1, experimental reproduction: this is the most direct method and the most respected method. The author disclosed the source code and data, and other researchers can reproduce the experimental results. If it can't be reproduced, there may be a problem. However, the disadvantage of this method is time-consuming and high cost, because each experiment needs to be repeated.

2. Comparison with other papers: If other papers use similar data sets and methods, the results can be compared and verified. But the premise of this method is that the results of other papers are reliable.

3. Use multiple data sets: If the paper uses multiple data sets, it can be verified by comparing the results of these data sets. The advantage of this method is that it can reduce the deviation caused by specific data sets.

4. Visualize the results: It is easier to find abnormal values or abnormal values by visualizing the results. This method is helpful to find possible problems in the data.

5. Use external benchmarks: If external benchmarks exist, they can be used for verification. But the premise of this method is that the external reference is reliable.

6. Data quality check: For example, check the noise level and abnormal values in the data set. This helps to find possible problems in the data.