[科研] | 101级别解释 | Benchmark

转载已于 2023-03-06 18:08:34 修改 · 225 阅读

CC 4.0 BY-SA版权

文章标签：

于 2023-01-21 23:26:44 首次发布

文章讨论了在机器学习中，基准测试是衡量解决方案性能的标准，主要关注训练数据量、测试数据以及相对于基准解决方案的准确性。为了超越基准，可能需要采用更复杂的算法或全局视角来提出更好的解决方案。

1. 直觉解释

Benchmark is standard against which you compare the solutions, to get a feel if the solutions are better or worse.

Benchmarking means a standard solution which already performs well. What would be the factors on which your solution will be tested? It’s mostly going to be on given the amount of training/test data, what is the accuracy with which your solutions is performing, as opposed to the benchmarked solution.
Of course they are not going to give it away, so that you can analyse and come up with something better. You just have to come up with a good solution and hope it works better than benchmarked solution. This can be achieved by implementing more sophisticated algorithms, or view things more holistically.