Hacker News Reader

Comment by travelhead | Hacker News Reader

What's more important, winning a competition or improving AI for the entire world? Do the ends justify the means? I Don't think we should be too hard on Baidu, considering they are attempting to improve their algorithm for the interest of humanity (or evil AI that will bring an end to mankind, depending on how you look at it).

View on HN

> I Don't think we should be too hard on Baidu, considering they are attempting to improve their algorithm for the interest of humanity

They are not improving the algorithm for the interest of humanity. They submitted and obtained test results much more frequently than allowed. With that information, they can tune parameters to more closely fit the test data.

Basically it is like scoring higher on an exam where some of the questions have been leaked. It does not suggest you now have a better understanding of the subject.

If they are building an algorithm which overfits to training data, their algorithm will lose performance on in the actual validation test.

It is NOT like looking at leaked answers. It's like re-reading the textbook more than everyone else. Definitely an unfair advantage however.

You misunderstand. They're using repeated attempts on the test set to improve their network. That is equivalent to training on the test set and that is unequivocably cheating.

Their actions were motivated to win the competition, not improve AI in general. There are many image sets the Baidu team could have used to test on. They used the competition's image set an unfair number of times to tweak their computation to score better on those specific images (not to improve it overall).

If those two motivations are not aligned then the competition is poorly designed.

They could have used an alternative data set as I assume all the other competitors are doing.

That argument might make sense if this competition was the only avenue in the world for improving AI.

Reply on news.ycombinator.com