WildML

I'd suggest searching for and reading blog posts of people who are studying deep learning. Academic papers are very pretentious, they are the tools for scholars to declare victories. They are not the best materials to understand. http://www.wildml.com/ https://medium.com/mlreview https://distill.pub/

Probably because the author followed this blog: http://www.wildml.com/2015/12/implementing-a-cnn-for-text-classificati… That blog used a 2d cnn because tensorflow didn't have a 1d version at the time of writing, so he just created a dummy 2nd dimension of length 1 and called it a day.

I have a pile of links for getting started with DL in my comment history you can use: https://news.ycombinator.com/item?id=10676455 What really helped advance my understanding from zero to knowledgeable novice was rewriting some existing code line by line (using expanded variable names and comments), and thinking about each line and what it does as you go. It's the software development equivalent of Hunter S. Thompson re-typing The Great Gatsby just to get the feel of writing a great novel. Here's one I did based on Denny Britz's tutorial: Britz's Original: http://www.wildml.com/2015/09/implementing-a-neural-network-from-scrat… My version: https://gist.github.com/sthware/c47824c116e6a61a56d9 HTH

You're right that JS is pretty ill-suited for implementing production-level Deep Learning architectures that operate on a lot of data, I think it's still well suited for explaining the concepts in a clear way. A lot of the models have intuitive (and visual) interpretations, something that can be nicely visualized with JS. As for Karpathy's implementation, you're right, he chose JS for that reason. His library is meant as a tool for experimenting and teaching, not to deploy these models in production. Op: Send me an email if you're interested in this (email in profile). I actually started something similar just recently at http://www.wildml.com/.

On the new age tech, these may help: https://hadoopweekly.com/ http://www.wildml.com/newsletter/ https://thomaswdinsmore.com/ http://www.dbms2.com/

Encoder-Decoder models are a very common technique in sequence-to-sequence models for deep learning. These models have had very big wins lately in NLP tasks such as translation, POS tagging, dialogue generation, etc. The Tensorflow documentation has an okay writeup about seq2seq models: https://www.tensorflow.org/tutorials/seq2seq The author of the library also has a small blurb about it on his blog: http://www.wildml.com/deep-learning-glossary/#seq2seq

You may wish to coordinate a bit with this?: http://www.wildml.com/deep-learning-glossary/ I came across that glossary a few days ago when looking to solve an RNN problem with a deadline. You may be pleased to know I think I may have solved it by switching to Keras, and this post actually helped unsticking me from a smaller problem too. Thanks, and congratulations on version 1.0!

It might look something like attention detailed in Show, Attend, Tell: http://arxiv.org/abs/1502.03044 Which attempts to visualize machine areas of attention that look like: http://www.wildml.com/wp-content/uploads/2015/12/Screen-Shot-2015-12-3…

There are two ways to consider this question, hacking as in "Mr. Robot" or hacking as in writing code for the joy of creation and discovery. Given the reference to 'writing software for corporations' I'll consider it the latter. It is profoundly important for engineers to periodically rekindle their sense of wonder if they wish to stay relevant and it sounds like what you are asking, if so that is super awesome. Unfortunately for different people it is different things. So you might have to flail around and try a few things first before you get where you want to be. The good news is that it has never been as easy to try different computing things as it is today. One piece of advice though is to be mindful of tooling. Which is to say think about the tools you use and if they help or hinder your wonder. Nothing puts the kabosh on wonder like slogging through a click and type interface for an hour for each attempt at making progress. So think about tools as well as what you're using those tools for. Perhaps take a week to try different ways of editing source code for example to find one that is the most frictionless. Finally, before we get to some suggestions, YOU MUST change how you 'score' your progress internally. On thing I have seen again and again is people who have been coding professionally for a while their 'score' is "projects completed in unit time" the bigger the number the better they feel, that was how their boss scored them and they adopted it for their own…

Agh, I remember there was this little thing that let you interactively play with neuron width and add and remove nodes, but I can't find it. Here's what else is in my bookmarks, aka what I've seen go past on here over the past few months that seemed approachably interesting: http://www.wildml.com/2015/09/implementing-a-neural-network-from-scrat… - a getting-started guide you can follow along with in IPython https://codewords.recurse.com/issues/five/why-do-neural-networks-think… - just playing around, but notable https://github.com/mbartoli/neural-animation - shows how to restyle video in manner loosely resemblant of Deep Dream http://neuralnetworksanddeeplearning.com/ - this book is online-only but it's CC so you could legally crawl it if you wanted a local copy, I haven't rated it for my own use yet http://nxxcxx.github.io/Neural-Network/ - a static 3D visualization you can zoom around, mostly kitsch but still kinda fun. I recommend you go through the cream of the crop and find what suits you the most though: https://hn.algolia.com/?query=neural&sort=byPopularity&prefix&…