Who are you, and what do you do?
I'm Ross Goodwin, and I'm a creative technologist, artist, hacker, data scientist, and recent graduate of NYU ITP. I am passionate about developing new forms and interfaces for written language, enabled by technology, and machine intelligence in particular. I've written two (rather long) essays about my recent experiments with writing and machine learning, which can be found here: I, II.
Most recently, I made a film with Oscar Sharp called Sunspring. It is, so far as we know, the first film ever created from a computer generated screenplay. I generated the screenplay by training an LSTM recurrent neural network on hundreds of science fiction screenplays, until it could adequately recognize patterns of individual characters to write dialogue and action descriptions on its own.
What hardware do you use?
I do the vast majority of my work on a 13" MacBook Pro (c. mid 2014). However, the cloud resources I leverage are another story entirely. At any given time, I'm running four to six Amazon Web Services instances to manage web scraping jobs, Twitter bots, web application projects, and other tasks and projects best suited for remote execution.
My primary computational workhorse is the NYU High Performance Computing facility, where I have access to 32 NVIDIA Tesla K80 GPUs, each with 24GB of GPU memory, along with large numbers of CPU cores and significant RAM allotments. I can simply SSH into the login node for this facility, and I have all these resources available from my laptop.
For portable applications, I've done a lot of work with various Raspberry Pi and Arduino models. More recently, I've been using the NVIDIA Jetson platform, which is like a Raspberry Pi with a GPU, for a variety of projects including generating the final screenplay for Sunspring as well as my Camera, Compass, Clock project.
Another piece of hardware I've been using is the Datamax O'Neil microFlash 4te thermal printer. I currently own three of them, and they're great for providing a tactile dimension to projects that might otherwise live entirely on a computer screen. They print much wider (over 4") than other thermal printers I've used, and they have built-in batteries, which come in handy for portable applications. While these printers are fairly expensive when purchased new, the manufacturer has been making the same printer model for over 10 years, and used ones can be picked up for quite a bit less than retail on eBay.
And what software?
Python was my first language, and I remain a devoted Python programmer. To me, it's still the most elegant and poetic programming language I've found, and the functionality of the standard library is incredible. Beyond that, my favorite Natural Language Processing library is Pattern, and I'm in the process of learning TensorFlow for machine learning.
In JavaScript, I'm a big fan of p5.js, which is developed by a lot of my friends from ITP, and was created by my former ITP professor Lauren McCarthy. (In Introduction to Computational Media at ITP, I learned the Java version of Processing from Daniel Shiffman, and I've found the paradigms I learned in that version carry over pretty smoothly to the JavaScript version.)
For machine learning, I've been using Torch, which uses the Lua programming language. Along with writing my own code, I've experimented extensively with Andrej Karpathy's incredible open source contributions, which include NeuralTalk2 and char-rnn. More recently, I switched to Justin Johnson's torch-rnn, which is 7x more memory efficient than char-rnn.
What would be your dream setup?
In the world of photography, there's a common condition called G.A.S. ("Gear Acquisition Syndrome"), which is pretty well known. I think many people are also prone to the same "gear lust" with respect to new computers. In general, I try not to lust after new equipment, so long as what I'm currently using remains adequate for what I want to do. So far, that's been the case, so I suppose my "dream setup" is my current setup.
That said, I'd absolutely love to own a personal machine learning rig with at least 4 K80 GPUs, running Ubuntu Linux, and a nice large monitor for designing web interfaces (and staring at smoothly rendered high-def fractal patterns). Or maybe one of these new NVIDIA deep learning appliances, which unfortunately cost about as much as a modest house in most places.