ml4a / ml4a.github.io

machine learning for artists

ml4a.github.io

Machine learning for artists

notes, demos, interactive visuals, art projects, references, further readings, etc.

more documentation to be added soon.

ml4a.github.io's Issues

10^80 atoms is a big lie

https://ml4a.github.io/ml4a/how_neural_networks_are_trained/

What 10^80 atoms in the universe you are talking about?

There is 5 x 10^21 atoms in a drop o water already. How many atoms would there be in Earth? The Earth is 10^6 times smaller than the Sun! How many Suns are there in the Universe? How many planets? 10^80 atoms in the Universe ? Ahahahha! I can only laugh about this. And what if the Dark Matter has atoms too? There is no way the Universe is so small that it only has 10^80 atoms!

Brute force and random search will be the algorithms of the future when we will make our computing devices with some other type of technology than moving the electrons. Nanotechnology will help. And then, the computers will be so cheap the we will brute force everything.

404 Error for video

https://ml4a.github.io/ml4a/machine-learning throws two 404 errors for missing videos (https://ml4a.github.io/images/video.mp4 and https://ml4a.github.io/images/video.webm). This video is the one to illustrate linear classifiers in 3 dimensions (I'm guessing)

redundant 'left' in chapter 4

@genekogan, I found redundant 'left' in this sentence (probably the second one can be omitted):
https://github.com/ml4a/ml4a.github.io/blame/master/_chapters/how_neural_networks_are_trained.md#L169

... and so solving the above linear regression will be left as an exercise left to the reader.

(sorry for pointing out small things, but when I translate I find these things)

multilingual javascript

from #37 (comment)

Currently the demo javascripts are hardcoded with english texts but they should be translated too. I tested i18next and it seems working well, which can be found in my branch:
master...micuat:js-multilng

translate _chapters_ko/how_neural_networks_are_trained.md

multilingual table of contents

We are missing toc in other languages - and I heard this will be useful for non English speakers. Should we translate this page
http://ml4a.github.io/ml4a/
to URLs like
http://ml4a.github.io/ml4a/cn/
http://ml4a.github.io/ml4a/es/
...

well, perhaps we should translate the front page as well.

Broken link in README

Apparently the README link should point to http://ml4a.github.io/

more fine-grained control over mathjax

right now, mathjax is externally hosted via CDN. hosting it locally makes it easier to work offline. additionally, it's not clear how to change some of the style parameters via CSS, like the size equations and figures should appear at. centralizing this in the main.css would be helpful for styling.

chapter 5 typo

missing closing )
https://github.com/ml4a/ml4a.github.io/blame/master/_chapters/convnets.md#L217

neural nets chapter needs regression example

the section in the neural nets chapter which says "Random set of dat, 3 cols. 1 regression value" needs a good example.

this would just describe what regression is and how it works. no code included -- it would not need to say much about how the solution is found (except maybe to link to the "training neural nets" chapter). it's a "magic trick" that shows what happens if a neural net has the correct weights for regression.

this leads into the classification example which is more complete already.

Typo in convnets chapter

"are simply the the image classes"

https://github.com/ml4a/ml4a.github.io/blame/master/_chapters/convnets.md#L115

term 'sample' in chapter 4 may be confusing

Hi @genekogan, in this line of chapter 4:
https://github.com/ml4a/ml4a.github.io/blame/master/_chapters/how_neural_networks_are_trained.md#L62
(sorry for using blame mode, it's just because github does not allow users to point to a line of rendered markdown)

This principle is closely related to what we call in machine learning “the curse of dimensionality.” Each dimension we add into a search space exponentially blows up the number of samples we require to get good generalization for any model learned from it. The curse of dimensionality is more often applied to datasets; simply put, the more columns or variables a dataset is represented with, the exponentially more samples in that dataset we need to understand it. In our case, we are thinking about the weights rather than the inputs, but the principle remains the same; high-dimensional space is enormous!

the first 'samples' refers to the sampling of the space, and the second one refers to elements of the dataset (if I understand correctly). Since in this chapter the former usage is more common, I suggest that the latter should be replaced by 'data' / 'element' etc to avoid confusion.

translate _chapters_ko/convnets.md

Just keep going.

demo images broken / where should we host them?

all the demos are broken because i removed the MNIST/CIFAR images from the repo to make it smaller. CIFAR alone is around 120MB. would it make sense to host the images somewhere else to keep the repo smaller (and likewise with other large content)? if so, where's a reliable place to do so? if we do that we save the space but the demos can't run on localhost without an internet connection.

create discussion forum/comment sections per page?

it could be useful to have space for people to ask for clarifications and have other discussions; probably better to have it per page, but not sure yet.

translator credits

@TiborUdvari @weakish @irealva @tornoteli @rickiepark

hi ml4a translators! right now, your names do not appear anywhere except github contributors. in general, none of the pages, including the book, guides, or demos have names right now. i'm trying to figure out a good way to list contributors. do you have any comments about how you might like to be credited? one option would be to put it at the top of the actual chapters (example: "translated by __" or "translated by __ and __") or we can possibly make a new page which lists the translations and the authors next to them. do you have any preferences?

What does the translation workflow look like?

Hey @genekogan 👋

After working with this, I thought it would be useful to add german translations since they are still missing. Is there some kind of workflow on how to do this?

I thought of throwing the chapters into https://www.deepl.com/translator and then proof-read the result. I mean… the translation for this should be done with the help of machine learning, right? 😄

working '_chapters_ko/neural_networks.md'

I started to translate _chapters_ko/neural_networks.md
If somebody already work in _chapters_ko folder, let me know about it.
Thanks. :)

Started to translate looking_inside_neural_nets to zh_CN

I wish I finish it in tomorrow. ;-)

translate _chapters_jp/neural_networks.md

working on it here:
https://github.com/micuat/ml4a.github.io
http://naotohieda.com/ml4a.github.io/ml4a/jp/neural_networks/

• preface
• Birth and rebirth of neural nets
• From linear classifiers to neurons
• Activation function
• Layers
• Regression
• More layers, more expressiveness
• Classification
• Classification of handwritten digits
• Further reading
• Next Chapter

No issue but does any one know if any east coast companies/startups are doing ML/Art?

I am a 28 years old artist, self taught myself ML for a year made a lot of progress, but then I got stuck. I decided this is my passion so I went back to school to study CS in a more formal way. I am about to finish my first year in college and I don't feel like I am being challenged so I am looking for internship opportunities. The problem is I don't know if any well known companies or start ups are doing 2d/3d animation + Machine Learning/Deep Learning on the east coast. Quick google search shows none. I know this is probably not the kind of post I should make for issues but I would like to try my luck. Does any one know any ML/Art internship opportunities?

translate _chapters_ko/looking_inside_neural_nets.md