SFV: Reinforcement Learning of Physical Skills from Videos: #ML
https://arxiv.org/abs/1810.03599v1
-
-
TrussFormer: 3D printing large kinetic structures:
-
The most effective way to deal with uncertainty is not control, but relaxation.
-
"Complex Adaptive Systems" - Keynote by Dave @Snowded:
"A framework allows you to look at things from different perspectives - a model seeks to represent reality. It is different in typology and taxonomy: A taxonomy forces you to put things into boxes, a typology says look at things from these perspectives" - @snowded
"It's a lot easier to train users to talk to IT people, than to train IT people to understand users" - @snowded
"The intervention was to say if you have 10y experience and somebody with 5y experience signs it off, you can break any rule - provided you documented it. So they created a rule about when rules can be broken. Which is kind of like recognising reality" - @snowded
One more recent talk by @snowded which is worth while:
https://www.youtube.com/watch?v=r8T7wlJ8DgM -
Mcluhan's diagnosis that "Augmentation" leads to "Amputation" is spot on. It is especially fascinating how Augmentation is related to "behaving like an insensitive, asocial idiot". Ego scales according to the size of the machine one controls, ethics & empathy don't.
#Augmentation #Ideas
-
"MIDI-VAE: Modeling Dynamics and Instrumentation of Music with Applications to Style Transfer": https://arxiv.org/abs/1809.07600 code: https://github.com/brunnergino/MIDI-VAE
"Symbolic Music Genre Transfer with CycleGAN": https://arxiv.org/abs/1809.07575 code: https://github.com/sumuzhao/CycleGAN-Music-Style-Transfer #Generative #Music #ML
-
Why we really really really like repetition in music:
-
"Attribute Enhanced Face Aging with Wavelet-based Generative Adversarial Networks":
https://arxiv.org/abs/1809.06647v1 #ML #Generative
-
Neural Animation and Reenactment of Human Actor Videos:
http://gvv.mpi-inf.mpg.de/projects/wxu/HumanReenactment/ -
Video-based Reconstruction of 3D People Models:
Paper: https://graphics.tu-bs.de/upload/publications/alldieck2018videopeople.pdf
Code: https://github.com/thmoa/videoavatars -
"My journey into fractals" - by @Bananaft:
https://medium.com/@bananaft/my-journey-into-fractals-d25ebc6c4dc2 -
Keynote by @plamere from Spotify on having fun with music and datascience:
-
"Neri Oxman's swarm of Fiberbots autonomously build architectural structures":
https://www.dezeen.com/2018/10/05/neri-oxman-fiberbots-mediated-matter-lab-mit-architectural-structures/ -
Slitscan Carnival
Creating Slitscan Images from Video in Python and MoviePy
-
"Plans, Takes, and Mis-takes": https://tidsskrift.dk/index.php/outlines/article/view/1964
"This paper analyzes what may have been a mistake by pianist Thelonious Monk playing a jazz solo in 1958". #Music #A mistake is the most beautiful thing in the world. It is the only way you can get to some place you’ve never been before. I try to make as many as I can. Making a mistake is the only way that you can grow. - Drummer E.W.Wainwright
-
Human-AI Collaborated Graffiti:
https://howtogeneratealmostanything.com/graffiti/2018/09/26/episode5.html -
Believe it or not: Designing a Human-AI Partnership for Mixed-Initiative Fact-Checking
-
Youtube-8M: https://research.google.com/youtube8m/ #ML
https://research.google.com/youtube8m/explore.htmlYouTube-8M is a large-scale labeled video dataset that consists of millions of YouTube video IDs, with high-quality machine-generated annotations from a diverse vocabulary of 3,800+ visual entities. It comes with precomputed audio-visual features from billions of frames and audio segments, designed to fit on a single hard disk. This makes it possible to train a strong baseline model on this dataset in less than a day on a single GPU! At the same time, the dataset's scale and diversity can enable deep exploration of complex audio-visual models that can take weeks to train even in a distributed fashion.
