Baidu Fesses Up To Cheating In Global Supercomputer Competition

We've heard of benchmarking scandals before, but usually they involve gaming benchmarks and tweaked drivers that run afoul of the rules to gain a competitive advantage. This time, however, it's Chinese search engine Baidu that's in hot water after it was discovered that its supercomputer cheated in a major artificial intelligence competition.

Prior to being caught, Dr. Ren Wu, head of the Baidu Heterogeneous Computing team, had boasted that his company was the top dog in computer intelligence. "We have great power in our hands -- much greater than our competitors," Dr. Wu said. The competitors he speaks of include both Google and Microsoft, a couple of tech titans based in the U.S.

Baidu's bad behavior involved the ImageNet large scale visual recognition competition (ILSVRC), which is one of the biggest AI competitions in the world. It's an annual event hosted by Stanford University, the University of North Carolina at Chapel Hill, and the University of Michigan.

Baidu Building

The test is a pretty intense image recognition evaluation that tasks computer systems with classifying objects from a set of 100,000 random images into 1,000 different categories. Per the rules for this year's competition, each team was allowed access to the database twice a week so that they could tweak their image recognition algorithm.

Baidu won the competition, but it later came to light that it was using several different accounts, which allowed the team to access the database more than 200 times in six months. At one point, it made over 40 entries in a span of five days, way over the four that was allowed.

Dr. Wu issued an apology for what he deemed was a "mistake" and said that his teams continues "to review the data." In the meantime, Baidu has been banned from participating in similar events by the organizers.

Show comments blog comments powered by Disqus