Shame ah! As the leading Internet company for cheating, he was banned for a year 5 January, Baidu supercomputer Minwa Computer vampire knight wigs Vision Challenge (ILSVRC) on made \u0026 ldquo; the world's best performance \u0026 rdquo ;, their systems in test dataset ImageNet classification error rate was 4.58%, beating the Google, Microsoft and other competitors. In the June 2, ILSVRC organizers noted that Baidu cheating during the test, far more than the prescribed number of times the frequency of testing. ImageNet allow participants to submit two sets of test results a week. However, Baidu five days in March, submitted more than 40 test results. In this regard, Baidu acknowledged violations, and has apologized. ImageNet test has been winning in artificial intelligence company Clarifai CEO MatthewZeiler represent: \u0026 ldquo; this practice is very bad, which is why the data to test on a separate server, and limit the number of visits of reasons. If you know the test data, you will revise your model parameters, how optimization can be. \u0026 Rdquo; currently, ImageNet organizers have asked Baidu to stop the next year to submit test scores ImageNet. A related report, Baidu has made the world's best performance in the benchmark test of artificial intelligence, 'Wall Street Journal' reported that Baidu has made the world's best performance in the benchmark AI (Artificial Intelligence Benchmark) test. The researchers said this is the secret weapon Baidu \u0026 ldquo; supercomputer Minwa \u0026 rdquo; have a great relationship. In the test, Minwa scanned images of a database with millions ImageNet, and \u0026 ldquo; learn \u0026 rdquo; according to the 1000 pre-defined categories of image classification. In this test, the lower the more precise identification-error rate, at present, the average person's recognition error rate of 5%. Microsoft's software is 4.94%, Google is 4.8%, while Baidu is 4.58%. It is worth mentioning that earlier Baidu in this test, the error rate was 5.98%. Compared to today, progress is relatively large. Perhaps because of this, in the next 18 months, the Baidu plans to build a computing speed more faster, more Cosplay Costumes advanced computer: a second can perform calculations 7 * 1015 supercomputer, which is sufficient among the top ten supercomputers in the world name. It is reported that Baidu's Minwa contains 36 server nodes, each node consists of two six-core Intel Xeon E5-2620 processors and four NVIDIA Tesla K40m GPU constitution. Minwa super computer has 1.7TB device memory and 6.9TB of main memory. Two reports, ILSVRC lucky star wigs organizers announce the details, Baidu apologized Beijing June 2 pm news, Baidu was traced to the existence of irregularities in the recent large-scale visual recognition Challenge (ILSVRC) in. ILSVRC organizers have announced specific details, while Baidu related team has apologized. The following is the full text of the message: ILSVRC community members: this is, after May 19, 2015 announcement of the follow-up statement, including more details and test server status. From November 28, 2014 to May 13, 2015, Baidu a team at least 30 accounts to test server was submitted at least 200 times, far more than can only be submitted twice a week limit. Which includes a high amount of a short time. For example, in the five days from March 15 to March 19 the years, more than 40 times the number submitted. Figure A shows the team associated with Baidu submit ImageNet account number. Figure B shows a cardcaptor sakura wigs comparison of these accounts and all other accounts the level of activity. Results Figure A Figure B obtained during this period in a recent report organizer of the team: ILSVRC community: Recently, ILSVRC organizers heterogeneous computing team contacted us and told us more than a week to allow the server to commit to ImageNet number of times (in our project life cycle of about 200 times). We apologize for this error, and will continue to evaluate the results. We have research paper 'Deep Image: Scaling up Image Recognition' put in a comment, and to provide relevant updates to learn more later. We support ImageNet challenging the fairness and transparency, honesty and commitment to the development of science. Wu Ren, Baidu heterogeneous computing team