常用AI术语对照 – TopQuant极宽量化

常用AI术语对照
zw-ps: 已按首字母排序

accuracy,准确率/精度
activation function,激活函数
approximate method,近似法
attention cue,注意力提示
attention mechanism,注意力机制
attention model,注意力模型
baseline,基准
batch,批量
bias,偏置
bidirectional recurrent neural network,双向循环神经网络
bigram,二元
binary classification,二元分类
block,块
broadcast,广播
channel,通道
class,类
classification,分类
classifier,分类器
collaborative filtering,协同过滤
concatenate,连结
context variable,上下文变量
context vector,上下文向量
context window,上下文窗口
context word,上下文词
continous bag-of-words (CBOW),连续词袋
converge,收敛
convex optimization,凸优化
convolutional kernel,卷积核
convolutional neural network,卷积神经网络
co-occurrence,共现
corpus,语料库
cost,成本
cross-entropy,交叉熵
data example,数据样例
data instance,数据示例
dataset,数据集
decision boundary,决策边界
decoder,解码器
dense,稠密
diverge,发散
dropout,暂退法
eigenvalue,特征值
empirical risk minimization,经验风险最小化
encoder,编码器
end-to-end,端到端
epoch,轮
error,误差
example,样例
feature,特征
feature map,特征图
filter,滤波器
fully-connected layer,全连接层
gated recurrent units (GRU),门控循环单元
generalization,泛化
generalization error,泛化误差
gradient,梯度
gradient clipping,梯度裁剪
gradient descent in one-dimensional space,一维梯度下降
Gram matrix,格拉姆矩阵
ground truth,真实值
hidden layer,隐藏层
hidden variable,隐藏变量
hierarchical softmax,层序softmax
hyperparameter,超参数
hypothesis,假设
identity mapping,恒等映射
image,图像
import,导入
independent and identically distributed (i.i.d.),独立同分布
instance,实例/示例
kernel regression,核回归
label,标签
Laplace smoothing,拉普拉斯平滑
learning rate,学习率
linear model,线性模型
linear regression,线性回归
local minimum,局部最小值
long short-term memory (LSTM),长短期记忆网络
loss function,损失函数
mean,均值
mean squared error,均方误差
metric,指标
minibatch,小批量
minibatch gradient,小批量梯度
momentum (method),动量法
multi-head attention,多头注意力
multilayer perceptron,多层感知机
negative sampling,负采样
normalization,规范化
numerical method,数值方法
object detection,目标检测
offset,偏移量
operator,运算符
optimizer,优化器
overfitting,过拟合
padding,填充
parameter,参数
partial derivative,偏导数
perplexity,困惑度
pipeline,流水线
pooling layer,汇聚层
precision,查准率
property,属性
random sampling,随机采样
receptive field,感受野
recurrent neural network (RNN),循环神经网络
saddle point,鞍点
self-attention,自注意力
sentiment analysis,情感分析
sequential partitioning,顺序分区
size,大小
skip-gram model,跳元模型
softmax regression,softmax回归
stop words,停用词
stride,步幅
subsample,子采样
support vector machine,支持向量机
token,词元
tokenize,词元化
tokenizer,词元分析器
training error,训练误差
transform,变换
trigram,三元
tune hyperparameter,调参
unbiased estimate,无偏估计
underfitting,欠拟合
uniform sampling,均匀采样
unigram,一元
unknown token,未知词元
upsample,上采样
vanishing gradients,梯度消失
variance,方差
vocabulary,词表
weight,权重
word embedding,词嵌入
word vector,词向量
zero tensor,零张量
Zipf’s law,齐普夫定律