过拟合Overfitting

Overfitting

解决什么问题What problem it solves

把练习题背熟了，换一套新题就不会做。Memorized the practice questions, then fails a different set of questions.

这是什么What it is

模型记住了训练数据里的噪声和细节，在新数据上反而变差。

训练分数一路涨、考试分数掉头向下，是它的典型信号。

The model memorises noise and details in the training data and gets worse on new data.

The classic signal: training score keeps climbing while the exam score turns downward.

为什么需要知道Why it matters

评测造假、榜单注水多半与它有关。看到「训练集上 99 分」要警惕。Inflated benchmarks and leaderboard games are usually tied to it. Treat "99 on the training set" with suspicion.

现在能做什么What you can do now

出处Sources