Profile picture
Paras Chopra @paraschopra
, 5 tweets, 2 min read Read on Twitter
1/ I trained a neural network to generate philosophy. @sia_steel named her Roomi.

Roomi's inspirations are Camus, Nietzsche, Wittgenstein, Feynman, David Hume, Stephen Hawking, and James Carse.

Here are some of the Roomi's unique philosophical sayings that I liked the best.
@sia_steel
@sia_steel
@sia_steel
In case you want to read more, and see how I generated them, here's how I did it: towardsdatascience.com/generating-new…

If Roomi keeps on making sense, I might also convert this into a bot :)
Missing some Tweet in this thread?
You can try to force a refresh.

Like this thread? Get email updates or save it to PDF!

Subscribe to Paras Chopra
Profile picture

Get real-time email alerts when new unrolls are available from this author!

This content may be removed anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

More from @paraschopra see all

Profile picture
@hardmaru i'm playing with generating art through CPNN blog.otoro.net/2016/03/25/gen…

The extra 'r' input doesn't seem to impact the output. Without r also images look very similar. Why is it there? What's the intuition of it being there?
BTW - I modified approach a little bit so instead of scanning x and y from 0 to max, it starts in the middle and the network outputs where in neighbourhood to move. Sort of like Langton's Ant. en.wikipedia.org/wiki/Langton%2…

It generated the attached. I love the (non)random drift.
And this.

(Sorry I'm DDOSing your twitter notifications, but I'm so in love with generating pretty pictures)
Read 3 tweets
Profile picture
1/ Why do I write? (A thread on writing)
2/ Writing is a way of having long conversations with myself
3/ I write when there’s an incredibly hard idea that I need to teach myself.

I write when I have a glimpse of an insight that’ll remain hidden until caught and penned down.
Read 13 tweets
Profile picture
1/ Was reading and thinking about why deep neural networks work so well on "natural" learning tasks (such as image classification).

Here are my notes.

(If some parts are vague or if they assume a familiarity with a concept you're not aware of, please let me know)
2/ First major insight was that the minibatch of data for gradient descent actually helps in generalization on unseen data.

Gradients of minibatch of data that are specific about that batch cancel over multiple runs and what remains is gradients that are generally applicable.
3/ It is known that neural networks are universal function approximators. That is, given a function they can approximate that function with arbitrary accuracy.

But now I think that's not an interesting result. What's interesting is that they give good answers on unseen data.
Read 19 tweets

Related threads

Profile picture
I trained a neural network on 1,228 types of cookies and apparently these are what cookies sound like to it.
aiweirdness.com/post/180892528… grandma's spritches, hand buttersacks, low fuzzy feats, walps, quitterbread bars, merry hunga poppers, lord's honey fight, fluffin coffee drops, spice biggersNutty Loispers *then Wholes
okay so i had a try at training the neural net on the cookie recipes themselves.
and um @@@@@ Noo No CA Cracker DECOR<br /> <br /> cookies, deserts, holiday<br /> <br /> <span class=1/2 cup brown sugar
1/2 cup sugar (or beaten
1 1/2 cup wheated sugar
1 tablespoon baking soda
1 cup brown sugar
3/4 cup brown sugar
2 egg whites
1/4 cup sugar
1 cup packed brown sugar
1 cup brown sugar
3 each egg whites
2 1/2 cup sugar
1 tablespoon powdered sugar

Cream the egg mixture and vanilla. Beat mixture to mix well. Add egg whites and vanilla and flAv" src="/images/1px.png" data-src="https://pbs.twimg.com/media/Dt1-fRlU8AECLNN.jpg">
here's another neural net cookie recipe. it was looking almost normal for a while there. @@@@@ NooN ----FILADONOUSED----<br /> 2 tablespoon egg whites<br /> <span class=1/2 cup white sugar
1/2 cup flour
1 1/2 teaspoon baking powder
1/2 teaspoon salt
1/2 cup brown sugar
1/2 cup brown sugar
1 teaspoon vanilla extract
1 1/2 cup sugar
1/2 cup wheater water
1/2 cup brown sugar
2 eggs
1/4 cup mark
1 cup butter or margarine
3 cup sugar
1 teaspoon vaniLle
1/2 cup sugar
1 tablespoon milk
1 cup flour
3/4 cup corn" src="/images/1px.png" data-src="https://pbs.twimg.com/media/Dt2LBozUwAAOsDM.jpg">
Read 5 tweets
Profile picture
I trained a neural network to invent new names for nail polish. Some of them are very strange indeed.
aiweirdness.com/post/175608335… Sea of Leather, I Dared, Murder Earth, Girls Burblipp, Chunky the Midnight Doodle, Collide Loopstorm, Batberry, OMG Power, Flame Dust
Batberry and Flame Dust are the only ponies. Every other polish name on this chart is very, very metal.
Here are some rejected neural net nail polish names. Love Rot, anyone? Song of the Booty?
Some of them sound like D&D spells too. Love Rot<br /> Fail Pink<br /> Wart Boy<br /> Blood 204<br /> Dead Pearl<br /> Social Mace<br /> Mooned Black<br /> Old With Anger<br /> Suffering Lights<br /> Acid Touch Green<br /> Evil Gold<br /> Charming Tick<br /> Diamond Flesh<br /> Song of the Booty<br /> Sorry Green<br /> Fail The Baby<br /> Sore Love Diamond Pink<br /> Ashes in the Green<br /> Creamscrap Golden Bop Dragon
Read 3 tweets
Profile picture
LOOKING FOR LOST HEIRS
by Pastor Bertrand L. Comparet,A.B., J.D.

Have you ever stopped to question the intense hatred there is in the world for #white #Christians?
"The written intention of our heavenly Father #Yahweh is expressed in His prayer when we pray, Thy will be done on EARTH as it is in HEAVEN.
"In keeping with this, is #Yahshua's [Jesus'] statement that #Yahweh's servants have been on earth for thousands of years.
Read 66 tweets
Profile picture
Yesterday I made a data set of AI headlines and played with an RNN (recurrent neural network) to generate new AI headlines.

It went...not so well. Unless you're psyched about "President Battlestar"

In the process of copying and pasting 1,000 headlines from TechCrunch, The Verge, Recode, The New York Times, The Washington Post, Slate, The New Yorker, The Atlantic, Reddit, The National Review, Fox News, the MIT Technology Review, and The Onion, I learned some things.
1. Each news source has a distinct style (journalists are prob cringing right now)

*TechCrunch*
a. [startup] raises [amount] [round] for [purpose]
b. [megacorp's] [product] [does a thing] [criticism of thing]
b. [corp] acquires [startup] to [innovation-sounding thing]
Read 11 tweets
Profile picture
So I'm doing a project where I'm training a neural network to generate new knitting patterns. It's going very well.
This is "Mystery lace", test-knit by DataSock.
ravelry.com/discuss/lazy-s… Close-up of the body knitted patch. Caption:
I'm training it on a combination of well-behaved swatch patterns from stitch-maps.com and a crowdsourced set of socks, tentacles, and unmentionables from ravelry.com . And some good sports at Ravelry have been kind enough to test-knit.
This was the first pattern the neural net made. It didn't have a title because the neural net hadn't learned to do titles yet.
Test-knit by emmasee100. ravelry.com/patterns/libra… It is a simple blue tangled cord. Maybe something knitted, maybe a knot from the last chapter of the scout handbook.Row 7 (RS): Slm, knit to last 1 sts, knit to last st, k1. -–<br /> 10: 12, 93, 10, 80, 20, 10, 18, 14, 15, 28, 28, 10, 88, and<br /> 140 stsritcheston ror 36 sts, 510, 18, 20, 10, 180, 19, 1/2 18<br /> sts). Round 97: K1it.<br /> Row 10: Knit.<br /> Row 1: Knit.<br /> Row 1 (RS): Knit.<br /> Row 1: K1, k2tog 3 times, k1, slm; rep from to last st, k6, k1<br /> tbl, knb, kfb a1 3 sts, knit to last st, kfbf, k1 t40 sts sts,<br /> knit to last sto sath.
Read 14 tweets
Profile picture
I trained a neural network to generate Halloween costumes. It had trouble w/the sexy costumes, but loved sharks. aiweirdness.com/post/166814009… Sexy Scare<br /> Sexy the Pumpkin<br /> Saxy Pumpkins<br /> Sexy the Pirate<br /> Sexy Pumpkin Pirate<br /> Sexy Gumb Man<br /> Sexy barber<br /> Sexy Gargles<br /> Sexy humblebee<br /> Sexy The GateSexy Lamp<br /> Sexy Ducty monster<br /> Sexy conchpaper<br /> Sexy the Bumble<br /> Sexy the Super bass<br /> Pretty zombie Space Suit<br /> sexy Drangers<br /> Sexy the SpockPunk Tree<br /> Disco Monster<br /> Spartan Gandalf<br /> Starfleet Shark<br /> A masked box<br /> Martian Devil<br /> Panda Clam<br /> Potato man<br /> Shark Cow<br /> Space Batman<br /> The shark knight<br /> Snape Scarecrow<br /> Gandalf the Good Witch<br /> Professor Panda<br /> Strawberry shark
update: someone has already drawn Sexy Pumpkin Pirate kurjanik.tumblr.com/post/166815605… Sketch of sexy pumpkin pirate. She does not look happy.
Read 14 tweets

Trending hashtags

#Yahweh #neurips4creativity #here #IoT #David #Genesis #100DaysOfMLCode #CrimesAgainstChildren #philosophy #Manassah #NeurIPS2019 #StephenHawking #SoDS #TFHub #Chronicles #commonwealth #Christianity #SIGGRAPH18 #CrimesAgainstHumanity #NeurIPS2018 #John #MachineLearning #Blockchain #SRGAN #ProjectLITbookclub

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just three indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member and get exclusive features!

Premium member ($3.00/month or $30.00/year)

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!