Recurrent neural network for generating stories about images
November 6, 2015 12:21 PM Subscribe
This experiment explores how to generate little romantic stories about images, using neural-storyteller, a recently published experiment by Ryan Kiros.
Lets get Real!Swift to cover Robo!Swift so we can have the Gibsonesque future we where promised and are doomed to get.
posted by The Whelk at 1:05 PM on November 6, 2015 [1 favorite]
posted by The Whelk at 1:05 PM on November 6, 2015 [1 favorite]
I have worked with story-telling algorithms for case-based-reasoning before, this page provides an excellent overview of basic approaches (I used a modified version of TALESPIN - which was used to recreate Aesop's fables). TALESPIN, from 1977, seems to be, uh... significantly better at telling stories than neural-storyteller (you'd have to couple it with a random generator of story state to begin the stories though. hmmm.... automated twitter bot thoughts...).
posted by combinatorial explosion at 1:05 PM on November 6, 2015 [1 favorite]
posted by combinatorial explosion at 1:05 PM on November 6, 2015 [1 favorite]
Just from a quick skim, wouldn't something like TALESPIN require a lot more explicit training and rule specification, vs. just pointing it at a few corpuses and letting it rip?
posted by en forme de poire at 1:17 PM on November 6, 2015
posted by en forme de poire at 1:17 PM on November 6, 2015
the mean of skip-thought vectors across Taylor Swift lyrics
What a weird future. Sad that my first thought was "I wonder how this will be used in phishing"
posted by RobotVoodooPower at 1:37 PM on November 6, 2015 [2 favorites]
What a weird future. Sad that my first thought was "I wonder how this will be used in phishing"
posted by RobotVoodooPower at 1:37 PM on November 6, 2015 [2 favorites]
Has anyone figured out how to set this up on their machines? I'm stuck at the part that says, "Next, you need to obtain the pre-trained skip-thoughts encoder," because it is not so clear what part of the encoder to wget.
posted by johngoren at 7:18 PM on November 6, 2015
posted by johngoren at 7:18 PM on November 6, 2015
I'm just trying to get win64 theano working.
Spoiler: It is difficult.
posted by PMdixon at 8:55 PM on November 6, 2015
Spoiler: It is difficult.
posted by PMdixon at 8:55 PM on November 6, 2015
I confess i don't even understand what is meant by the "lasagna" in the generator.py's "import lasagna." Maybe we could write a nice installation script
posted by johngoren at 10:56 PM on November 6, 2015
posted by johngoren at 10:56 PM on November 6, 2015
« Older I'll take the sheath style with a brush train in... | "We failed you." Newer »
This thread has been archived and is closed to new comments
posted by poe at 12:43 PM on November 6, 2015 [5 favorites]