Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

DeepMind has open-sourced the heart of AlphaGo and AlphaZero

twitter.com

Post reply on HN
Print view View original

71–80 of 91 posts

Re: DeepMind has open-sourced the heart of AlphaGo and AlphaZero

#71

Post by sebstefan »

rhdunn wrote: post #39
londons_explore wrote: post #16

Worth noting that while AlphaGo and AlphaZero are incredible achievements, the amount of actual code to implement them isn't very much. If you have the research paper, someone in the field could reimplement them in a few days. Then there is the large compute cost for training them to produce the trained weights. So, opensourcing these bits of work without the weights isn't as major a thing as you might imagine.

There is Leela Zero ( https://github.com/leela-zero/leela-zero ) for Go and lc0/Leela Chess ( https://github.com/orgs/LeelaChessZero/repositories ) for Chess, where both provide trained weights. The Leela Chess project specifically have been working for a long time on training and refining the weights for Chess, as well as providing the code -- they allow you to see the history and performance over time for the vario…

I'm moderately into chess but I have never heard of Leela

I'm shocked to discover it's been rated higher than AlphaZero & Komodo and just slightly below Stockfish

Re: DeepMind has open-sourced the heart of AlphaGo and AlphaZero

#72

Post by ragnarsson »

CSMastermind wrote: post #70
rhdunn wrote: post #39

Earlier quoted context omitted.

There is Leela Zero ( https://github.com/leela-zero/leela-zero ) for Go and lc0/Leela Chess ( https://github.com/orgs/LeelaChessZero/repositories ) for Chess, where both provide trained weights. The Leela Chess project specifically have been working for a long time on training and refining the weights for Chess, as well as providing the code -- they allow you to see the history and performance over time for the vario…

If I'm not mistaken Stockfish has it's own neural network implementation as well correct?

yes https://cp4space.hatsya.com/2021/01/08/the-neural-network-of...

Re: DeepMind has open-sourced the heart of AlphaGo and AlphaZero

#73

Post by mensetmanusman »

lern_too_spel wrote: post #69
mensetmanusman wrote: post #49

Earlier quoted context omitted.

If only there existed a distributed way to incentivize calculation of AI weightings while also providing a currency to encourage scale…

As many people have pointed out before, the distributed currency part adds energy waste. BOINC accomplishes the same without the waste.

Of course there will be trade-offs of inefficiency, that is always the case with distribution.

Is it valuable to have open source AI systems is the countering question to that…

Re: DeepMind has open-sourced the heart of AlphaGo and AlphaZero

#74

Post by swyx »

brilee wrote: post #62
londons_explore wrote: post #16

Worth noting that while AlphaGo and AlphaZero are incredible achievements, the amount of actual code to implement them isn't very much. If you have the research paper, someone in the field could reimplement them in a few days. Then there is the large compute cost for training them to produce the trained weights. So, opensourcing these bits of work without the weights isn't as major a thing as you might imagine.

> If you have the research paper, someone in the field could reimplement them in a few days. Hi I did this while I was at Google Brain and it took our team of three more like a year. The "reimplementation" part took 3 months or so and the rest of the time was literally trying to debug and figure out all of the subtleties that were not quite mentioned in the paper. See https://openreview.net/forum?id=H1eerhIpLV

alltime classic Hacker News moment: "heh someone in the field could write this in a few days" "Hi its me 3 of us literally work at google brain and it took us a year"

Re: DeepMind has open-sourced the heart of AlphaGo and AlphaZero

#75

Post by swyx »

knicholes wrote: post #65
Synaesthesia wrote: post #10

Please do the same for Alphastar, the starcraft AI, that would be great.

I talked to the guy who led AlphaStar at Blizzcon. I asked if I could get the weights. He said that balance changes and map changes makes keeping the model updated prohibitive.

sure but even frozen outdated weights would be very instructive to study

Re: DeepMind has open-sourced the heart of AlphaGo and AlphaZero

#76

Post by codeflo »

d4rkp4ttern wrote: post #37
codeflo wrote: post #4

“The heart of” can mean many things, is this runnable or not? Also, they should open source the weights as well so that their claims can finally be verified independently.

And let us not forget those secret random seeds!

Is this a joke I don't get? Because it's a bit unlikely there's something interesting in random seeds.

Re: DeepMind has open-sourced the heart of AlphaGo and AlphaZero

#77

Post by abraxas »

brilee wrote: post #62
londons_explore wrote: post #16

Worth noting that while AlphaGo and AlphaZero are incredible achievements, the amount of actual code to implement them isn't very much. If you have the research paper, someone in the field could reimplement them in a few days. Then there is the large compute cost for training them to produce the trained weights. So, opensourcing these bits of work without the weights isn't as major a thing as you might imagine.

> If you have the research paper, someone in the field could reimplement them in a few days. Hi I did this while I was at Google Brain and it took our team of three more like a year. The "reimplementation" part took 3 months or so and the rest of the time was literally trying to debug and figure out all of the subtleties that were not quite mentioned in the paper. See https://openreview.net/forum?id=H1eerhIpLV

How come you weren't able to just get it from DeepMind given that they are a subsidiary of Google? Is there a lot of red tape involved in exchanging IP like that?

Re: DeepMind has open-sourced the heart of AlphaGo and AlphaZero

#78

Post by amj2 »

londons_explore wrote: post #27
tiagod wrote: post #24

Earlier quoted context omitted.

> Then there is the large compute cost for training them to produce the trained weights. And as far as I understand, the training code is where the secret sauce lies.

generally no... Secret sauce is in the ML compiler and accelerator used, but all those improvements simply lower the cost of training a model. You could still do it on a regular GPU, it would just take you more time. In the case of Google, they probably used TPU chips that you can't get direct 'bare metal' access to anyway, so none of that code would have helped. The actual optimizer used and parameters (like the lea…

You should pencil out on a napkin just how long "more time" is. Here, i'll get you started:

1600 inferences per move * 1ms per inference * 250 moves/game * 30M games played = 12B seconds. 140k days; muzero with gumbel brought down the 1600 to ~40, but either way, you need some more scale.

It turns out a lot of the difficulties, judgment calls, and implementation details involve data pipelining. Some of those choices affect the final skill ceiling you reach. Which ones? How much? Are they path dependent? Well, you'll need to run it more than once...

Re: DeepMind has open-sourced the heart of AlphaGo and AlphaZero

#79

Post by amj2 »

abraxas wrote: post #77
brilee wrote: post #62

Earlier quoted context omitted.

> If you have the research paper, someone in the field could reimplement them in a few days. Hi I did this while I was at Google Brain and it took our team of three more like a year. The "reimplementation" part took 3 months or so and the rest of the time was literally trying to debug and figure out all of the subtleties that were not quite mentioned in the paper. See https://openreview.net/forum?id=H1eerhIpLV

How come you weren't able to just get it from DeepMind given that they are a subsidiary of Google? Is there a lot of red tape involved in exchanging IP like that?

They were & are very protective of the AlphaGo "brand", is the best-case explanation.

Re: DeepMind has open-sourced the heart of AlphaGo and AlphaZero

#80

Post by dllthomas »

westurner wrote: post #66
brilee wrote: post #62

Earlier quoted context omitted.

> If you have the research paper, someone in the field could reimplement them in a few days. Hi I did this while I was at Google Brain and it took our team of three more like a year. The "reimplementation" part took 3 months or so and the rest of the time was literally trying to debug and figure out all of the subtleties that were not quite mentioned in the paper. See https://openreview.net/forum?id=H1eerhIpLV

Replication crisis: https://en.wikipedia.org/wiki/Replication_crisis : > The replication crisis (also called the replicability crisis and the reproducibility crisis) is an ongoing methodological crisis in which the results of many scientific studies are difficult or impossible to reproduce. Because the reproducibility of empirical results is an essential part of the scientific method,[2] such failures undermine the c…

I think I agree with everything you've said here, but just want to note that while we absolutely should (where relevant) expect published code including automated tests, we should not typically consider reproduction that reuses that code to be "replication" per se. As I understand it, replication isn't merely a test for fraud (which rerunning should typically detect) and mistakes (which rerunning might sometimes detect) but also a test that the paper successfully communicates the ideas such that other human minds can work with them.
Post reply on HN