深度学习词汇 - 中英对照词典(内容由AI生成) Deep Learning Vocabulary - English-Chinese Dictionary

Deep Learning Vocabulary - English-Chinese Dictionary

深度学习词汇 - 中英对照词典

Created: 2025-12-08 22:18:14


Core Concepts / 核心概念

1. normalization ˌnɔːrməlaɪˈzeɪʃən

中文: 归一化,标准化
Definition: The process of scaling data to a standard range or distribution
例句 / Example:

  • Feature normalization helps stabilize training. (特征归一化有助于稳定训练)
  • We apply z-score normalization to the input data. (我们对输入数据应用 z-score 标准化)

2. normalized ˈnɔːrməlaɪzd

中文: 归一化的,标准化的
Definition: Data that has been scaled to a standard range
例句 / Example:

  • The normalized features have zero mean and unit variance. (归一化后的特征具有零均值和单位方差)
  • Normalized data converges faster in gradient descent. (归一化数据在梯度下降中收敛更快)

3. deep learning diːp ˈlɜːrnɪŋ

中文: 深度学习
Definition: A subset of machine learning using neural networks with multiple layers
例句 / Example:

  • Deep learning models require large amounts of data. (深度学习模型需要大量数据)
  • Normalization is crucial in deep learning contexts. (在深度学习场景中,归一化至关重要)

4. feature ˈfiːtʃər

中文: 特征
Definition: An individual measurable property or characteristic of a phenomenon
例句 / Example:

  • Each feature should be normalized independently. (每个特征应独立归一化)
  • The model uses two features with very different scales. (模型使用两个量级差异很大的特征)

5. scaling ˈskeɪlɪŋ

中文: 缩放
Definition: The process of adjusting the range of values
例句 / Example:

  • Feature scaling prevents one feature from dominating. (特征缩放防止某个特征主导)
  • Min-max scaling maps values to 0, 1 range. (最小最大缩放将值映射到 0, 1 范围)

Statistical Terms / 统计术语

6. mean miːn

中文: 均值,平均值
Definition: The average value of a set of numbers
例句 / Example:

  • Zero mean normalization subtracts the mean from each value. (零均值归一化从每个值中减去均值)
  • The mean of normalized features should be approximately zero. (归一化特征的均值应接近零)

7. variance ˈveriəns

中文: 方差
Definition: A measure of how spread out values are from the mean
例句 / Example:

  • Unit variance means the standard deviation equals 1. (单位方差意味着标准差等于 1)
  • Normalization adjusts both mean and variance. (归一化同时调整均值和方差)

8. standard deviation (std) ˈstændərd ˌdiːviˈeɪʃən

中文: 标准差
Definition: The square root of variance, measuring data dispersion
例句 / Example:

  • We divide by standard deviation in z-score normalization. (在 z-score 归一化中,我们除以标准差)
  • The std of normalized data should be close to 1. (归一化数据的标准差应接近 1)

9. zero mean ˈzɪroʊ miːn

中文: 零均值
Definition: Having an average value of zero
例句 / Example:

  • Z-score normalization produces zero mean features. (Z-score 归一化产生零均值特征)
  • Zero mean helps center the data distribution. (零均值有助于将数据分布居中)

10. unit variance ˈjuːnɪt ˈveriəns

中文: 单位方差
Definition: Variance equal to 1 (standard deviation = 1)
例句 / Example:

  • Normalized features have unit variance. (归一化特征具有单位方差)
  • Unit variance ensures consistent feature scales. (单位方差确保特征尺度一致)

Optimization Terms / 优化术语

11. gradient ˈɡreɪdiənt

中文: 梯度
Definition: The vector of partial derivatives indicating the direction of steepest ascent
例句 / Example:

  • Normalization stabilizes gradients during training. (归一化在训练过程中稳定梯度)
  • Large gradients can cause training instability. (大梯度可能导致训练不稳定)

12. gradient descent ˈɡreɪdiənt dɪˈsent

中文: 梯度下降
Definition: An optimization algorithm that minimizes loss by following gradients
例句 / Example:

  • Gradient descent converges faster with normalized features. (使用归一化特征时,梯度下降收敛更快)
  • We use gradient descent to train the regression model. (我们使用梯度下降训练回归模型)

13. convergence kənˈvɜːrdʒəns

中文: 收敛
Definition: The process of approaching a stable solution
例句 / Example:

  • Normalization speeds up convergence. (归一化加速收敛)
  • The model shows faster convergence with normalized inputs. (使用归一化输入时,模型收敛更快)

14. learning rate ˈlɜːrnɪŋ reɪt

中文: 学习率
Definition: A hyperparameter controlling the step size in optimization
例句 / Example:

  • Normalization allows larger learning rates. (归一化允许使用更大的学习率)
  • Adjust the learning rate to see stability differences. (调整学习率以观察稳定性差异)

15. stabilize ˈsteɪbəlaɪz

中文: 稳定
Definition: To make something steady or consistent
例句 / Example:

  • Normalization helps stabilize gradients. (归一化有助于稳定梯度)
  • Feature scaling stabilizes the training process. (特征缩放稳定训练过程)

Loss & Evaluation / 损失与评估

16. loss lɔːs

中文: 损失
Definition: A measure of how far predictions are from actual values
例句 / Example:

  • The final loss without normalization was 13822. (未归一化的最终损失为 13822)
  • Normalization significantly reduces the loss. (归一化显著降低损失)

17. loss landscape lɔːs ˈlændskeɪp

中文: 损失景观
Definition: The shape of the loss function in parameter space
例句 / Example:

  • Normalization creates a smoother loss landscape. (归一化创建更平滑的损失景观)
  • A smooth loss landscape enables better optimization. (平滑的损失景观有助于更好的优化)

Normalization Methods / 归一化方法

18. z-score ziː skɔːr

中文: Z 分数,标准分数
Definition: A normalization method: (x - mean) / std
例句 / Example:

  • Z-score normalization subtracts mean and divides by std. (Z-score 归一化减去均值并除以标准差)
  • We apply z-score normalization to the input features. (我们对输入特征应用 z-score 归一化)

19. min-max scaling mɪn mæks ˈskeɪlɪŋ

中文: 最小最大缩放
Definition: Scaling data to a fixed range, typically 0, 1
例句 / Example:

  • Min-max scaling maps values to 0, 1 range. (最小最大缩放将值映射到 0, 1 范围)
  • Min-max scaling is an alternative to z-score normalization. (最小最大缩放是 z-score 归一化的替代方法)

20. per-channel pər ˈtʃænəl

中文: 按通道的
Definition: Applied separately to each channel (e.g., RGB channels)
例句 / Example:

  • Per-channel image normalization processes each color channel separately. (按通道图像归一化分别处理每个颜色通道)
  • We use per-channel normalization for image data. (我们对图像数据使用按通道归一化)

21. image normalization ˈɪmɪdʒ ˌnɔːrməlaɪˈzeɪʃən

中文: 图像归一化
Definition: Normalizing pixel values in images
例句 / Example:

  • Image normalization is essential for CNN training. (图像归一化对 CNN 训练至关重要)
  • Per-channel image normalization improves model performance. (按通道图像归一化提高模型性能)

Data & Model Terms / 数据与模型术语

22. regression rɪˈɡreʃən

中文: 回归
Definition: A supervised learning task predicting continuous values
例句 / Example:

  • We generate synthetic two-feature regression data. (我们生成合成的双特征回归数据)
  • Linear regression benefits from feature normalization. (线性回归受益于特征归一化)

23. synthetic sɪnˈθetɪk

中文: 合成的,人工的
Definition: Artificially created rather than naturally occurring
例句 / Example:

  • We use synthetic data for demonstration purposes. (我们使用合成数据进行演示)
  • Synthetic datasets help test normalization effects. (合成数据集有助于测试归一化效果)

24. magnitude ˈmæɡnɪtuːd

中文: 量级,大小
Definition: The size or scale of something
例句 / Example:

  • Features should have comparable magnitude after normalization. (归一化后,特征应具有可比较的量级)
  • Different feature magnitudes can cause training issues. (不同的特征量级可能导致训练问题)

25. feature dominance ˈfiːtʃər ˈdɑːmɪnəns

中文: 特征主导
Definition: When one feature overshadows others due to scale differences
例句 / Example:

  • Normalization reduces feature dominance. (归一化减少特征主导)
  • Feature dominance can bias the model's predictions. (特征主导可能使模型预测产生偏差)

Training Terms / 训练术语

26. trace treɪs

中文: 轨迹,追踪
Definition: A record of values over time (e.g., loss trace)
例句 / Example:

  • We compare gradient descent loss traces with/without normalization. (我们比较归一化前后的梯度下降损失轨迹)
  • The loss trace shows faster convergence with normalization. (损失轨迹显示归一化后收敛更快)

27. stability stəˈbɪlɪti

中文: 稳定性
Definition: The quality of being steady and consistent
例句 / Example:

  • Normalization improves training stability. (归一化提高训练稳定性)
  • Plotting losses helps visualize stability differences. (绘制损失有助于可视化稳定性差异)

28. step step

中文: 步骤,步数
Definition: A single iteration in the training process
例句 / Example:

  • The first 5 steps show rapid loss reduction. (前 5 步显示损失快速下降)
  • Each training step updates the model parameters. (每个训练步骤更新模型参数)

Additional Terms / 附加术语

29. input ˈɪnpʊt

中文: 输入
Definition: Data fed into a model
例句 / Example:

  • Normalization scales inputs to comparable ranges. (归一化将输入缩放到可比较的范围)
  • Raw inputs often need normalization before training. (原始输入在训练前通常需要归一化)

30. comparable ˈkɑːmpərəbəl

中文: 可比较的
Definition: Similar enough to be compared fairly
例句 / Example:

  • Normalization makes features have comparable magnitude. (归一化使特征具有可比较的量级)
  • Features with comparable scales train more efficiently. (具有可比较尺度的特征训练更高效)

Quick Reference / 快速参考

English 中文 Phonetic Key Usage
normalization 归一化/标准化 ˌnɔːrməlaɪˈzeɪʃən Feature scaling technique
gradient 梯度 ˈɡreɪdiənt Optimization direction
convergence 收敛 kənˈvɜːrdʒəns Training progress
learning rate 学习率 ˈlɜːrnɪŋ reɪt Step size parameter
loss 损失 lɔːs Error measure
z-score Z 分数 ziː skɔːr Normalization method
variance 方差 ˈveriəns Data spread measure
feature 特征 ˈfiːtʃər Input variable

Study Tips / 学习建议

  1. Practice with examples: Use the example sentences to understand context. (通过例句理解语境)
  2. Focus on pronunciation: Pay attention to phonetic symbols for key terms. (关注重点词汇的音标)
  3. Connect concepts: Understand how normalization relates to gradients and convergence. (理解归一化与梯度、收敛的关系)
  4. Read code comments: Look at normalize_demo.py for practical usage. (查看代码注释了解实际用法)

Last Updated: 2025-12-08 22:18:14

相关推荐
火山引擎开发者社区4 小时前
没有长期记忆,Agent 谈何持续进化?一图看懂火山 Mem0:解锁 Agent 持续学习与进化之路
人工智能
冬奇Lab8 小时前
Workflow 系列(06):安全——跨步骤注入传播与四层防御
人工智能·工作流引擎
冬奇Lab8 小时前
每日一个开源项目(第149篇):RAG-Anything - 把图片、表格、公式当成一等公民的多模态 RAG 框架
人工智能·开源
米小虾8 小时前
AI Agent 安全实战指南:当智能体开始"不听话",开发者该如何应对?
人工智能·安全·agent
IT_陈寒10 小时前
Vite的热更新突然不香了,排查三小时差点砸键盘
前端·人工智能·后端
阿里云大数据AI技术12 小时前
构建高转化海外电商搜索:阿里云OpenSearch行业算法版的全链路智能优化策略实战
人工智能·搜索引擎
Awu122712 小时前
⚡从零开发 Agent CLI(五)实现一个可治理、可扩展的工具系统
前端·人工智能·claude
字节跳动视频云技术团队12 小时前
让 Agent 成为音视频工作台:AI MediaKit CLI + Skill 发布
人工智能·音视频开发
魏祖潇12 小时前
framework 整合实战——DDD/TDD/SDD 三件套在 framework 仓的真实落地
人工智能·后端