Message from Why Tea in MacGuyver - Skills & Academics #stem

2017-12-19 15:49:46 UTC  

Oh, I forgot about this since I have the whole server muted. So yes, relativity just says that objects moving through space can't go faster than light, space itself expanding isn't really covered by that, since it's not moving through anything. I haven't really studied GR so I don't know all about that though.

2017-12-19 15:57:11 UTC  

"Well, as I understand it, the modern view of light isn't so much that its a particle/wave as it is something that exists in spacetime itself."
Modern physics doesn't need to make a hard distinction between 'objects' and light, the same set of equations governs them. They both exist in spacetime, it's where everything exists, and they both have wave particle duality. Electrons have wave properties too, if you send them though a narrow opening, they behave in a way that a pure particle never could.

2017-12-19 15:57:31 UTC  

is there any way to get notifications from just one channel but not the others in a server?

2017-12-19 16:14:06 UTC  

I think you can mute channels.

2017-12-20 07:00:37 UTC  
2017-12-24 01:00:01 UTC New discord dedicated to sci + tech (and culture) with an emphasis on futurology and emerging tech (note: this server is not identitarian exclusive)

2017-12-24 09:30:20 UTC  

^ To be clear, this is not a replacement of this channel or server

2017-12-24 20:09:56 UTC  

Anyone into machine learning, deep learning, etc? @here

2017-12-24 20:14:10 UTC  

What’s that?

2017-12-24 20:56:26 UTC  

AI, I think.

2017-12-25 00:10:01 UTC  

@Why Tea I do some at work and did some for my degree

2017-12-25 00:10:29 UTC  

Do you ? @Why Tea

2017-12-25 00:11:52 UTC  

Cool. I've been reading some on coding it. My employer (big big company) is a huge player in it, but I'm not involved in it at all at work.

2017-12-25 00:12:44 UTC  

Is this a good place to start (sure looks like it) and what other resources do you think are worth looking at?

2017-12-25 00:14:01 UTC  

@Deleted User Yes, one way / model of dealing with AI.

2017-12-25 00:28:44 UTC  

@Why Tea what resource to use and what to focus on depends a lot on what you're actually working on.

Are you doing some sort of image or signal processing? If so then yes Neural Networks sounds good and that site is a good start.

But Neural Networks aren't useful for every problem (though some academics are trying to push the idea that they are) so step one I would ask are you absolutely sure they are necessary for your work.

You can tell what machine learning things to focus on by laying out what type of inputs and outputs you are doing. The dimensions your data can fit are usually sequential vs non-sequential (i.e. a sentence is a sequence of words, an image is just a single vector of numbers) and categorical/symbolic vs cardinal vs ordinal.

2017-12-25 00:31:22 UTC  

Right, right. Thanks. But I'm not working on anything *specific* yet, as I'm trying to learn.

2017-12-25 00:31:47 UTC  

Of course one of the best ways to learn is to pick a thing you'd like to do, and try to solve the problem, but I'm not there yet.

2017-12-25 00:41:53 UTC  
2017-12-25 00:42:02 UTC  

Excellent, thanks!

2017-12-25 00:42:13 UTC  

I love the MIT courses that are available.

2017-12-25 00:42:24 UTC  

MIT open courseware on Artificial Intelligence

2017-12-25 00:44:10 UTC  

You're welcome, yeah they are great

2017-12-25 01:12:34 UTC  

I’ve also heard good things about the Coursera online courses for this.

2017-12-25 01:13:29 UTC  

I took a Coursera course in Data Analytics learning R and they did a better job than Texas A&M’s graduate class on the same topic (I took both).

2017-12-25 01:17:31 UTC  

Yeah Coursera's first course was a pretty good Machine Learning Course by Andrew Ng, but last year they changed the format of the Coursera website SO MUCH that I now find it virtually unusable. That's partly by design, they give very little away for free anymore and they tightly restrict when materials are available.

2017-12-26 03:03:23 UTC  

So just pay for it.

2017-12-26 03:45:00 UTC  

That probably works for some people, but ever since after I finished school I usually study a subject in spurts in a way that doesn't fit well with Coursera's pricing model. I say this as someone who has paid for their courses and generally didn't get my money's worth before they expired.

2017-12-26 03:46:10 UTC  

The best experience I had with a paid Coursera course was when I was doing a programming course with other people. I feel like that model works best for Coursera because that's closest to the classroom model they are going for

2017-12-26 03:48:20 UTC  

When I study on my own though, I like to dive deep into something in a few goes when all of a sudden a few spare hours open up. Coursera is not conducive to this though. Courses start periodically so if it doesn't happen to be the first week of their course then you are SOL. And then the longer courses they have roll out the course material one week at a time, preventing a dive in.

2017-12-26 03:49:10 UTC  

That's why I like to just have all the course materials available at once, all the time. I don't think Coursera has that model for any of the courses I was interested in though.

2017-12-29 00:02:02 UTC  

I do some machine learning light for work and have a masters in data science from Berkeley

2017-12-29 00:02:10 UTC  

I can answer basic questions

2017-12-29 00:02:27 UTC  
2017-12-29 00:03:28 UTC  

It's a lot of hype concealing a clever compilation of basic techniques but a very powerful concept once understood.

2017-12-29 00:05:22 UTC  

It's most useful applications involve large datasets. Currently there are innumerable opportunities for application. The limiting factor is usually expertise and leadership buy in.

2017-12-29 00:06:20 UTC  

Even still one of the most widely applied machine learning techniques is linear regression.

2017-12-29 00:07:04 UTC  

This fact reveals the simplicity at the core of the concept.

2017-12-29 00:08:18 UTC  

Its a beautiful simplicity though and there is a growing library of well documented algorithms for various applications. Even still > 80% of the work is just wrangling the data and pre processing.

2017-12-29 00:12:07 UTC  

If you want to learn check out look at winning solutions for problems that interest you.