First of all, data relations

Because the data used by the author is relatively private, most people can't get it. In this case, even if the author provides the source code, the reader can't get the data, so the algorithm can't be reproduced.

This situation is very common in domestic academic circles, and the data can't be found by others. Just like an Olympic math teacher, he works out an Olympic math problem by himself, solves it by himself, and then writes a paper on the problem-solving process. This kind of paper is often not convincing enough and the story is not strong enough.

Second, the hardware reasons

Many algorithms of deep learning are miraculously created. For example, some algorithms of Google and facebook are trained by powerful hardware.

Ordinary researchers don't have such powerful hardware resources, and it is estimated that they can't reach their 1% computing power, so they can't reproduce the algorithm at all.

Third, data division and training methods.

Some papers disclose code and data at the same time, but the problem of data division is not mentioned in the paper. If there is less data, different division will lead to different results.

Fourth, well-known reasons.

Everyone knows this truth, so I won't say that I understand it too well. This situation appears in the papers of many domestic authors. This is rare in public data.