Top all-time · #2106
大トロ ・ Machine Learning
Personal blog · blog.otoro.netaidesignprogramming
大トロ ・ Machine Learning is a personal blog at blog.otoro.net about ai, design and programming. Since 2015, 35 different Hacker News readers have linked it 71 times, most often “Recurrent neural network playing slime volleyball. Can you beat them?”. The last post was in 2022, but the archive is still widely read.
Too few sampled comments (18) to say how often HN recommends it.?
Why HN links it
HN readers link David Ha's blog on neural networks and evolution, covering neuroevolution, evolution strategies, neural slime volleyball, hyper-networks, and handwriting and kanji generation experiments.
- Recommended for · 5 comments
- Called an excellent graphical explanation, a nice visual primer for newcomers, pretty comprehensive, and a great blog; one commenter loved the work.
Summarised by Claude from a sample of 18 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
ml ・ design
Most-linked posts?
- Recurrent neural network playing slime volleyball. Can you beat them?5 people · 8 mentions · first linked 2015-04-13
- Collective Intelligence for Deep Learning: A Survey of Recent Developments4 people · 6 mentions · first linked 2022-10-04
- Show HN: Neural Network Evolution Playground with Backprop NEAT4 people · 5 mentions · first linked 2016-05-09
- A Visual Guide to Evolution Strategies4 people · 5 mentions · first linked 2017-10-29
- Evolving Stable Strategies4 people · 5 mentions · first linked 2017-11-13
- Handwriting Generation Demo in TensorFlow4 people · 4 mentions · first linked 2015-12-12
- Hyper Networks: An RNN that changes the weights of an RNN4 people · 4 mentions · first linked 2016-09-30
- Recurrent Neural Network Tutorial for Artists (2017)3 people · 4 mentions · first linked 2017-01-01
- Recurrent Net Dreams Up Fake Chinese Characters in Vector Format with TensorFlow3 people · 3 mentions · first linked 2015-12-28
- Creatures avoiding planks3 people · 3 mentions · first linked 2015-05-07
What Hacker News says?
http://blog.otoro.net/2015/01/27/neuroevolution-algorithms/ Otoro explains neuroevolution and demonstrates a Silme Volleyball opponent evolving
Comment linking /2015/01/27/neuroevolution-algorithms/
on “NeuroEvolution – Flappy Bird”Submitted A Visual Guide to Evolution Strategies as a story · 260 points
HN discussionOtoro has an excellent page graphically explaining the neuroevolution of an opponent for Slime Volleyball. http://blog.otoro.net/2015/03/28/neural-slime-volleyball/
Comment linking /2015/03/28/neural-slime-volleyball/
on “NeuroEvolution – Flappy Bird”Submitted Creatures avoiding planks as a story · 183 points
HN discussionanother one: Hyper-Networks https://arxiv.org/abs/1609.09106 and a blog post to go with it: http://blog.otoro.net/2016/09/28/hyper-networks/
Comment linking /2016/09/28/hyper-networks/
on “Learning to Learn in TensorFlow”Submitted Recurrent Neural Network Tutorial for Artists (2017) as a story · 141 points
HN discussionThe next article in the series, Evolving Stable Strategies, is also up. http://blog.otoro.net/2017/11/12/evolving-stable-strategies/
Comment linking /2017/11/12/evolving-stable-strategies/
on “A Visual Guide to Evolution Strategies”Submitted Hyper Networks: An RNN that changes the weights of an RNN as a story · 135 points
HN discussionTopology generation is the new feature engineering, it's not surprising that it's the first thing to go. I have been working on advancing HyperNEAT for this very purpose - similar, but not quite the same to: http://blog.otoro.net/2016/05/07/backprop-neat/ Non-sequential and non-hierarchical topologies are the future, and it makes sense that machines should generate them
Comment linking /2016/05/07/backprop-neat/
on “Darpa Goes “Meta” with Machine Learning for Machine Learning (2016)”Submitted Recurrent Net Dreams Up Fake Chinese Characters in Vector Format with TensorFlow as a story · 124 points
HN discussionFor those new to the subject, here is a nice visual primer to Evolution Strategies. http://blog.otoro.net/2017/10/29/visual-evolution-strategies...
Comment linking /2017/10/29/visual-evolution-strategies/
on “The quest to evolve neural networks through evolutionary algorithms”Submitted Permutation-Invariant Neural Networks for Reinforcement Learning as a story · 109 points
HN discussionHi This post outlines a version that doesn't start pre-trained http://blog.otoro.net/2015/05/13/neural-slime-volleyball-evo...
Comment linking /2015/05/13/neural-slime-volleyball-evolution-demo/
on “Show HN: Recurrent neural net plays 'neural slime volleyball' in JavaScript”Submitted Show HN: Neural Network Evolution Playground with Backprop NEAT as a story · 73 points
HN discussionPretty comprehensive details here. http://blog.otoro.net/2015/03/28/neural-slime-volleyball/
Comment linking /2015/03/28/neural-slime-volleyball/
on “Show HN: Recurrent neural net plays 'neural slime volleyball' in JavaScript”Submitted Generating Large Images from Latent Vectors as a story · 60 points
HN discussionThanks for the references So you think basically recording human actions (possibly in some feature extracted way) in part of the 'replay memory' for DQN would work well? I have also been reading about deepmind's recent survey of combining deep learning methods with actor critic models What I also want to explore is the possibility to use evolution to evolve q-functions, which shouldn't be so hard to do, rather than…
Comment linking /2015/04/08/evolutionary-function-approximation-for-reinforcement-learning/
on “Show HN: Recurrent neural net plays 'neural slime volleyball' in JavaScript”Submitted Evolving Stable Strategies as a story · 32 points
HN discussion*references Teaching Deep Convolutional Neural Networks to Play Go by Clarke & Storkey http://arxiv.org/abs/1412.3409 Move Evaluation in Go Using Deep Convolutional Neural Networks by Maddison , Huang , Silver & Sutskever http://arxiv.org/abs/1412.6564 Tesauro's TD-Gammon http://webdocs.cs.ualberta.ca/~sutton/book/ebook/node108.htm... Playing Atari with Deep Reinforcement Learning , Mnih et al. https://www.cs.toront…
Comment linking /2015/01/27/neuroevolution-algorithms/
on “Show HN: Recurrent neural net plays 'neural slime volleyball' in JavaScript”Submitted Generating Large Images from Latent Vectors – Part Two as a story · 18 points
HN discussionPosted a small write up here with regards to the evolution setup used: http://blog.otoro.net/2015/05/13/neural-slime-volleyball-evo...
Comment linking /2015/05/13/neural-slime-volleyball-evolution-demo/
on “Show HN: Recurrent neural net plays 'neural slime volleyball' in JavaScript”Submitted Show HN: Neural Network Evolution Playground with Backprop NEAT as a story · 10 points
HN discussionHi Thomas I manually take the outputs of a feed forward network and out them back as inputs in the next time step, kind of making it a fully connected RNN. Here are some details of the implementation: http://blog.otoro.net/2015/03/28/neural-slime-volleyball/
Comment linking /2015/03/28/neural-slime-volleyball/
on “Show HN: Recurrent neural net plays 'neural slime volleyball' in JavaScript”Submitted Hyper Networks – 大トロ as a story · 8 points
HN discussionThe number of weights depends on the network architecture. In this case fully connected , 1 weight per connection . 16 inputs to 7 neurons + 7 neurons to 3 outputs + 7 biases 16x7 + 7x3 + 7 = 140 http://blog.otoro.net/wp-content/uploads/sites/2/2015/03/sli... The number of neurons is a decidable.
Comment linking /wp-content/uploads/sites/2/2015/03/slime_rnn.png
on “Show HN: Recurrent neural net plays 'neural slime volleyball' in JavaScript”Submitted Show HN: Neurogram – Creating art by evolving neural networks in Javascript as a story · 7 points
HN discussion[reply to darkmighty comment below, thread depth limitation] Indeed this is exactly it, evolution is a global method, learning is local. I am reading John Gomez's thesis where he compares and combines learning and evolution http://www.cs.utexas.edu/users/nn/downloads/papers/gomez.phd... Otoro's post on the evolution of his slime volleyball thinking is well worth a read. http://blog.otoro.net/2015/01/27/neuroevolutio…
Comment linking /2015/01/27/neuroevolution-algorithms/
on “Show HN: Recurrent neural net plays 'neural slime volleyball' in JavaScript”Submitted Hyper Networks as a story · 6 points
HN discussionhaha, I like that. Yeah, there are a total of 140 weights/biases in the neural network. This stuff is not bleeding edge - just taking techniques popular in academia and industry many years ago to make a fun game mainly for my own educational purpose. Basically recently I saw a lot of work on AI's playing ATARI games and wanted to learn reinforcement learning techniques, and the best way to learn is to build a game!…
Comment linking /2015/03/28/neural-slime-volleyball/
on “Show HN: Recurrent neural net plays 'neural slime volleyball' in JavaScript”Submitted Recurrent Neural Network Tutorial for Artists Using JavaScript and P5.js as a story · 4 points
HN discussion
