AI Glossary | 人工智能詞匯表

A

Algorithm (算法)

English: A set of step-by-step instructions or rules designed to perform a specific task. In AI, algorithms process data and learn patterns that enable decision-making or predictions.
中文：一組分步驟的指令或規則，用於執行特定任務。在人工智能中，算法對數據進行處理並學習模式，從而實現決策或預測。

Artificial Intelligence (AI) (人工智能)

English: A field of computer science focused on creating machines or software capable of exhibiting human-like intelligence—such as reasoning, learning, and problem-solving.
中文：計算機科學的一個領域，旨在創建能夠表現出類似人類智能（如推理、學習和解決問題）能力的機器或軟件。

Artificial Neural Network (ANN) (人工神經網絡)

English: A computational model inspired by biological neural networks in the human brain. ANNs are composed of interconnected nodes (“neurons”) that process information in layers to learn patterns.
中文：一種受人類大腦生物神經網絡啟發的計算模型。人工神經網絡由相互連接的節點（“神經元”）組成，通過分層處理信息來學習模式。

Artificial general intelligence (AGI) is a type of artificial intelligence (AI) that matches or surpasses human cognitive capabilities across a wide range of cognitive tasks. This contrasts with narrow AI, which is limited to specific tasks.^[1] Artificial superintelligence (ASI), on the other hand, refers to AGI that greatly exceeds human cognitive capabilities. AGI is considered one of the definitions of strong AI.

Agentic AI

Agentic AI uses sophisticated reasoning and iterative planning to autonomously solve complex, multi-step problems.

---

B

Backpropagation (反向傳播)

English: An algorithm for training neural networks by propagating the error backward from the output layer to the input layer, adjusting the weights to minimize error.
中文：一種用於訓練神經網絡的算法，通過將誤差從輸出層向輸入層反向傳播，並調整權重以最小化誤差。

Batch (批次)

English: A subset of data used to train a model at one time. In mini-batch gradient descent, the training set is divided into small batches for updates to model parameters.
中文：一次用於訓練模型的數據子集。在小批量梯度下降中，訓練集被分割成小批次，以便對模型參數進行更新。

Bias (偏差)

1. Model bias (模型偏差): Systematic error that causes a model to consistently deviate from the true values.
   中文：導致模型持續偏離真實值的係統性誤差。

2. Bias term (in neurons) (神經元中的偏置項): A constant value added to the weighted input of a neuron to shift the activation function.
   中文：加到神經元加權輸入中的一個常數值，用於對激活函數進行平移。

---

C

Classification (分類)

English: A supervised learning task where the goal is to predict a discrete label (e.g., “spam” vs. “not spam” or identifying which category an image belongs to).
中文：一種監督學習任務，目標是預測離散標簽（例如，“垃圾郵件”與“非垃圾郵件”或識別圖像所屬的類別）。

Clustering (聚類)

English: An unsupervised learning task that involves grouping data points such that points in the same group (cluster) are more similar to each other than to those in other clusters.
中文：一種無監督學習任務，將數據點進行分組，使同一組（簇）內的點彼此之間比與其他簇的點更加相似。

Convolutional Neural Network (CNN) (卷積神經網絡)

English: A type of neural network especially suited for tasks involving image and spatial data. It uses filters or “kernels” to detect patterns such as edges and textures in images.
中文：一種特別適合處理圖像和空間數據的神經網絡。它使用濾波器或“卷積核”來檢測圖像中的邊緣、紋理等模式。

 

Cosmos/WFM model

Cosmos is a family of world foundation models(WFM), or neural networks that can predict and generate physics-aware virtual environments. Speaking at the headline keynote at CES 2025, Huang said Nvidia Cosmos models will be open sourced. "The ChatGPT moment for robotics is coming. Like large language models, world foundation models are fundamental to advancing robot and AV development," said Huang. "We created Cosmos to democratize physical AI and put general robotics in reach of every developer."

world foundation models (WFMs) will be as important as large language models, but physical AI developers have been underserved. WFMs will use data, text, images, video and movement to generate and simulate virtual worlds that accurately models environments and physical interactions. Nvidia said 1X, Agility Robotics and XPENG, and autonomous developers Uber, Waabi and Wayve are among the companies using Cosmos.

 

---

D

Data Mining (數據挖掘)

English: The process of discovering patterns and insights from large data sets, using methods like clustering, classification, and association rule mining.
中文：從大型數據集中發現模式和洞見的過程，通常使用聚類、分類和關聯規則挖掘等方法。

Dataset (數據集)

English: A collection of data points or examples used to train, validate, and/or test machine learning models.
中文：用於訓練、驗證和/或測試機器學習模型的一組數據點或示例的集合。

Deep Learning (深度學習)

English: A subset of machine learning that involves neural networks with multiple layers (deep neural networks). These deeper architectures can automatically learn high-level representations from data.
中文：機器學習的一個分支，使用具有多層結構的神經網絡（深度神經網絡）。這些深層架構能夠自動從數據中學習高層次表示。

Dimensionality Reduction (降維)

English: Techniques like Principal Component Analysis (PCA) or t-SNE that reduce the number of input features while preserving as much information as possible.
中文：例如主成分分析（PCA）或 t-SNE 等技術，用於在盡可能保留信息的前提下減少輸入特征的數量。

---

E

Epoch (輪次)

English: One complete pass through the entire training dataset during the training process of a machine learning model.
中文：在機器學習模型的訓練過程中，對整個訓練數據集進行一次完整遍曆的過程。

Evaluation Metrics (評估指標)

English: Methods or formulas used to measure the performance of a model. Examples include accuracy, precision, recall, F1 score, and ROC AUC.
中文：用於衡量模型性能的方法或公式。例如準確率（accuracy）、精確率（precision）、召回率（recall）、F1 值（F1 score）以及 ROC AUC。

---

F

Feature (特征)

English: An individual measurable property or characteristic of a phenomenon being observed (e.g., the pixel intensity in an image, or the temperature reading in a climate dataset).
中文：對所觀察現象的可測量屬性或特性（例如圖像中的像素強度，或氣候數據集中的溫度讀數）。

Feature Engineering (特征工程)

English: The process of using domain knowledge to create new features from existing raw data to improve model performance.
中文：運用領域知識從原始數據中創建新特征以提升模型性能的過程。

Fine-tuning (微調)

English: The process of taking a pretrained model (often on a large dataset) and adjusting its parameters on a new, related task or dataset.
中文：對已經在大型數據集上預訓練過的模型進行參數調整，以適應新的、相關任務或數據集的過程。

---

G

Generalization (泛化)

English: A model’s ability to perform accurately on new, unseen data, rather than just on the training data.
中文：指模型在未見過的新數據上仍能準確表現的能力，而不僅僅是在訓練數據上表現良好。

Generative Adversarial Network (GAN) (生成對抗網絡)

English: A system of two neural networks: a generator that creates synthetic data and a discriminator that tries to distinguish real data from fake. They train each other in a competitive process.
中文：由兩個神經網絡組成的係統：一個生成器用於生成合成數據，另一個判別器用於區分真實數據與虛假數據。它們在競爭的過程中相互訓練。

Gradient Descent (梯度下降)

English: An optimization algorithm that iteratively adjusts parameters to minimize a loss function by moving in the direction of the steepest descent in the parameter space.
中文：一種優化算法，通過在參數空間沿最陡下降方向迭代調整參數，來最小化損失函數。

Generative AI?

Generative AI enables users to quickly generate new content based on a variety of inputs. Inputs and outputs to these models can include text, images, sounds, animation, 3D models, or other types of data.

---

H

Hyperparameters (超參數)

English: Configuration settings external to a model that are not learned from the training data (e.g., learning rate, number of layers, or number of neurons in each layer).
中文：模型外部的配置設置，不從訓練數據中學習（例如學習率、層數或每層神經元的數量）。

Heuristic (啟發式方法)

English: A practical method or approach to problem-solving that isn’t guaranteed to be optimal but is fast and good enough for many situations.
中文：一種解決問題的實用方法或策略，無法保證絕對最優，但在很多情況下速度快且效果足夠好。

---

I

Inference (推理)

English: The process of using a trained model to make predictions or decisions on new, unseen data.
中文：使用訓練好的模型對新的、未見過的數據進行預測或決策的過程。

Interpretability (可解釋性)

English: Refers to how understandable the decisions or reasoning of a model are to humans. Interpretable models help users trust the outputs.
中文：指模型的決策或推理對於人類而言的可理解程度。可解釋的模型能幫助用戶信任其輸出。

---

L

Label (標簽)

English: The correct answer or category assigned to a data point in supervised learning (e.g., the numeric value in a regression problem or class category in classification).
中文：在監督學習中分配給數據點的正確答案或類別（例如回歸問題中的數值，或分類中的類別標簽）。

Latent Space (潛在空間)

English: A compressed, abstract representation of the data typically learned by deep neural networks. It captures the underlying patterns and structure in a lower-dimensional space.
中文：深度神經網絡通常學習到的對數據的壓縮和抽象表示，以較低維度來捕捉數據的內在模式和結構。

Learning Rate (學習率)

English: A hyperparameter that controls how much to adjust the model’s parameters in response to the estimated error each time model weights are updated.
中文：一種超參數，用於控製在每次模型權重更新時，針對估計誤差調整模型參數的幅度。

Loss Function (Cost Function) (損失函數/代價函數)

English: A function that measures the disparity between a model’s predictions and the actual ground truth labels. Training aims to minimize this function.
中文：衡量模型預測值與實際真實標簽之間差異的函數。訓練的目標就是使該函數的值最小化。

---

M

Machine Learning (ML) (機器學習)

English: A subset of AI focusing on algorithms that learn from data to make predictions or decisions without being explicitly programmed.
中文：人工智能的一個分支，側重於通過數據學習來做出預測或決策，而不是基於顯式編程。

Model (模型)

English: A representation of the learned patterns or relationships in data—often implemented as a mathematical function or computational architecture.
中文：對數據中所學到的模式或關係的表示，通常以數學函數或計算架構的形式實現。

Momentum (動量)

English: In gradient-based optimization, a method that helps to accelerate gradient descent by moving more consistently in the direction of minima, reducing oscillations.
中文：在基於梯度的優化中，一種通過在朝最小值方向更連續地移動來加速梯度下降並減少振蕩的方法。

---

N

Natural Language Processing (NLP) (自然語言處理)

English: A field of AI that deals with understanding, interpreting, and generating human language. Applications include text classification, machine translation, and chatbots.
中文：人工智能的一個領域，涉及對人類語言的理解、解釋和生成。應用包括文本分類、機器翻譯和聊天機器人等。

Normalization (歸一化/標準化)

English: Transforming data to have specific statistical properties (like a mean of 0 and a standard deviation of 1) or scaling data to a specific range. This can improve training stability and speed.
中文：將數據轉換為具有特定統計特性的形式（如均值為0、標準差為1）或將數據縮放至特定範圍，可提高訓練的穩定性和速度。

---

O

Overfitting (過擬合)

English: When a model learns the training data too well, including noise or random fluctuations, and performs poorly on new, unseen data.
中文：模型對訓練數據（包括噪聲或隨機波動）學習得過於透徹，導致在新數據上的表現不佳。

One-Hot Encoding (獨熱編碼)

English: A method of representing categorical variables as binary vectors, where exactly one element is “1” and all others are “0.”
中文：將類別型變量表示為二進製向量的方法，其中僅有一個元素為“1”，其他均為“0”。

---

P

Parameter (參數)

English: A variable in a model (like a weight or bias in a neural network) that is learned during training.
中文：模型中的一個變量（如神經網絡中的權重或偏置），在訓練過程中被學習。

Pretraining (預訓練)

English: Initial training on a large, generic dataset (such as billions of words of text), allowing the model to learn general features. This model can then be fine-tuned on a task-specific dataset.
中文：在大型、通用數據集（如海量文本）上進行的初始訓練，使模型學習到通用特征。隨後可以在特定任務的數據集上進行微調。

Precision (精確率)

English: In a classification context, the proportion of positive predictions that are actually correct. Formula:

Precision=True PositivesTrue Positives+False Positivestext{Precision} = frac{text{True Positives}}{text{True Positives} + text{False Positives}}Precision=True Positives+False PositivesTrue Positives?

中文：在分類任務中，模型預測為正類的實例中真正為正類所占的比例。公式：

Precision=真正例真正例+假正例text{Precision} = frac{text{真正例}}{text{真正例} + text{假正例}}Precision=真正例+假正例真正例?

---

Q

Q-Learning (Q學習)

English: A type of Reinforcement Learning algorithm that learns an action-value function $Q$ indicating the expected reward for taking a given action in a particular state.
中文：一種強化學習算法，通過學習一個動作價值函數 $Q$，來指示在特定狀態下執行給定動作的期望回報。

---

R

Recall (Sensitivity) (召回率/敏感度)

English: In a classification context, the proportion of actual positives that are correctly identified. Formula:

Recall=True PositivesTrue Positives+False Negativestext{Recall} = frac{text{True Positives}}{text{True Positives} + text{False Negatives}}Recall=True Positives+False NegativesTrue Positives?

中文：在分類任務中，實際為正類的實例中被正確識別為正類的比例。公式：

Recall=真正例真正例+假負例text{Recall} = frac{text{真正例}}{text{真正例} + text{假負例}}Recall=真正例+假負例真正例?

Regularization (正則化)

English: Techniques to prevent overfitting by penalizing large weights or complex models (e.g., L1 and L2 regularization, dropout in neural networks).
中文：通過懲罰過大的權重或過於複雜的模型來防止過擬合的技術（例如 L1 與 L2 正則化，以及神經網絡中的 dropout）。

Reinforcement Learning (RL) (強化學習)

English: A learning paradigm where an agent learns by interacting with an environment, receiving rewards or penalties for actions, and aims to maximize the cumulative reward.
中文：一種學習範式，智能體通過與環境交互並根據行動獲得獎勵或懲罰來學習，目標是最大化累積獎勵。

---

S

Semi-Supervised Learning (半監督學習)

English: A learning approach that uses a small amount of labeled data and a large amount of unlabeled data to build better models.
中文：一種學習方法，使用少量有標簽數據和大量無標簽數據來構建更好的模型。

Stochastic Gradient Descent (SGD) (隨機梯度下降)

English: A variant of gradient descent where parameters are updated for each training example (or a small batch) rather than the entire dataset, speeding up computations.
中文：梯度下降的一種變體，每個訓練樣本（或小批量）都會更新參數，而不是等待處理整個數據集，從而加快計算速度。

Supervised Learning (監督學習)

English: A machine learning task that uses labeled data to train models. The algorithm learns by comparing its predictions with the known ground truth.
中文：一種使用帶標簽數據來訓練模型的機器學習任務。算法通過將預測結果與已知真實值進行比較來學習。

---

T

Tensor (張量)

English: A multi-dimensional array used in numerical computing frameworks like TensorFlow or PyTorch for representing data.
中文：一種多維數組，用於在諸如 TensorFlow 或 PyTorch 等數值計算框架中表示數據。

Transfer Learning (遷移學習)

English: A technique where a model trained on one task is repurposed on a second, related task, leveraging the learned features.
中文：將已經在某一任務上訓練好的模型應用於另一個相關任務，並利用其已學到的特征的技術。

Transformer (Transformer模型)

English: A deep learning architecture primarily used in NLP (e.g., BERT, GPT). It relies on self-attention mechanisms rather than convolutions or recurrent modules.
中文：一種主要用於自然語言處理（如 BERT、GPT）領域的深度學習架構，依賴於自注意力機製，而不是卷積或循環結構。

Training (訓練)

English: The process of feeding data to a model so it can learn the relationships or patterns within that data.
中文：將數據輸入模型的過程，以便模型學習數據中的關係或模式。

---

U

Underfitting (欠擬合)

English: When a model is too simple or has not learned enough structure from the data, resulting in poor performance even on training data.
中文：模型過於簡單或沒有充分學習數據的結構，即使在訓練數據上也表現不佳。

Unsupervised Learning (無監督學習)

English: A family of algorithms that discover patterns in data without labeled examples, such as clustering or dimensionality reduction.
中文：無需標簽示例而在數據中發現模式的一類算法，如聚類或降維。

---

V

Validation Set (驗證集)

English: A portion of data held out from training to tune model hyperparameters and prevent overfitting.
中文：從訓練中留出的一部分數據，用來調節模型的超參數並防止過擬合。

Variance (方差)

English: How much a model’s predictions for a given point vary across different training sets. High variance may indicate the model is overfitting.
中文：模型對同一個數據點在不同訓練集中預測結果的變化程度。高方差可能表明模型出現過擬合。

Vectorization (向量化)

English: Expressing computations (such as model operations) in terms of vectors and matrices for efficient parallel processing, often used in deep learning frameworks.
中文：以向量和矩陣的形式表達計算（如模型操作）以實現高效的並行處理，這在深度學習框架中常被使用。

---

W

Weight (權重)

English: A parameter in a neural network that is multiplied by an input value. Adjusting weights is central to training neural networks.
中文：神經網絡中的一種參數，用於與輸入值相乘。調整權重是訓練神經網絡的核心。

---

X

XAI (Explainable AI) (可解釋的人工智能)

English: Methods and techniques focused on understanding and interpreting the decisions and inner workings of AI models, ensuring transparency.
中文：專注於理解和解釋人工智能模型決策及其內部機製的方法和技術，以確保透明度。