Life In 19x19

But transfer learning is all about using a trained NN for related new tasks (e.g. play go under another board size, ruleset or komi), while I'm talking of expanding a NN for the same task.

From AlphaGo Zero paper, one can see that less "blocks", i.e. a shallower neural network, leads to a faster learning process but plateau faster too. Later on I read somewhere about the nature of the residual network in use and notice one thing: by default a residual "block" will copy data from ...

To be honest I'm not a fan of the button (complications for little benefit, since the result is still not as nice as Japanese). But I think I see where it's problems originate from. From a fairness perspective, shouldn't it go like this? (complementing my earlier list )
"Fair" button : Less ...

I said "side-effect" because the gameplay deviate from the 'optimal jigo sequence', telling God to insert a premature pass / take button before all stones played out in order to score a win. The intended effect, I assume, is to avoid the situation of mainstream fractional komi that make all 'optimal ...

The actual "minuses" of button go / first passer win jigo now looks like the side-effect of players may in some rare cases, as illustrated above, want to pass or take button before everything has been played out, thus changed the nominal play sequence. The result looks fair but the gameplay looks ...

Apparently, 7x7 Go has been weakly solved , but the article is in Chinese and I can't make heads or tails of it (Google Translate doesn't help).
Some other points made in the Chinese article (not in order):

⋅ The need of komi stems from existence of tengen, as for a board without ...

[*]For Li Zhe's 7x7 optimal solutions, the first three moves are unique; the first seven moves branch into the five core gameplay sequences in illustration, each with irrelevant minor variations later on of course
Is there a link to the actual solutions?
They are posted in the Chinese article ...

Apparently, 7x7 Go has been weakly solved , but the article is in Chinese and I can't make heads or tails of it (Google Translate doesn't help).
Some other points made in the Chinese article (not in order):

⋅ The need of komi stems from existence of tengen, as for a board without ...

Some go knowledge was involved, but indirectly. A winner is determined (by a simple, non-ML portion of the program) at the terminal state of each game. For training the network, the position on the board is given, along with a history of recent moves, and an estimated winning probability given ...

Button go, anyone?
I've mentioned this multiple times in this thread already. It would be the perfect solution, but no one seems excited about it.
I guess button go being unpopular is because upon introduction of a new object "button", people think of the game as a go-variant, not seeing it a ...

Herman's explanation further strengthen my suspicion that 6.5 komi may be a reasonable komi value for area scoring, while Luigi's explanation repeated what I learnt from the Sensei wiki I linked.

6.5 komi may be less "fair" than 7.5 komi, maybe not. But assuming 6.5 komi is exactly as fair/unfair ...

Unfortunately for China, it doesn't make sense to use 6.5 komi under Chinese counting rule.
The rationale of Chinese counting rule can't use 6.5 komi never make sense to me. Even after reading https://senseis.xmp.net/?WhyChineseKomiShouldIncreaseBy2PointSteps , although it does ...

And 6.5 komi under Japanese rules is almost equivalent to 7 komi plus button under Chinese rules. The message is clear...
Agree, but I'm still standing my statement.
In Chinese rule both statistics and AI agreed that white has advantage with 7.5 komi.
In Japanese rule DeepZen also agree that 6.5 ...

I don't think there will be any 100% "fair" value for komi as long as we make it tie-breaking, no matter it is breaking by fraction value or by extra rules.

Let's assume two Gods play Go in a minimax manner (because Gods have infinite computational power so they can brute-force the whole game tree ...

A better yet about-the-same-difficulty-to-code algorithm will be let opponent play inside first. Then this side try to capture them.