课堂 Quiz 英文自测|S11-S20
现有材料范围内的非官方复习,含英文术语、成立条件与自拟练习解析。
手机上宽表格可左右滑动;自测答案默认收起,点击题目下方按钮查看。
非官方自拟练习,考试安排与范围以讲师通知为准。 先用英文作答,再逐题展开解析。
测验类型与范围边界
本课这里准备的是 In-class Quiz,不是 Midterm。现有课程信息写明三次课堂测验,共 9%,每次 3%,约 30 分钟、约 20 道二择一判断题。考试安排与范围以讲师通知为准。
Quiz 1 按课表为 10-08,现已过该日期;L03 的 p.4 公告范围 L1+L2+L3。本册保留这部分供查漏。Quiz 2 的准备材料见复习资料与英文自测 3。
文末 20 道英文判断题均为补充·自拟练习。建议每题写 T/F,再写一条理由或反例;这会比只背正误更容易发现 always、only、necessarily 等限定词中的陷阱。
复习资料 · 英文自测 1 · 英文自测 2 · 英文自测 3
英文自测
S11 · Naive assumption
Naive Bayes assumes that features are mutually independent without conditioning on the class.
S11 · 查看答案与解析
False. “The assumption is conditional independence given the class.” 漏掉 given the class 就改变了假设。
S12 · Gaussian assumption
Every Naive Bayes classifier requires all features to have Gaussian distributions.
S12 · 查看答案与解析
False. “Gaussian Naive Bayes is one variant; categorical features can use categorical distributions.” 高斯与条件独立是不同假设。
S13 · Density
A probability density evaluated at a point must lie between zero and one.
S13 · 查看答案与解析
False. 均匀分布在 [0,0.5] 上的密度为 2,积分才是 1。“A density value is not a point probability.”
S14 · Classification scores
For a fixed input, dividing all class scores by the same positive number does not change the maximizing class.
S14 · 查看答案与解析
True. “A common positive normalization factor preserves the ordering.” 前提是固定输入、公共正数;公共零分母时不能这样除。
S15 · Calibration
Normalizing Naive Bayes scores guarantees that predicted probabilities are perfectly calibrated.
S15 · 查看答案与解析
False. “Normalization makes the values sum to one, but does not guarantee calibration.” 模型假设错误等仍可能导致概率偏差。
S16 · Parameter count
A dense network of widths 3→2→1, with a bias for each non-input neuron, has 11 trainable parameters.
S16 · 查看答案与解析
True. (3+1)×2+(2+1)×1=11。若无偏置只有 8;要按题设口径计算。
S17 · Depth without nonlinearity
Composing only affine layers produces another affine transformation.
S17 · 查看答案与解析
True. W₂(W₁x+b₁)+b₂=(W₂W₁)x+(W₂b₁+b₂)。因此仅增加仿射层不提供新的非线性边界。
S18 · Chain rule
When a variable affects the loss through two paths, backpropagation should keep only the larger path contribution.
S18 · 查看答案与解析
False. “Sum the contributions from all paths.” 正负贡献甚至会抵消,取最大值不是链式法则。
S19 · BP and optimization
Backpropagation by itself specifies the learning rate and updates all parameters.
S19 · 查看答案与解析
False. “Backpropagation computes gradients; the optimizer applies an update rule.” 学习率来自优化设置,不由 BP 单独指定。
S20 · Global optimum
Training a neural network with backpropagation guarantees a global minimum of its training objective.
S20 · 查看答案与解析
False. “The objective is generally non-convex, and backpropagation is a gradient computation procedure.” 计算梯度正确不等于优化达到全局最优。
个人整理的学习笔记,不是官方材料;数字与结论以课件和讲师为准。