explicitly right on the algorithm to self a given problem.

An example of a problem like this is when you want to make a computer categorize images s.

So let's say that you want your computer to understand that this picture contains a dog on this other picture contains a car.

You can solve this problem with machine money by giving a machine learning algorithm a bunch of labeled examples.

And then are your machine learning algorithm, which might be, for example, on your network will learn to label new images into either dogs or course.

So that's basically what machine learning does.

It's a way for a computer to learn panners from a bunch of examples.

Anyway, after that, I asked them if they could share some interesting applications of machine learning.

One application Helen right there talked about is how some people have been trying thio use machine learning to generate things like art, poetry and music s.

So I asked them how you would actually go about doing that and then Joe right there.

Explain that for the example of poetry, you could just take a bunch of points you like and train what's called a language model on those poems.

And then that language model will basically learn these, that the score regularities in those poems and it will learn to generate similar new poems similar to the given ones.

After that, Pippen right there talked about how he applied machine learning to astronomy because that's what he is really interested in.

He was basically trying to help astronomers find more supernova in the space.

You know, astronomers need to We got a lot of images to find supernova, and he saw that this is basically just a classification problem.

Eso In this problem, you just need to classify each image into either containing a supernova or not containing a supernova.

And he told us that he was able to apply this relatively new technology called convolution a ll your networks to solve this problem on this approach was actually better than the state of art method.

Mark right there shared with us was how they helped the bank with their credit card customers.

They basically built a model that would predict who's payments would be delayed based on the past customer information.

And they also built a model to predict if each customer would respond to a phone call to remind them to pay.

So where you would actually make a pretty good guess?

That model bank particular was our flagship project.

Some of you tomorrow, people, people in this room who have been processed, okay.

And after that, I asked them, What are the best resource is for learning machine learning?

I I It's supposed to be a more of, ah, hands on practical course In particular, I recommended to this audience because it targets people who are already coders on, and it's very well done.

It was developed by a guy named Jeremy Howard and also Rachel Thomas are both very interesting and very big teachers.

And then Mark here recommended the machine learning course on course era by on Junik from Stuff Word.

It's also a really popular course on apparently, it's more theoretical and more mathematical than Fast II.

Another bookmark recommended is this online, freely available book called Neural Networks and Deep Learning by physicists Michael Nielsen.

Okay, so those are some of the machine learning learning resource is they recommend it, but such in your right here took a slightly different approach for him.

He got started by learning So the six, because machine learning is partially based on statistics and he didn't recommend any particular resource is for that before steps.

I would personally recommend getting started with Kana Kami and course ERA and just go from there.

Uh, you know, I would say the same thing for linear algebra two anyway, After that, Joe added.

This this kind of two paths that people part argue for there's like a top down approach to learning in the bottom up approach.

Top down approach is start with practical things.

First start with how to do it and then get into the nitty gritty of like what is the underlying mathematics and the bottom up approach is start with the underlying mathematics and work your way to the practical stuff.

Uh, if I had to guess, I would say top down approach will work better on average for this audience.

Uh, but you have to try for yourself and see, See what works Had top down approach being fast.

Thought I fast I die Bottom up approach being like learning statistics on DDE reading Deep Learning Book by Ian Goodfellow or something like that.

Okay, so one thing they all recommend it is working on a machine learning project so you can learn from the experience, eh?

So I asked them, How should people find a good machine learning project to work on something that that is all the motivation?

So that's Helen's advice on I asked them How should people go about finding good deficits?

They recommended a few resources for that, Like Google's data sets on after the event.

They also sent me an article that lists good that assets for machine learning.

So I'm gonna put a link to that in the description below two.

But after that, Marco So said this for many projects you'll find there isn't a nice curate they've set.

If there's like something you want to do our bill, you'll have to come up with creative ways of building yourself, whether it's scraping, scraping and then actually, a few of them also recommended this website called Chicago, which is a machine learning competition websites.

First, I found the helpful for me to get started, an initiative that also has been a few bones data science around Toronto US.

And these places are good or talking to other people should feel talking to leaders, feel it points the other machine learning engineers and scientists.

And I say eventually, Chicago is excellent.

I think I have huge skill that machine there engineers, if you want to go into machine learning industry, have this working with real life.

If you're interested in actually where I feel like a lot of non profits, actually people to help with the data's times because we don't have in house science, So a good way to go about getting your hands on a real date, actually, as well it's hitting Thio Good is to help with that.

So there's one more last night that it's just a lot of big years, and then you would eventually.

After those questions, I let the audience ask questions.

What are the questions we got was Why did you choose Machine?

Learning out of everything on here was Danny's answer.

You know, what's something guys that I'm good at and that could be good at it and then use my school's well, and that's something I enjoy.

I am pretty like machine learning because and data science.

And when I got into it, I found that it was using a lot of what I read Renew from physics.

No, it's about explaining the world in novel ways.

It's playing physical phenomena through mathematics, always and the power is your power machine.

Learning that blew away the traditional predicted classical analytics Really true.

There's a lot of promise in future, more so than loved.