📚 IB Computer Science: Introduction to Machine Learning Key Points | IB 计算机：机器学习入门 考点精讲
Machine learning is a subfield of artificial intelligence that empowers computers to learn from data and improve their performance on specific tasks without being explicitly programmed. For IB Computer Science students, understanding the core concepts of ML—such as supervised, unsupervised, and reinforcement learning—provides a vital foundation for exploring modern computational problem-solving and data-driven decision making.
机器学习是人工智能的一个分支，它使计算机能够从数据中学习，并在特定任务上提高性能，而无需显式编程。对于IB计算机科学的学生来说，理解监督学习、无监督学习和强化学习等核心概念，为探索现代计算问题解决和基于数据的决策奠定了重要基础。
1. What is Machine Learning? | 什么是机器学习？
Machine learning (ML) refers to algorithms that build a model based on sample data, known as training data, in order to make predictions or decisions without being explicitly programmed to do so. The key idea is that the system identifies patterns, learns from examples, and generalises to new, unseen inputs.
机器学习（ML）指的是根据样本数据（即训练数据）构建模型的算法，目的是无需显式编程即可进行预测或决策。其核心思想是系统识别模式、从示例中学习，并能推广到新的未见输入。
ML is often contrasted with traditional rule-based programming. In traditional programming, a human writes explicit rules (e.g., if temperature > 30 then turn on fan). In ML, the algorithm discovers the rules from the data itself. This makes ML particularly powerful for problems where manual rule-crafting is impractical, such as image recognition or natural language processing.
ML常与传统的基于规则的编程形成对比。在传统编程中，人类编写明确的规则（例如，如果温度 > 30 则打开风扇）。在ML中，算法从数据本身中发现规则。这使得ML对于那些手工制定规则不切实际的问题（如图像识别或自然语言处理）特别强大。
2. Types of Machine Learning | 机器学习的类型
The three main paradigms of machine learning are supervised learning, unsupervised learning, and reinforcement learning. Each addresses a different type of problem and uses a different type of data. A clear understanding of their distinctions is essential for selecting the right approach.
机器学习的三种主要范式是监督学习、无监督学习和强化学习。每种范式解决不同类型的问题并使用不同类型的数据。清楚地理解它们之间的区别对于选择正确的方法至关重要。
In supervised learning, the model is trained on a labelled dataset, meaning each training example is paired with an output label (the target). The goal is to learn a mapping from inputs to outputs so it can predict labels for new data. Unsupervised learning works with unlabelled data and seeks to discover hidden structures or patterns, such as groupings or clusters. Reinforcement learning involves an agent that learns to make decisions by interacting with an environment, receiving rewards or penalties in return.
在监督学习中，模型在带标签的数据集上进行训练，这意味着每个训练示例都配有输出标签（目标）。其目标是学习从输入到输出的映射，以便能够预测新数据的标签。无监督学习处理无标签数据，旨在发现隐藏的结构或模式，例如分组或聚类。强化学习则涉及一个智能体，它通过与环境交互来学习做出决策，并获得奖励或惩罚作为反馈。
3. Supervised Learning in Detail | 监督学习详解
Supervised learning is divided into two main categories: classification and regression. Classification predicts a discrete category or class label, such as spam or not spam. Regression predicts a continuous numerical value, such as the price of a house.
监督学习分为两大类：分类和回归。分类预测一个离散的类别或类标签，如垃圾邮件或非垃圾邮件。回归预测一个连续的数值，如房屋价格。
Common supervised learning algorithms include linear regression for regression tasks, and logistic regression, decision trees, and support vector machines (SVM) for classification tasks. The training process involves minimising a loss function, such as mean squared error (MSE) for regression: MSE = (1/n) Σ(yᵢ − ŷᵢ)², where yᵢ is the actual value and ŷᵢ is the predicted value.
常见的监督学习算法包括用于回归任务的线性回归，以及用于分类任务的逻辑回归、决策树和支持向量机（SVM）。训练过程涉及最小化损失函数，例如回归中的均方误差（MSE）：MSE = (1/n) Σ(yᵢ − ŷᵢ)²，其中yᵢ是实际值，ŷᵢ是预测值。
Underfitting and overfitting are important concepts in supervised learning. A model that is too simple may underfit and fail to capture the underlying trend. A model that is too complex may overfit and essentially memorise the training data, including its noise, leading to poor generalisation to new data.
欠拟合和过拟合是监督学习中的重要概念。过于简单的模型可能会欠拟合，无法捕捉潜在趋势。过于复杂的模型可能会过拟合，实际上记住了训练数据，包括其中的噪声，导致对新数据的泛化能力差。
4. Unsupervised Learning in Detail | 无监督学习详解
Unsupervised learning aims to find patterns or structures in unlabelled data. The two most common tasks are clustering and dimensionality reduction. Clustering groups similar data points together; for example, customer segmentation in marketing. Dimensionality reduction reduces the number of features while preserving essential information, aiding visualisation and speeding up computation.
无监督学习旨在发现无标签数据中的模式或结构。最常见的两种任务是聚类和降维。聚类将相似的数据点归为一组；例如，市场营销中的客户细分。降维在保留基本信息的同时减少特征数量，有助于可视化并加速计算。
The K-means algorithm is a classic clustering method. It partitions data into K clusters by minimising the within-cluster sum of squares. The algorithm initialises K centroids, assigns each data point to the nearest centroid, recomputes centroids as the mean of the assigned points, and repeats until convergence.
K-means 算法是一种经典的聚类方法。它通过最小化簇内平方和将数据划分为K个簇。该算法初始化K个质心，将每个数据点分配给最近的质心，重新计算质心作为所分配点的均值，并重复直到收敛。
Principal Component Analysis (PCA) is a widely used dimensionality reduction technique. It transforms the data into a new coordinate system where the greatest variance lies on the first principal component, the second greatest on the second, and so on. This allows us to keep only the top components while discarding less informative dimensions.
主成分分析（PCA）是一种广泛使用的降维技术。它将数据变换到一个新的坐标系中，使得数据方差最大的方向作为第一主成分，第二大的作为第二主成分，以此类推。这样我们可以只保留前几个主成分，而丢弃信息量较少的维度。
5. Reinforcement Learning Overview | 强化学习概述
Reinforcement learning (RL) is inspired by behavioural psychology: an agent learns by interacting with an environment, performing actions, and observing the results. The agent receives a reward signal that indicates the immediate goodness of an action. The goal is to learn a policy that maximises cumulative reward over time.
强化学习（RL）受行为心理学启发：智能体通过与环境交互、执行动作并观察结果来学习。智能体接收一个奖励信号，表明该动作的即时好坏程度。目标是学习一个策略，使得随时间累积的奖励最大化。
The core elements of RL are the state, action, reward, and policy. The agent’s policy maps states to actions. The Q-learning algorithm, a model-free method, estimates the value of taking a given action in a given state, known as the Q-value. The update rule is: Q(s,a) ← Q(s,a) + α [r + γ maxₐ’ Q(s’,a’) − Q(s,a)], where α is the learning rate, γ is the discount factor, r is the reward, s’ is the next state.
RL的核心要素是状态、动作、奖励和策略。智能体的策略将状态映射到动作。Q-learning 算法是一种无模型方法，它估计在给定状态下采取特定动作的价值，即Q值。更新规则为：Q(s,a) ← Q(s,a) + α [r + γ maxₐ’ Q(s’,a’) − Q(s,a)]，其中α是学习率，γ是折扣因子，r是奖励，s’是下一个状态。
RL has been successfully applied in game playing (e.g., AlphaGo), robotics, and autonomous driving. In IB Computer Science, it illustrates how agents can autonomously learn complex behaviours through trial and error.
强化学习已成功应用于游戏（如AlphaGo）、机器人和自动驾驶。在IB计算机科学中，它展示了智能体如何通过试错自主地学习复杂行为。
6. Data Preprocessing | 数据预处理
Real-world data is rarely clean and ready for analysis. Data preprocessing transforms raw data into a format suitable for machine learning. Key steps include handling missing values, normalisation, and encoding categorical variables.
现实世界的数据很少是干净且可直接用于分析的。数据预处理将原始数据转换成适合机器学习的格式。关键步骤包括处理缺失值、归一化和分类变量编码。
Missing values can be dealt with by removing the instances, or more commonly by imputation—replacing them with the mean, median, or a predicted value. Normalisation scales features to a common range, typically [0,1] or standardisation to zero mean and unit variance, which helps algorithms that rely on distance calculations (e.g., K-means, SVM). Categorical features such as colour (‘red’,’blue’) are often one-hot encoded into binary vectors.
缺失值可以通过删除该实例来处理，或者更常用的是通过插补——用均值、中位数或预测值替换它们。归一化将特征缩放到一个共同的范围，通常是[0,1]，或标准化为零均值和单位方差，这有助于依赖距离计算的算法（如K-means、SVM）。分类特征如颜色（“红”、“蓝”）通常被独热编码为二进制向量。
Feature engineering, the process of creating new meaningful features from existing ones, can significantly boost model performance. For example, from a date column, extracting day of the week or month as a new feature can reveal cyclic patterns.
特征工程是从现有特征创建有意义的新的特征的过程，可以显著提高模型性能。例如，从日期列中提取星期几或月份作为新特征，可以揭示周期性模式。
7. Overfitting and Underfitting | 过拟合与欠拟合
Overfitting occurs when a model learns the training data too well, capturing noise and random fluctuations instead of the underlying pattern. It performs excellently on training data but poorly on unseen test data. Underfitting happens when the model is too simplistic to capture the complexity of the data, resulting in poor performance on both training and test sets.
过拟合发生在模型对训练数据学习得过好，捕捉到了噪声和随机波动而非潜在模式。它在训练数据上表现优异，但在未见过的测试数据上表现不佳。欠拟合则发生在模型过于简单，无法捕捉数据的复杂性，导致在训练集和测试集上都表现不佳。
To combat overfitting, several regularisation techniques are used. Common methods include L1 and L2 regularisation, which add a penalty term to the loss function based on the magnitude of model coefficients. Cross-validation, especially k-fold cross-validation, provides a robust estimate of model performance on unseen data and helps in tuning hyperparameters. Early stopping halts training when performance on a validation set stops improving.
为防止过拟合，可使用多种正则化技术。常见的方法包括L1和L2正则化，它们根据模型系数的大小给损失函数添加惩罚项。交叉验证，尤其是k折交叉验证，能对模型在未见数据上的性能提供稳健估计，并有助于超参数调优。早停法在验证集性能不再提升时停止训练。
Model complexity and the amount of data play crucial roles. A more complex model with many parameters requires more data to avoid overfitting. Gathering more training data, data augmentation, and pruning decision trees are practical strategies to achieve better generalisation.
模型复杂度和数据量起着关键作用。参数众多的复杂模型需要更多数据来避免过拟合。收集更多训练数据、数据增强和剪枝决策树是实现更佳泛化能力的实用策略。
8. Evaluation Metrics | 评估指标
Choosing the right evaluation metric is vital to correctly assess a machine learning model. For classification, accuracy is the proportion of correct predictions among total predictions, but it can be misleading with imbalanced datasets. Precision, recall, and F1-score provide a more nuanced view.
选择合适的评估指标对于正确评估机器学习模型至关重要。对于分类任务，准确率是正确预测占总预测的比例，但在不平衡数据集中可能具有误导性。精确率、召回率和F1分数提供了更细致的视角。
Precision measures the fraction of true positives among all predicted positives: Precision = TP / (TP + FP). Recall (sensitivity) measures the fraction of true positives among all actual positives: Recall = TP / (TP + FN). The F1-score is the harmonic mean of precision and recall: F1 = 2 × (Precision × Recall) / (Precision + Recall). These metrics are derived from the confusion matrix, which tabulates TP, TN, FP, FN.
精确率衡量所有预测为正的样本中真正为正的比例：精确率 = TP / (TP + FP)。召回率（灵敏度）衡量所有实际为正的样本中被正确预测为正的比例：召回率 = TP / (TP + FN)。F1分数是精确率和召回率的调和平均数：F1 = 2 × (精确率 × 召回率) / (精确率 + 召回率)。这些指标源于混淆矩阵，该矩阵列出了TP、TN、FP、FN。
For regression, common metrics are Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). R-squared (R²) indicates the proportion of variance in the dependent variable that is predictable from the independent variables; a score close to 1 indicates a good fit.
对于回归任务，常见的指标是平均绝对误差（MAE）、均方误差（MSE）和均方根误差（RMSE）。R平方（R²）表示因变量的方差中可由自变量解释的比例；接近1的分数表明拟合良好。
9. Introduction to Neural Networks | 神经网络入门
Artificial neural networks (ANNs) are computing systems inspired by biological neural networks. They consist of interconnected nodes (neurons) organised in layers: an input layer, one or more hidden layers, and an output layer. Each connection has a weight that is adjusted during learning.
人工神经网络（ANN）是受生物神经网络启发的计算系统。它们由相互连接的节点（神经元）组成，这些节点分层组织：输入层、一个或多个隐藏层以及输出层。每个连接都有一个权重，在学习过程中进行调整。
A neuron computes a weighted sum of its inputs, adds a bias, and then applies an activation function to produce the output. Common activation functions include sigmoid: σ(x) = 1/(1 + e⁻ˣ), which squashes values between 0 and 1; and ReLU (Rectified Linear Unit): f(x) = max(0, x), which introduces non-linearity while being computationally efficient.
一个神经元计算其输入的加权和，加上偏置，然后应用激活函数产生输出。常见的激活函数包括 sigmoid：σ(x) = 1/(1 + e⁻ˣ)，它将值压缩在0到1之间；以及ReLU（线性整流单元）：f(x) = max(0, x)，它在计算高效的同时引入非线性。
Training a neural network typically involves backpropagation and gradient descent. The algorithm calculates the gradient of the loss function with respect to each weight by the chain rule, propagating errors backward from the output layer. Weights are then updated to minimise the loss. Deep learning refers to neural networks with many hidden layers, enabling the learning of hierarchical feature representations.
训练神经网络通常涉及反向传播和梯度下降。该算法利用链式法则计算损失函数相对于每个权重的梯度，将误差从输出层向后传播。然后更新权重以最小化损失。深度学习指具有多个隐藏层的神经网络，能够学习层次化的特征表示。
10. Ethical Considerations and Societal Impact | 伦理考量与社会影响
As machine learning systems become increasingly integrated into daily life, it is crucial for IB students to consider the ethical implications. Bias in training data can lead to discriminatory outcomes, for example in hiring algorithms or facial recognition systems that perform poorly on certain demographic groups.
随着机器学习系统日益融入日常生活，IB学生必须考虑其伦理影响。训练数据中的偏见可能导致歧视性结果，例如在招聘算法或面部识别系统中，对某些人群的表现不佳。
Transparency and explainability are major concerns. Many ML models, especially deep neural networks, act as “black boxes”, making it difficult to understand how a decision was reached. This lack of interpretability can conflict with the need for accountability in sensitive areas like healthcare and criminal justice.
透明度和可解释性是主要关切点。许多ML模型，尤其是深度神经网络，充当“黑箱”，使得理解决策是如何做出的变得困难。这种可解释性的缺乏可能与医疗保健和刑事司法等敏感领域问责的需要相冲突。
Privacy is another significant issue. Machine learning often relies on vast amounts of personal data. Data anonymisation and consent mechanisms are essential, but re-identification risks persist. Moreover, automation driven by ML could displace jobs, raising socioeconomic challenges. Responsible development of ML includes fairness, accountability, and transparency (FAT) principles.
隐私是另一个重大问题。机器学习常常依赖大量个人数据。数据匿名化和同意机制至关重要，但重新识别的风险仍然存在。此外，由ML驱动的自动化可能取代工作岗位，引发社会经济挑战。负责任的ML发展包括公平、问责和透明（FAT）原则。
Published by TutorHao | Computer Science Revision Series | aleveler.com
更多咨询请联系16621398022（同微信）
屏轩国际教育cambridge primary/secondary checkpoint, cat4, ukiset,ukcat,igcse,alevel,PAT,STEP,MAT, ibdp,ap,ssat,sat,sat2课程辅导，国外大学本科硕士研究生博士课程论文辅导