Baidu admits cheating in international supercomputer competition

A group of scientists from Baidu has apologised for cheating on a complex image-recognition test in May, leading to the Chinese technology giant being banned from similar events for the next year, according to a South China Morning Post report last week.

The report cited a statement from Dr Ren Wu, a leader of the Baidu Heterogeneous Computing team, who apologised for Baidu's "mistake" in the image-recognition contest, and added that the company would continue to review the report.

Speaking at the Embedded Vision Summit in May, Ren said Baidu is leading the race in computer intelligence, with the Chinese company scoring a record low of 4.58 percent error rate in objects classification, beating Google's 4.82 percent error rate and Microsoft's 4.9 percent, said the report.

The ImageNet Large Scale Visual Recognition Challenge (ILSVRC) is one of the largest academic challenges in computer vision worldwide, which is held every year to test state-of-the-art technology in image understanding.

According to the South China Morning Post, the test "requires a computer to classify objects from a set of 100,000 random images into 1,000 different categories". This year, "the ILSVRC organisers allowed each team to access the test database twice a week during the span of their project in order to improve the performance of the image-recognition system".

In the statement issued by Baidu, Ren admitted that its project exceeded the allowable number of weekly submissions to the ImageNet servers by over 200 times during the lifespan of the test.

A NetEase report further indicated that Baidu, in fact, submitted results over 40 times in five days of March, which also set up 30 accounts to submit more than 200 results during a six-month period.

Baidu is banned from participating in the test next year, said the NetEase report.