Name: Sentdex Channel in 2020
Uploaded: 2020-03-31T18:21:04.000Z
Duration: 10 min 14 s
Description: Thousands of YouTube videos with English-Chinese subtitles! Now you can learn to understand native speakers, expand your vocabulary, and improve your pronunciation...

Present continuous

The imperative

And welcome to a stick of the Centex Channel for 2020 address.

So we start off 2020 with, As most of you are probably aware, the neural networks from scratch announcement slash Kickstarter, which I am happy to say, has been going better than I honestly could have expected.

We've got hundreds of people in the draft document.

Um, it's going really well, thank you to everybody who supported that.

So moving forward, we didn't get a couple of pretty common questions on that announcement.

So one is people asking, Will you cover confidence and recurrent nets?

And so the initial plan was just neural network from scratch.

There's so much information to cover just on that topic, too.

So the plan was not to do confidence in the book, necessarily, but it is possible.

And so I thought, maybe what we could do is like a stretch goal for the Kickstarter.

But honestly, it's not a function of money.

Um, and by I I mean we which I'll talk about in a moment.

But maybe we can add it to like the e book version or something like that later on.

The other question I've got was, when will we begin the releasing of the actual videos?

For that Siri's, I'd still don't totally know.

But I'm going to guess somewhere in the next 1 to 3 months will start the release, but that's going to be a very long running series.

The series is likely thio Match that I don't know.

I I also haven't fully decided how how much, um, you know how brief to be in certain things and how much I should leave the book to be the more in depth version or not.

So I'm still kind of trying to figure out what I want the balance to be there, but yes Oh, probably one of three months.

The reason for that is I want to get the entire draft the first draft.

Dunn's all the code for that Siri's is done.

So I want to make sure the way I'm explaining things is good for people.

So I want to know where are the common pain points So we can fix that before releasing the videos and, of course, publishing the book.

So over the next few months, the main stuff is going to be neural networks from scratch.

The Kickstarter campaign still has, like, 20 days to go.

I think if you do want to support you goto n n f s dot I oh, uh, and support the book or by the book.

For some reason, um, 2020 is going to be mainly a I focused.

It won't just be neural networks from scratch.

It can't possibly just make that the only thing we talk about on this channel, eh?

The next most likely video you'll see on the channel will be facial recognition, but there's many other things I'd liketo do.

So the main kind of deep Siri's that we're gonna be doing first will be neural network from scratch.

And then the next topic is most likely to be natural language stuff.

I've been getting lots of requests to do natural language content.

Um, as many of you guys know, that's actually how I got my start with programming.

In general, this was long before there was, like deep learning but doing sentiment analysis, using machine learning to do to predict stock prices and stuff.

That's that's what got me into programming in general.

What I'd like to do is continue the chat by the text to speech and the speech to text.

If we combine all three of those things, something really cool can happen.

So So I know that's what I'm interested in doing, and I know that that's what Daniel is also interested in doing.

Daniel has been working with me now full time for the last year.

He's actually been with me since before that, but it wasn't full time.

Um, so many of you guys know who Daniel is?

He's danos in the discord, but yes, he's been working with me now for, um, over a year and basically What he's able to do is, is kind of he's able to go off and do like R and D for things while I'm doing something else.

Toe have that because so much of my time goes into just the R and D for something, and often it's the case that it's just for whatever reason, it needs to be scrapped or can't really do it on the channel for because maybe it'll be too long to advance.

We're missing a lot of background information that would just make it too crazy to do stuff like that.

So while I was doing the quantum computer programming serious, he was doing the research for Can we get neural networks from scratch?

And actually, it was starting to look like man, this is super long.

So, um and then that's kind of where the idea of like, Well, what if we make this into more like a book?

Basically, at this point, uh, and then do the videos and code for free, but then sell like the hard copy of the book, especially because for topics like this, you really do wanna have, like, you're gonna have multiple sittings.

It's useful to have a hard copy and then also the draft access, which I to me that's the most valuable part being able to, like, highlight where you're confused and stuff.

So anyway, that's been kind of cool to see how that's going.

Eso far that's been working seemingly quite well.

Some people were asking, Hey, why don't you just put it up on get Hub and then share that with collaborators on get, huh?

But really, the valuable part is to be ableto directly in line.

Highlight and post comments or even some people have been posting grammar fixes and syntax issues and stuff like that.

And so I don't know that there's any better way than good.

So that's kind of what we're using now moving forward.

What I'd like to do is continue that, but then make our own kind of version of that on and host that upon neural networks from scratcher and NFs dot io.

Um, and I think that would be a good way of moving forward But for now, Google docks.

It is so aside from Daniel, uh, also have a new machine over here called a box with two exes.

Uh, this is from a company called Box at Based out of Austin, Texas.

They sent that to me on that one is mine to keep.

It's very similar to the data science workstation that we had before, only slightly better in a few ways and actually cheaper in price.

You'll be seeing more of that at some point.

Once I've had it for a little bit, I'll do a full kind of review off the machine.

The main takeaway is 96 Gigs of RAM, 256 gigs of Ram Um, and 48 threads off CPU Power.

I'm very excited to have that, and what that means for you is I can go through much quicker rnd times for things.

So, for example, one of the main reasons the GT a Siri's ground to a halt was every model took like, a week or more to train, and that didn't mean it was gonna be better than the previous one.

It was a test right, and most of the time the tests are not better.

So just it's I stopped being able to justify the GPU time for that.

I needed to spend time doing other things that were more productive.

First of all, it's it's like 10 times more powerful than what I was doing that GT a series on.

So so that alone is good, but I can also be training multiple models for multiple projects at the same time, or multiple models for the same project at the same time, whichever you want, which, obviously, for any norm network, that's great.

Also, for, like, reinforcement learning we can we can get those things done much quicker, so super excited for what that is alone gonna bring to the channel.