Machine Learning Fundamentals · 免费预览
批量分类与准确率
小批量通过一次矩阵-矩阵乘法、随后对每一行做 softmax 来完成分类，模型输出的硬判决是其概率行的 argmax。
1. Two operations over a minibatch
Two operations over a minibatch
2. From probabilities to a decision
From probabilities to a decision
3. When the argmax is not enough
When the argmax is not enough
4. Accuracy, and why it is not the objective
Accuracy, and why it is not the objective
5. Why accuracy is not the training loss
A classifier reports accuracy rather than softmax cross-entropy on its validation set. Which statement explains why cross-entropy serves as the training objective and accuracy does not?
6. The argmax of a probability row
A softmax row is for a four-class problem. The predicted class index is ____.