校准置信度
让预测概率更值得信赖
校准置信度 是 CoddyKit 上的免费 Deep Learning Academy 课时。 这是第 4 节课，共 4 节。 你可以在下方免费阅读本课时的完整内容 — 然后在浏览器中使用内置代码编辑器和全天候 AI 导师进行实践。 这是 Deep Learning Academy 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 Deep Learning Academy 课程共包含 4 节课。
本课时的部分内容尚未翻译，以英文显示。
Confidence Is a Promise
When a model says 90 percent, it should be right about 90 percent of those times. Calibration measures whether that promise holds.
Overconfident by Default
Modern deep nets often output probabilities that are too high. They sound certain even when they are frequently wrong.
Why It Matters
In medicine or finance, decisions hinge on the number, not just the label. A trustworthy probability is as vital as the prediction.
The Reliability Diagram
Plot predicted confidence against actual accuracy in bins. A perfectly calibrated model traces the diagonal line.
Expected Calibration Error
ECE sums the gap between confidence and accuracy across bins into one number. Lower means better-aligned probabilities.
Softmax Is Not Calibrated
A confident-looking softmax output is not a guarantee. The raw scores need a correction step to mean what they say.
Temperature Scaling
The simplest fix divides the logits by a learned temperature before softmax, softening overconfident scores.
probs = torch.softmax(logits / temperature, dim=-1)Fit on Validation Data
You tune the single temperature on a held-out validation set, leaving the model weights and its accuracy untouched.
Accuracy Stays the Same
Temperature scaling only rescales confidence; it never changes which class wins, so your accuracy is preserved.
Other Calibration Methods
Beyond temperature, Platt scaling and isotonic regression can map raw scores to honest probabilities too.
Check, Then Trust
Always verify calibration before you act on probabilities. A well-calibrated model lets you set thresholds with confidence. ✅
Quick Check
Recall the simplest method for fixing an overconfident neural network.
Recap
Good calibration means confidence matches accuracy. Check it with reliability diagrams and ECE, then fix it with temperature scaling. 🎯
常见问题解答
「校准置信度」课时是免费的吗？
是的 — 「校准置信度」的完整文本可在网页上免费阅读。要进行交互式练习（内置代码编辑器和全天候 AI 导师）并解锁 Deep Learning Academy 课程的其余内容，请升级到 CoddyKit PRO。 Deep Learning Academy 课程共包含 4 节课。
「校准置信度」这节课中我会学到什么？
让预测概率更值得信赖 你通过在浏览器中直接运行的动手代码来练习 Deep Learning Academy，全天候 AI 导师会在你学习这节课的过程中回答你的问题。
学习 Deep Learning Academy 需要有经验吗？
无需任何先前经验。CoddyKit 上的 Deep Learning Academy 课程适合初学者到高级学习者，你可以从这里开始或从头开始，按照自己的节奏学习。 这是第 4 节课，共 4 节。
「校准置信度」课时需要多长时间？
大多数 CoddyKit 课程大约需要 5–10 分钟。每节课都很精短且互动，所以你能稳步进步，并在网页和应用中从离开的地方继续。
我能在这节 Deep Learning Academy 课中编写并运行代码吗？
能。每节 Deep Learning Academy 课都包含内置代码编辑器，你可以在浏览器中直接编写并运行真实代码，并获得即时 AI 反馈 — 无需本地设置。