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

81–90 of 91 posts

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

#81

Post by westurner »

dllthomas wrote: post #80
westurner wrote: post #66

Earlier quoted context omitted.

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 dete…

Sources of variance; Experimental Design, Hardware, Software, irrelevant environmental conditions/state, Data (Sample(s)), Analysis

Can you run the notebook again with the exact same data sample (input) and get the same charts and summary statistics (output)? Is there a way to test the stability of those outputs over time?

Can you run the same experiment (the same 'experimental design'), ceteris paribus (everything else being equal) and a different sample (input) and get a very similar output? Is it stable, differentiable, independent, nonlinear, reversible; Does it converge?

Now I have to go look up the definitions for Replication, Repeatability, Reproducibility

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

#82

Post by rcme »

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

One thing that's not appreciated by many who haven't tried to implement a NN is how subtle bugs can be. When you look at code for a NN, it's generally pretty simple. However, what happens when your code doesn't produce the output you were expecting? When that happens, it can be very difficult and time consuming to find the subtle issue with your code.

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

#83

Post by whamlastxmas »

swyx wrote: post #74
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

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"

I don't know the subject matter well enough to make the call, but it's possible the OP is making a general statement that's generally true even if it's not in this specific context of it taking a year.

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

#84

Post by throwaway81523 »

mtlmtlmtlmtl wrote: post #53
mensetmanusman wrote: post #50

Earlier quoted context omitted.

AZ revolutionized chess engine architecture and then moved on to other fields. Fast followers who incrementally improve upon the breakthroughs are very rarely recognized. E.g. inventor of the blue LED versus those who improve the efficiency by .1%

Stockfish' architecture is in no way revolutionised by AZ. It has gone its own way.

The current versions of Stockfish (designated NNUE) are definitely AZ influenced. In fact it was implemented in collaboration with the LCZero devs.

https://stockfishchess.org/blog/2020/introducing-nnue-evalua...

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

#85

Post by lern_too_spel »

mensetmanusman wrote: post #73
lern_too_spel wrote: post #69

Earlier quoted context omitted.

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…

The point is you can develop open source AI cheaper without blockchain.

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

#86

Post by gowld »

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?

That defeats the purpose of reimplementing it?

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

#87

Post by runnerup »

Synaesthesia wrote: post #10

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

I personally hope these types of AI's make it into future games as bots to play against in single-player, and also as an agent to run some of the background stuff while you're focusing on other things.

For example, in Starcraft, when you A-move a large part of your army across the map, an AlphaStar-like AI could decide which units to prioritize it to attack once it reaches a group of enemy units, or what formation to move in (automatically keeping colossus in back, observer ahead, balls of marines split apart into smaller clumps), or when your Terran army encounters a significant force along the way, the tanks in back could auto-siege / auto-unsiege.

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

#88

Post by Synaesthesia »

runnerup wrote: post #87
Synaesthesia wrote: post #10

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

I personally hope these types of AI's make it into future games as bots to play against in single-player, and also as an agent to run some of the background stuff while you're focusing on other things. For example, in Starcraft, when you A-move a large part of your army across the map, an AlphaStar-like AI could decide which units to prioritize it to attack once it reaches a group of enemy units, or what formation to…

That was one of the things I was hoping for, better ai, but also there are many unanswered questions in Starcraft. Like which race is the "best", what is an optimal strategy, and I thought these questions could be explored more.

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

#89

Post by EricLeer »

Jensson wrote: post #57
leereeves wrote: post #47

Earlier quoted context omitted.

Like Bell Labs of old, Google has excellent AI researchers but they haven't built many (any?) AI products people can use.

Search? Gmail completing your emails? They have lots of AI products, they just wait until they are more solid before releasing stuff. Now they will rush out a competitor to ChatGPT before it is ready, but they would use it for something given time.

Don't forget translate!

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

#90

Post by mtlmtlmtlmtl »

throwaway81523 wrote: post #84
mtlmtlmtlmtl wrote: post #53

Earlier quoted context omitted.

Stockfish' architecture is in no way revolutionised by AZ. It has gone its own way.

The current versions of Stockfish (designated NNUE) are definitely AZ influenced. In fact it was implemented in collaboration with the LCZero devs. https://stockfishchess.org/blog/2020/introducing-nnue-evalua...

The link you linked confirms what I already pointed out in another subthread. NNUE was invented by Yu Nasu for the game of Shogi. It bears very little resemblance to the AZ neural net. It's much more shallow, and inspired by work that pre-exists AZ, from the computer Shogi community. The way it trains is different, inference is different. The only commonality is the fact it's a neural net. But neural nets have been used for chess before in various capacities; Deepmind didn't invent this. All they did was throw a bunch of compute at it and do a big PR stunt then walk away.

I'm not sure where you get the LCZero connection from.

Post reply on HN