| English Terminology | 中文術語 |
|---|---|
| neural networks | 神經網絡 |
| activation function | 激活函數 |
| hyperbolic tangent | 雙曲正切函數 |
| bias units | 偏置項 |
| activation | 激活值 |
| forward propagation | 前向傳播 |
| feedforward neural network | 前饋神經網絡 |
| Backpropagation Algorithm | 反向傳播算法 |
| (batch) gradient descent | （批量）梯度下降法 |
| (overall) cost function | （整體）代價函數 |
| squared-error | 方差 |
| average sum-of-squares error | 均方差 |
| regularization term | 規則化項 |
| weight decay | 權重衰減 |
| bias terms | 偏置項 |
| Bayesian regularization method | 貝葉斯規則化方法 |
| Gaussian prior | 高斯先驗概率 |
| MAP | 極大后驗估計 |
| maximum likelihood estimation | 極大似然估計 |
| activation function | 激活函數 |
| tanh function | 雙曲正切函數 |
| non-convex function | 非凸函數 |
| hidden (layer) units | 隱藏層單元 |
| symmetry breaking | 對稱失效 |
| learning rate | 學習速率 |
| forward pass | 前向傳導 |
| hypothesis | 假設值 |
| error term | 殘差 |
| weighted average | 加權平均值 |
| feedforward pass | 前饋傳導 |
| Hadamard product | 阿達馬乘積 |
| forward propagation | 前向傳播 |
| off-by-one error | 缺位錯誤 |
| bias term | 偏置項 |
| numerically checking | 數值檢驗 |
| numerical roundoff errors | 數值舍入誤差 |
| significant digits | 有效數字 |
| unrolling | 組合擴展 |
| learning rate | 學習速率 |
| Hessian matrix Hessian | 矩陣 |
| Newton's method | 牛頓法 |
| conjugate gradient | 共軛梯度 |
| step-size | 步長值 |
| Autoencoders | 自編碼算法 |
| Sparsity | 稀疏性 |
| neural networks | 神經網絡 |
| supervised learning | 監督學習 |
| unsupervised learning | 無監督學習 |
| hidden units | 隱藏神經元 |
| the pixel intensity value | 像素灰度值 |
| IID | 獨立同分布 |
| PCA | 主元分析 |
| active | 激活 |
| inactive | 抑制 |
| activation function | 激活函數 |
| activation | 激活度 |
| the average activation | 平均活躍度 |
| sparsity parameter | 稀疏性參數 |
| penalty term | 懲罰因子 |
| KL divergence | KL 散度 |
| Bernoulli random variable | 伯努利隨機變量 |
| overall cost function | 總體代價函數 |
| backpropagation | 后向傳播 |
| forward pass | 前向傳播 |
| gradient descent | 梯度下降 |
| the objective | 目標函數 |
| the derivative checking method | 梯度驗證方法 |
| Visualizing | 可視化 |
| Autoencoder | 自編碼器 |
| hidden unit | 隱藏單元 |
| non-linear feature | 非線性特征 |
| activate | 激勵 |
| trivial answer | 平凡解 |
| norm constrained | 范數約束 |
| sparse autoencoder | 稀疏自編碼器 |
| norm bounded | 有界范數 |
| input domains | 輸入域 |
| vectorization | 矢量化 |
| Logistic Regression | 邏輯回歸 |
| batch gradient ascent | 批量梯度上升法 |
| intercept term | 截距 |
| the log likelihood | 對數似然函數 |
| derivative | 導函數 |
| gradient | 梯度 |
| vectorization | 向量化 |
| forward propagation | 正向傳播 |
| backpropagation | 反向傳播 |
| training examples | 訓練樣本 |
| activation function | 激活函數 |
| sparse autoencoder | 稀疏自編碼網絡 |
| sparsity penalty | 稀疏懲罰 |
| average firing rate | 平均激活率 |
| Principal Components Analysis | 主成份分析 |
| whitening | 白化 |
| intensity | 亮度 |
| mean | 平均值 |
| variance | 方差 |
| covariance matrix | 協方差矩陣 |
| basis | 基 |
| magnitude | 幅值 |
| stationarity | 平穩性 |
| normalization | 歸一化 |
| eigenvector | 特征向量 |
| redundant | 冗余 |
| variance | 方差 |
| smoothing | 平滑 |
| dimensionality reduction | 降維 |
| regularization | 正則化 |
| reflection matrix | 反射矩陣 |
| decorrelation | 去相關 |
| Principal Components Analysis (PCA) | 主成分分析 |
| zero-mean | 均值為零 |
| mean value | 均值 |
| eigenvalue | 特征值 |
| symmetric positive semi-definite matrix | 對稱半正定矩陣 |
| numerically reliable | 數值計算上穩定 |
| sorted in decreasing order | 降序排列 |
| singular value | 奇異值 |
| singular vector | 奇異向量 |
| vectorized implementation | 向量化實現 |
| diagonal | 對角線 |
| Softmax Regression | Softmax回歸 |
| supervised learning | 有監督學習 |
| unsupervised learning | 無監督學習 |
| deep learning | 深度學習 |
| logistic regression | logistic回歸 |
| intercept term | 截距項 |
| binary classification | 二元分類 |
| class labels | 類型標記 |
| hypothesis | 估值函數/估計值 |
| cost function | 代價函數 |
| multi-class classification | 多元分類 |
| weight decay | 權重衰減 |
| self-taught learning | 自我學習/自學習 |
| unsupervised feature learning | 無監督特征學習 |
| autoencoder | 自編碼器 |
| semi-supervised learning | 半監督學習 |
| deep networks | 深層網絡 |
| fine-tune | 微調 |
| unsupervised feature learning | 非監督特征學習 |
| pre-training | 預訓練 |
| Deep Networks | 深度網絡 |
| deep neural networks | 深度神經網絡 |
| non-linear transformation | 非線性變換 |
| represent compactly | 簡潔地表達 |
| part-whole decompositions | “部分-整體”的分解 |
| parts of objects | 目標的部件 |
| highly non-convex optimization problem | 高度非凸的優化問題 |
| conjugate gradient | 共軛梯度 |
| diffusion of gradients | 梯度的彌散 |
| Greedy layer-wise training | 逐層貪婪訓練方法 |
| autoencoder | 自動編碼器 |
| Greedy layer-wise training | 逐層貪婪訓練法 |
| Stacked autoencoder | 棧式自編碼神經網絡 |
| Fine-tuning | 微調 |
| Raw inputs | 原始輸入 |
| Hierarchical grouping | 層次型分組 |
| Part-whole decomposition | 部分-整體分解 |
| First-order features | 一階特征 |
| Second-order features | 二階特征 |
| Higher-order features | 更高階特征 |
| Linear Decoders | 線性解碼器 |
| Sparse Autoencoder | 稀疏自編碼 |
| input layer | 輸入層 |
| hidden layer | 隱含層 |
| output layer | 輸出層 |
| neuron | 神經元 |
| robust | 魯棒 |
| sigmoid activation function | S型激勵函數 |
| tanh function | tanh激勵函數 |
| linear activation function | 線性激勵函數 |
| identity activation function | 恒等激勵函數 |
| hidden unit | 隱單元 |
| weight | 權重 |
| error term | 偏差項 |
| Full Connected Networks | 全連接網絡 |
| Sparse Autoencoder | 稀疏編碼 |
| Feedforward | 前向輸送 |
| Backpropagation | 反向傳播 |
| Locally Connected Networks | 部分聯通網絡 |
| Contiguous Groups | 連接區域 |
| Visual Cortex | 視覺皮層 |
| Convolution | 卷積 |
| Stationary | 固有特征 |
| Pool | 池化 |
| features | 特征 |
| example | 樣例 |
| over-fitting | 過擬合 |
| translation invariant | 平移不變性 |
| pooling | 池化 |
| extract | 提取 |
| object detection | 物體檢測 |
| DC component | 直流分量 |
| local mean subtraction | 局部均值消減 |
| sparse autoencoder | 消減歸一化 |
| rescaling | 縮放 |
| per-example mean subtraction | 逐樣本均值消減 |
| feature standardization | 特征標準化 |
| stationary | 平穩 |
| zero-mean | 零均值化 |
| low-pass filtering | 低通濾波 |
| reconstruction based models | 基于重構的模型 |
| RBMs | 受限Boltzman機 |
| k-Means | k-均值 |
| long tail | 長尾 |
| loss function | 損失函數 |
| orthogonalization | 正交化 |
| Sparse Coding | 稀疏編碼 |
| unsupervised method | 無監督學習 |
| over-complete bases | 超完備基 |
| degeneracy | 退化 |
| reconstruction term | 重構項 |
| sparsity penalty | 稀疏懲罰項 |
| norm | 范式 |
| generative model | 生成模型 |
| linear superposition | 線性疊加 |
| additive noise | 加性噪聲 |
| basis feature vectors | 特征基向量 |
| the empirical distribution | 經驗分布函數 |
| the log-likelihood | 對數似然函數 |
| Gaussian white noise | 高斯白噪音 |
| the prior distribution | 先驗分布 |
| prior probability | 先驗概率 |
| source features | 源特征 |
| the energy function | 能量函數 |
| regularized | 正則化 |
| least squares | 最小二乘法 |
| convex optimization software | 凸優化軟件 |
| conjugate gradient methods | 共軛梯度法 |
| quadratic constraints | 二次約束 |
| the Lagrange dual | 拉格朗日對偶函數 |
| feedforward architectures | 前饋結構算法 |
| Independent Component Analysis | 獨立成分分析 |
| Over-complete basis | 超完備基 |
| Orthonormal basis | 標準正交基 |
| Sparsity penalty | 稀疏懲罰項 |
| Under-complete basis | 不完備基 |
| Line-search algorithm | 線搜索算法 |
| Topographic cost term | 拓撲代價項 |
來源：國際電子商情
