Recently visited EMNLP 2018 and, in addition to the hard science, saw there a lot of funny things. Some of them:
* a team of 9 authors from Columbia University, NYC wrote an article on the identification of aggression between the gangs of New York and the death of gangs' individual members based on the Twitter messages flow bit.ly/2Adqt5o
* a group of researchers From the University of Texas, Austin, classifies the usage of swears and vulgar expressions in online communication (6 types, from expressing emotions to signal group identity) bit.ly/2R4v1Ck using a model trained on a specially assembled dataset, which they also published here: bit.ly/2zmVmob
* a pair of scientists from KAIST, South Korea, took the annals of the Joseon dynasty (Korean government chronicle 14-19 centuries) and use them to train their Conversational Decision Making Model to predict the decision of the King based on the preceeding debates between advisers bit.ly/2OX72TR. There are 13K cases handled by 15 different kings, each judgement is reduced to 5 classes, dataset and code posted here bit.ly/2r1wzSc.
* a team of 9 authors from Columbia University, NYC wrote an article on the identification of aggression between the gangs of New York and the death of gangs' individual members based on the Twitter messages flow bit.ly/2Adqt5o
* a group of researchers From the University of Texas, Austin, classifies the usage of swears and vulgar expressions in online communication (6 types, from expressing emotions to signal group identity) bit.ly/2R4v1Ck using a model trained on a specially assembled dataset, which they also published here: bit.ly/2zmVmob
* a pair of scientists from KAIST, South Korea, took the annals of the Joseon dynasty (Korean government chronicle 14-19 centuries) and use them to train their Conversational Decision Making Model to predict the decision of the King based on the preceeding debates between advisers bit.ly/2OX72TR. There are 13K cases handled by 15 different kings, each judgement is reduced to 5 classes, dataset and code posted here bit.ly/2r1wzSc.
Suddenly: CariGAN -- a neural network trained to draw caricatures from photographs. Article: stanford.io/2QeZeBv pictures: stanford.io/2P13IH4 project page: bit.ly/2zmVVi6
Comrade Lawrence Kesteloot bit.ly/2zrrDut wrote a raytracer to reproduce the cover of Pink Floyd's Dark Side of the Moon album. The code and details are here: bit.ly/2zsab8Z.
Picture showing the superiority of the bicycle (and fish) over other miserable types of transport. Mass VS normalized energy, loglog. Found on Twitter, the original source is this book: bit.ly/2P5VDRq
Playing with different text generation neural networks, at some point I came up with the idea to generate names of different colors. It turned out that the idea is not so new, but I did not like the results of previous attempts, and there were several ideas that I wanted to check myself. Therefore, I made a network that generates color names and adapted an interactive visualizer to explore these names.
You can see the visualization here (better use a 600+px width screen): bit.ly/2rb805f
Also, the post with some details is here: bit.ly/2BGzaXS
You can see the visualization here (better use a 600+px width screen): bit.ly/2rb805f
Also, the post with some details is here: bit.ly/2BGzaXS
People are divided into those who understand regular expressions and the rest. The former should like this regexp-puzzles site bit.ly/1yX1mwj the latter is better to read some FAQ first bit.ly/2QaTipz
Today started Advent of Code 2018, the third annual advent calendar with programming puzzles. Every day from 1 to 25 December you get two tasks described in the form of cyber-Christmas tales, and decorated in cozy ascii-graphics.
Based on the past two years, I should say that the average complexity of the tasks is low, but some of them occasionally could be really tricky. In general this is not a leetcode on hardmode, but rather an opportunity to improve knowledge or learn a new language in a game form. This year I think to try it with Go lang.
Link: https://adventofcode.com/2018
Based on the past two years, I should say that the average complexity of the tasks is low, but some of them occasionally could be really tricky. In general this is not a leetcode on hardmode, but rather an opportunity to improve knowledge or learn a new language in a game form. This year I think to try it with Go lang.
Link: https://adventofcode.com/2018
Friday's post on Wednesday! The idea is not so new, but Mario Klingemann seems to have brought it to a new level of quality: he took BigGAN bit.ly/2EKcAkB (the most powerful method of generating images by the neural network currently) and has forced it to generate porn detector triggering images. Then he published it all on Tumblr, which just recently declared Holy war on NSFW-content bit.ly/2zFOHpj. So, if you want to take a look at the "neural network generative porn", it will be available here for another 5 days: bit.ly/2LaP4Ne.
This recalled me another story -- about the year 2011 or 2012 guys from the Yandex computer vision team made a powerful pornfilter to sanitize the results of Image search. As a side product they launched the internal pilot service Rule34, which for any given query searched on the Internet for relevant pictures, but with a reverse boost of that filter. A main page of this search was standard, so it has a typical block of popular safe queries, like this bit.ly/2zREdTN. Amazing things could be discovered by that service even with the most innocent queries, such as "abstract figures" and "birch forest", to say nothing about the "secrets of the Kremlin" and "wedding dresses". Pity, it never made it into the production.
This recalled me another story -- about the year 2011 or 2012 guys from the Yandex computer vision team made a powerful pornfilter to sanitize the results of Image search. As a side product they launched the internal pilot service Rule34, which for any given query searched on the Internet for relevant pictures, but with a reverse boost of that filter. A main page of this search was standard, so it has a typical block of popular safe queries, like this bit.ly/2zREdTN. Amazing things could be discovered by that service even with the most innocent queries, such as "abstract figures" and "birch forest", to say nothing about the "secrets of the Kremlin" and "wedding dresses". Pity, it never made it into the production.
Once, in my first year at university, I studied C, and to make it more fun, my friends and I analyzed the principles of some works from IOCCC bit.ly/2PzRtBJ, the international obfuscated C code contest which has been taking place since 1984. I just checked it and it's still pretty alive today. Take, for example, this year's entry bit.ly/2QUzLOc, which could be compiled into different programms, if you rotate and reflect its source code, designed in the form of a square. Recommended to all the fans of hardcore debug.
There was also an attempt bit.ly/1rlsDVE to do a similar annual contest for Perl, but it did not last long. According to my version, there was too high a risk of singularity, and the project had to be closed, and the involved participants were isolated.
Well, this year, in accordance with my concept of JS is the new ASM bit.ly/2MbycoY, Arvid Gerstmann is organizing IOJCC bit.ly/2UCpi97, the same style competition in Javascript crafts. Submission of works is already open and will last until March 1, 2019. Meanwhile, the inspiration for the contest can be found in the Dwitter-e bit.ly/2Gh07FK - twitter-like project, where each post is a 140-character javascript code.
There was also an attempt bit.ly/1rlsDVE to do a similar annual contest for Perl, but it did not last long. According to my version, there was too high a risk of singularity, and the project had to be closed, and the involved participants were isolated.
Well, this year, in accordance with my concept of JS is the new ASM bit.ly/2MbycoY, Arvid Gerstmann is organizing IOJCC bit.ly/2UCpi97, the same style competition in Javascript crafts. Submission of works is already open and will last until March 1, 2019. Meanwhile, the inspiration for the contest can be found in the Dwitter-e bit.ly/2Gh07FK - twitter-like project, where each post is a 140-character javascript code.
Suddenly: a letter from Claude Shannon (father of information theory) bit.ly/20DRhC6 to Warren McCulloch (father of mathematical model of neuron) bit.ly/2eSkb1q about Ron Hubbard (father of scientology and dianetics) bit.ly/2SKKxDO. This correspondence mentioned in bit.ly/2S28Woo
Just found a rather old project on a painterly "harmonization" -- add a foreign object to the image and style it to the original. Seems to be easy, but I feel a lot of potential. Article: bit.ly/2SPHp9E code and more pictures: bit.ly/2GVC5fq
A detailed and very interesting flow-chart of the world religions history from historian John B. Sparks, the 1952 year's edition by Rand McNally. Detailed zoom version is here: bit.ly/2UT2Oka
Rubric "Corner of cybernecrophile":
The interactive fiction is an ancient chthonic genre of computer games, sometimes also called text adventures or simply adventures. Games of this genre are something in the middle between cross between an interactive book, an ADnD game with a good DM, and a classic quest, but without graphics. Actually, the Quest genre was born from interactive fiction when good enough monitors appeared.
If you have not heard about it at all, you can read some very basic introductory information about the genre here bit.ly/2Cp7lUr, also there is a great documentary Get Lamp bit.ly/2SZ2bDP. The peak of the popularity of these games came at the beginning of the 80th bit.ly/2GvtZhV, then there were years of oblivion, but a small subculture is still alive and occasionally pleases its audience with interesting indie games. It is quite difficult to play such games, especially if you did not start in childhood -- the genre has a quite high entry threshold, but if you still want to try, I recommend using Frotz or Gargoyle as an interpreter, and start with some simple and nice games, for example, with The Wizard Sniffer bit.ly/2ScIx7p or Hunger Daemon bit.ly/2Gv3xVK -- they are small has a good built-in hint systems.
There is a reason of such post today: this year marks the 20th anniversary of the classic Anchorhead game in the lovecraft horror genre bit.ly/2SbahJG, and the masters of the genre celebrated it with a tribute, for which 84 authors made one location each, and then put it all together into something called Cragne Manor bit.ly/2LqmebI. They published it literally a week ago, so I myself had not yet had time to look (maybe on holidays), but some of the names in the list of authors are impressive.
The interactive fiction is an ancient chthonic genre of computer games, sometimes also called text adventures or simply adventures. Games of this genre are something in the middle between cross between an interactive book, an ADnD game with a good DM, and a classic quest, but without graphics. Actually, the Quest genre was born from interactive fiction when good enough monitors appeared.
If you have not heard about it at all, you can read some very basic introductory information about the genre here bit.ly/2Cp7lUr, also there is a great documentary Get Lamp bit.ly/2SZ2bDP. The peak of the popularity of these games came at the beginning of the 80th bit.ly/2GvtZhV, then there were years of oblivion, but a small subculture is still alive and occasionally pleases its audience with interesting indie games. It is quite difficult to play such games, especially if you did not start in childhood -- the genre has a quite high entry threshold, but if you still want to try, I recommend using Frotz or Gargoyle as an interpreter, and start with some simple and nice games, for example, with The Wizard Sniffer bit.ly/2ScIx7p or Hunger Daemon bit.ly/2Gv3xVK -- they are small has a good built-in hint systems.
There is a reason of such post today: this year marks the 20th anniversary of the classic Anchorhead game in the lovecraft horror genre bit.ly/2SbahJG, and the masters of the genre celebrated it with a tribute, for which 84 authors made one location each, and then put it all together into something called Cragne Manor bit.ly/2LqmebI. They published it literally a week ago, so I myself had not yet had time to look (maybe on holidays), but some of the names in the list of authors are impressive.
Flickr
Some Things Never Change
Explore Jason "Textfiles" Scott's 53507 photos on Flickr!
Phases of the Moon calendar for 2019 bit.ly/2BAtKMV from Michael Bostock, the author D3js.org, worked for many years as a data journalist for NYT. By the way, I recommend to study his other works: bit.ly/28Y3uwb, Tafti himself praises him very much on.ft.com/2rPEaDW :)
Recently, a new issue bit.ly/2QKYQf1 of SEEDS was published. It's an annual anthology on procedural content generation -- music, graphics, games, text, etc. It's a project of ProcJam.com team which is aBritish hackathon on the same topics.
As the author of the project Neural defense and Yandex's Autopoet I cannot but rejoice this caricature about one of the first generators of poetry, AUTO-BEATNIK, 1962 bit.ly/2BFWxPR. I remember once I gave a lecture about my experiments at the "Good text" school, so "real poets" listened to me with just the same kind of faces :)
Accidentally found out that Donald Knuth (one of the main scientists of the twentieth century in information science) is still read once a year a Christmas lecture on algorithms at Stanford. This year, the 24th annual lecture bit.ly/2GCjObu is about the Dancing Links algorithm bit.ly/2V9vVzR. For me it resonates personally with the Zaliznyak's annual lectures on a birch-bark documents decryption bit.ly/2EJVotT that Andrey was reading before his death last year, and this year continued to read Alexei Gippius.
The post for enhancing the connectivity of this blog: By the link from the SEEDS t.me/pathetic_low_frequenciers_eng/96 I found the project bit.ly/2bvwIEx of a raytracer https://t.me/pathetic_low_frequenciers_eng/78 on the card, winning IOCCC 2011 https://t.me/pathetic_low_frequenciers_eng/89.