Slide 1

Slide 1 text

Dreams, Drugs & ConvNets Piotr Migdał, PhD deepsense.io / freelancer http://p.migdal.pl/ 2 Mar 2017, Studenckie Koło Naukowe Neurobiologii UW

Slide 2

Slide 2 text

No content

Slide 3

Slide 3 text

No content

Slide 4

Slide 4 text

Deep Learning progress • image recognition, neural style, word analogies, per-char translations, playing ATARI games, Go, [no idea what’s next] • fast-paced (more than my quantum physics PhD):
 6 month ago a breakthrough, now a baseline • (no questions about Singularity please!)

Slide 5

Slide 5 text

Artificial neural networks • Multidimensional arrays • Simple matrix operations (add, multiply) • Floats (not spikes) • Simple activation functions (sigmoid, ReLU) • Cost function and back-propagation • A lot of weights (e.g. 100M)

Slide 6

Slide 6 text

Many layers https://research.facebook.com/publications/deepface-closing-the-gap-to-human-level-performance-in-face-verification/

Slide 7

Slide 7 text

Convolutions http://setosa.io/ev/image-kernels/

Slide 8

Slide 8 text

Patterns activating channels https://blog.keras.io/how-convolutional-neural-networks-see-the-world.html

Slide 9

Slide 9 text

Big questions • creating general AI or sentient beings • transferring our minds (mind uploading) “I mean, if we're able to save even just a small piece of ourselves, why wouldn't we do that?” - SOMA https://www.youtube.com/watch?v=BZTfi1jv-EE

Slide 10

Slide 10 text

Flesh vs machines:
 practical questions • Cross-inspiration: • Machine Learning to Human Cognition • Human Cognition to Machine Learning • Common abstractions • Learning in general • Hierarchical signal processing

Slide 11

Slide 11 text

NSFW, 18+, etc

Slide 12

Slide 12 text

http://www.techrepublic.com/article/why-microsofts-tay-ai-bot-went-wrong/ Tay.AI (+ 4chan)

Slide 13

Slide 13 text

The biggest lie of every
 parent, pet owner
 and neural network trainer: “I’ve never shown it that, it must have learned it by itself!”

Slide 14

Slide 14 text

Stereotypes http://p.migdal.pl/2017/01/06/king-man-woman-queen-why.html

Slide 15

Slide 15 text

If you’ve never had a trip… https://youtu.be/cPKq7JuQDvg?t=256

Slide 16

Slide 16 text

If you’ve never had a trip… https://youtu.be/z7_U6y8kJaY?t=72

Slide 17

Slide 17 text

Deep dreams: forcing ConvNet
 to see things https://research.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html

Slide 18

Slide 18 text

• What should I use:
 THC, LSD, DMT? • A good GPU! http://alexgrey.com/shop/st-al.html

Slide 19

Slide 19 text

Grocery trip https://www.youtube.com/watch?v=DgPaCWJL7XI

Slide 20

Slide 20 text

Forcing networks to see things https://research.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html

Slide 21

Slide 21 text

Neural style transfer also video: https://www.youtube.com/watch?v=Khuj4ASldmU

Slide 22

Slide 22 text

With own photos to play with https://deepart.io/ or https://prisma-ai.com/

Slide 23

Slide 23 text

What is sexual? http://blog.clarifai.com/what-convolutional-neural-networks-see-at-when-they-see-nudity/#.WLhGmRIrLLh

Slide 24

Slide 24 text

What is sensual? http://blog.clarifai.com/what-convolutional-neural-networks-see-at-when-they-see-nudity/#.WLhGmRIrLLh

Slide 25

Slide 25 text

And this? (yes, ConvNet is also fooled)

Slide 26

Slide 26 text

Sketch to picture http://affinelayer.com/pixsrv/

Slide 27

Slide 27 text

Sketch to picture: let’s get rogue! http://affinelayer.com/pixsrv/

Slide 28

Slide 28 text

https://raw.githubusercontent.com/Newmu/dcgan_code/master/images/faces_arithmetic_collage.png https://github.com/255BITS/HyperGAN Picture analogies

Slide 29

Slide 29 text

https://open_nsfw.gitlab.io/ Image generation

Slide 30

Slide 30 text

Learning what is important • Object consistency • Toddler needs to learn, so do ConvNets http://blogs.scientificamerican.com/illusion-chasers/what-little-babies-see-that-you-no-longer-can/

Slide 31

Slide 31 text

Uncertainty and learning http://www.nature.com/articles/ncomms11609

Slide 32

Slide 32 text

Retrain your own network:
 McCollough effect http://www.michaelbach.de/ot/col-McCollough/

Slide 33

Slide 33 text

Trypophobia
 (don’t google it!) visit here instead: https://en.wikipedia.org/wiki/Trypophobia

Slide 34

Slide 34 text

Trypophobia detector https://github.com/grzegorz225/trypophobia-detector (a 5-day workshop was enough for 88% accuracy!)

Slide 35

Slide 35 text

And if you
 don’t want to code… http://www.sciencemag.org/news/2015/11/pigeons-spot-cancer-well-human-experts https://www.youtube.com/watch?v=flzGjnJLyS0

Slide 36

Slide 36 text

Questions? • If you want to learn: • http://lumiverse.io/series/neural-networks-demystified • https://gist.github.com/stared/7de2908b9bcba01c39ee3c591875a23c • http://www.deeplearningbook.org/ • Keras (on top of TensorFlow) in Jupyter Notebook • https://medium.com/@mateuszsieniawski/keras-with-gpu-on-amazon-ec2-a-step-by- step-instruction-4f90364e49ac#.4i1norivh • https://www.erowid.org/ http://p.migdal.pl/ + I do like emails: [email protected]