Archive for March, 2016

The non-existence of Android backup and restore

People change phones. They want their programs and data to show up on their new phones. Apple has solved this problem. Somehow, Google has not. As explained here, settings can be synced through Google’s sync API. It is however useless except for Google’s own apps and whoever uses their API (maybe nobody).

But as far as full-system backup and restore options go, you either have to root your system and use Nandroid or Titanium to backup to the phone storage itself, or you have to rely on adb backup. adb is Android Device Bridge, something that is accessed through the Developer Mode on Android. It feels like Google has given up on this feature midway through and just left it flopping around, because it simply does not work. Although I haven’t had trouble getting backup to work (‘adb backup -apk -shared -all -f [file]‘), I could not get restore (‘adb restore [file]‘) to work automatically because of this bug (which incidentally is also obsoleted).
(Read the article)

Google marks bugs it doesn’t want to deal with as ‘obsolete’

Apparently Google has a habit of marking bugs it doesn’t want to deal with as obsolete. According to Google’s own Life of a Bug explainer, a bug categorized as “Obsolete” means,

Obsolete: Similar to Unreproducible, but with a reasonable certainty that the bug did exist in the reported version but was already fixed in a later release.

…which is a lie. I give two pieces of solid evidence for why this is a lie.
(Read the article)

The AlphaGo paper is a topological counting game that has been challenging for a computer to play due to its global phase-transition like game dynamics, requiring accurate reads of position early in the game. When the AlphaGo paper (preprint) first came out in January purporting a new system that plays Go at a human-professional level, I didn’t have a chance to read it thoroughly, partly because the paper is badly written (yes, I said it), and partly because it was a lot of information. Now that the DeepMind Challenge Match against Lee Sedol is happening, with a surprise nearly every game, I thought to give it another go (no pun intended).

The main objective is to extract from the paper four key pieces of information: (1) How did computers play Go before, and what was the state-of-the-art before AlphaGo? (2) What is the novelty of AlphaGo from which it derives its improved performance? (3) How is AlphaGo actually implemented? (4) What are the future and the limitations of this approach?
(Read the article)