A Neural Algorithm of Artistic Style

sliders to regions of the latent space; quite an obvious thing to do, if you want to try getting fine-grained control over the output. (A 2015 paper built on this technique: https://arxiv.org/abs/1508.06576) I was using spaCy back in 2017–2018-time. It represents sentences as vectors, that you can do stuff like cosine similarity on. If I'd been more interested in the field back then, I could have put two and two together, and realised you could train a net on labelled images (with supervised learning) to map a spaCy model's space to StyleGAN's, which would be a text-to-image model. It was very much imaginable back before April of 2020; a wealthy non-researcher hobbyist could've made one, using off-the-shelf tools. If I were better at literature searches, I could probably find you an example of someone who'd done that, or something like it! --- ¹: See e.g. here: they tell me GPT-4 can translate more than just explicitly-specified meaning, and the "evidence" doesn't even manage that. https://news.ycombinator.com/item?id=35530316 (They also think translating the title of a game is the same as translating the game, for some reason; that confusion was probably my fault.)

Copilot or Codewhisper producing code, let's think of examples like some of recent advances using Convolutional Neural Networks [2] What will the Music Industry or Cinema Industry do, if a company uses the Corpus of Music or Movies without asking for further permission? For example, to produce a commercial product, that would allow a user, to create music on the style of Pat Metheny, AC/DC, Abba, or a Cinematic experience on the style of Lord of the Rings? Surely both the user and the product vendor would be called upon it? [1] "The Most Important Court Decision For Data Science and Machine Learning": https://towardsdatascience.com/the-most-important-supreme-co... [2] https://arxiv.org/pdf/1508.06576v1.pdf

translations, time/cpu to produce a good model, and I think you would be astonished by the results. Data's complexity is of little hindrance here, because the algorithm does not try to understands the data (as we humans like to think we're doing when reading), it's just trying to infer a solution based on learned behaviors. And it's 'just' ML, AI will come when your computer tell you that this specific text makes it sad, for example.

whenever you ask a grad student to run some numerical simulation expect them to come back saying "Hey I found a closed-form expression!" - The reason the above is absurd is that these trained NN's aren't really solutions to the optimization problem, but approximations. So this is really saying I have a problem, I don't know how to solve it but I can produce a infinite sequence of approximations. Now I'm gonna truncate this sequence of approximations, and call this a closed form solution. The analogy in highschool math would be computing an infinite sum that doesn't converge, but now let's instead just add to some large N, and call this a closed-form solution. [1] e.g. https://arxiv.org/pdf/1508.06576.pdf

Faster R-CNN - have hardly any maths in. You can count the equations on one hand in both those papers. No doubt the authors know their stuff, but how readable are those papers compared to e.g. a 100-page proof? Which did I learn more from? They're a combination of two things: intuitive network architecture and a clever loss function. The first thing is a combination of intuition and programming, the second involves a little maths, but nothing outrageous. [1] https://arxiv.org/abs/1508.06576 [2] https://arxiv.org/abs/1506.01497

in derived work.. should it be "okay" for people to publish to arxiv and just.. leave it there and not put it in a conference or journal? Should such work be validated by citation? It's a legitimate question that I don't know the answer to. If you get an idea from there, you can't just.. not cite it.. and yet, I feel like arxiv should be used only as long as the work will eventually get a proper publication. Which more and more it seems is not a given. I'm really just responding to the statement on the posted link, which is just a header, "Citation", and an Arxiv-bibtex entry. No "submitted to X..", "in press", etc., or anything. Again, it's not that I'm totally against the idea, but I think there must be some happy medium between "peer reviewed work" and "well-written but unreviewed article". Typically this used to be conferences, but conferences are expensive, and with people uploading their pre-conference unaccepted publications and those getting citations, I mean.. where will this end? I was a bit shocked recently while writing an ML article to find that the "official" original reference, that you see cited everywhere, for the whole concept of "style transfer" appears to be an arxiv paper [1]. My reservation doesn't so much come from the fact that the authors put their work there, but more that this is what people are actively choosing to cite over their peer-reviewed conference publication [2]. When does "pre-print" become just "print"? (Currently Google Scholar reports 216 citations for the former and 74 citations for the latter. Of course they were published a few months apart so it's not a great comparison but still, just an example..) [1]: https://arxiv.org/abs/1508.06576 [2]: http://www.cv-foundation.org/openaccess/content_cvpr_2016/ht...

was to test whether evaluating Deep Learning networks in WebGL is doable. To our knowledge(but we may be wrong!), our implementation is the first implementation ever to attempt GPU accelerating neural networks with WebGL And we hope that this implementation will provide a foundation for people who, like us, wish to experiment with Deep Learning and WebGL The GPU implementation can be found here(https://github.com/Erkaman/regl-cnn/blob/gh-pages/src/gpu.js) Note that this network will probably be slower than the corresponding network implemented on the CPU. This is because of the overhead associated with transferring data to and from the GPU. But in the future we will attempt implementing more complex networks in the browser, such as Neural Style(https://arxiv.org/pdf/1508.06576v2.pdf), and then we think that we will see a significant speedup compared to the CPU. Lastly, if anyone has any questions, I will be glad to answer them here.

was to test whether evaluating Deep Learning networks in WebGL is doable. To our knowledge(but we may be wrong!), our implementation is the first implementation ever to attempt GPU accelerating neural networks with WebGL And we hope that this implementation will provide a foundation for people who, like us, wish to experiment with Deep Learning and WebGL The GPU implementation can be found here(https://github.com/Erkaman/regl-cnn/blob/gh-pages/src/gpu.js) Note that this network will probably be slower than the corresponding network implemented on the CPU. This is because of the overhead associated with transferring data to and from the GPU. But in the future we will attempt implementing more complex networks in the browser, such as Neural Style(https://arxiv.org/pdf/1508.06576v2.pdf), and then we think that we will see a significant speedup compared to the CPU. Lastly, if anyone has any questions, I will be glad to answer them here.