Introduction to Machine Learning | 机器学习入门

📚 Introduction to Machine Learning | 机器学习入门

Machine learning is a branch of artificial intelligence that enables computers to learn from data without being explicitly programmed for every scenario. It is transforming how we interact with technology, from personalised recommendations on streaming platforms to voice assistants that understand natural language. For IGCSE CCEA Computer Science, you need to grasp the core concepts, types of learning, key terminology, and the ethical implications of this rapidly growing field.

机器学习是人工智能的一个分支，它让计算机能够从数据中学习，而不必为每一种情况都进行显式编程。它正在改变我们与技术互动的方式，从流媒体平台上的个性化推荐到理解自然语言的语音助手。对于 IGCSE CCEA 计算机科学来说，你需要掌握核心概念、学习类型、关键术语以及这一快速发展领域的伦理影响。

1. What is Machine Learning? | 什么是机器学习？

Machine learning allows systems to improve their performance on a task over time by recognising patterns in data. Instead of following fixed rules written by a programmer, the system builds a mathematical model based on examples. This model can then make predictions or decisions when presented with new data.

机器学习通过识别数据中的模式，让系统随着时间的推移提升在某一任务上的表现。系统并不遵循程序员编写的固定规则，而是基于示例构建一个数学模型。然后，当面对新数据时，该模型可以进行预测或做出决策。

A classic example is an email spam filter. Rather than manually creating a set of rules for every spam keyword, the filter is trained on thousands of emails labelled as ‘spam’ or ‘not spam’. The algorithm learns the characteristics of spam messages and applies this knowledge to classify future emails.

一个经典例子是电子邮件垃圾邮件过滤器。不是手动为每个垃圾邮件关键词编写一组规则，而是用数千封被标记为“垃圾邮件”或“非垃圾邮件”的邮件来训练过滤器。算法会学习垃圾邮件的特征，并利用这些知识对未来的邮件进行分类。

2. Machine Learning vs Traditional Programming | 机器学习与传统编程的区别

In traditional programming, a developer writes explicit instructions: the program receives input, processes it according to predefined rules, and produces output. The logic is fixed and does not change unless the code is modified. With machine learning, the logic is not written directly; instead, the system learns the mapping between inputs and outputs from a large set of examples.

在传统编程中，开发人员编写显式指令：程序接收输入，根据预定义的规则进行处理，然后产生输出。逻辑是固定的，除非修改代码，否则不会改变。而在机器学习里，逻辑不是直接编写的；相反，系统通过大量示例来学习输入与输出之间的映射关系。

Consider a program that identifies cats in photos. With traditional programming, you would need to specify rules about shapes, edges and colours — an almost impossible task given the variety of cat appearances. A machine learning approach feeds the program many labelled photos of cats and non-cats, and the algorithm works out the distinguishing features by itself.

设想一个识别照片中猫的程序。使用传统编程，你需要指定关于形状、边缘和颜色的规则 —— 鉴于猫外观的多样性，这几乎是不可能完成的任务。机器学习的方法则是向程序提供许多已标记的猫和非猫的照片，算法会自行找出区分特征。

Aspect	Traditional Programming	Machine Learning
Logic source	Programmer-defined rules	Learned from data patterns
Adaptability	Static; needs code changes	Can improve with more data
Best for	Well-defined problems	Problems with complex patterns

中文对照：

方面	传统编程	机器学习
逻辑来源	程序员定义的规则	从数据模式中学习
适应性	静态；需修改代码	可随更多数据改进
适用场景	定义明确的问题	具有复杂模式的问题

3. Types of Machine Learning | 机器学习的类型

Machine learning is typically divided into three main categories based on how the system learns from data: supervised learning, unsupervised learning and reinforcement learning. Each type uses a different approach to training and is suitable for different kinds of problems.

机器学习通常根据系统从数据中学习的方式分为三大类：监督学习、无监督学习和强化学习。每种类型采用不同的训练方法，并适用于不同类型的问题。

Supervised learning uses labelled datasets, where each training example has an input and a known correct output.
监督学习 使用带标签的数据集，每个训练示例都有一个输入和一个已知的正确输出。
Unsupervised learning works with unlabelled data, looking for hidden structures or groupings.
无监督学习 处理无标签数据，寻找隐藏的结构或分组。
Reinforcement learning involves an agent that learns to make decisions by receiving rewards or penalties from its environment.
强化学习 涉及一个智能体，它通过从环境中获取奖励或惩罚来学习如何做出决策。

4. Supervised Learning in Detail | 监督学习详解

In supervised learning, the algorithm is given a dataset where the inputs (features) are paired with the correct outputs (labels). The goal is to learn a function that maps inputs to outputs, so the model can predict labels for new, unseen data. This is similar to a student learning with an answer key.

在监督学习中，算法接收的数据集约定了输入（特征）与正确输出（标签）的配对。其目标是学习一个将输入映射到输出的函数，从而使模型能够预测新的、未见过的数据的标签。这就像学生拿着答案来学习。

Two common tasks are classification and regression. Classification predicts a discrete category, such as whether an email is spam or not. Regression predicts a continuous numeric value, such as the price of a house based on its features.

两种常见的任务是分类和回归。分类预测离散的类别，例如一封邮件是否是垃圾邮件。回归预测连续的数值，例如根据房屋的特征预测其价格。

5. Unsupervised Learning: Finding Hidden Patterns | 无监督学习：寻找隐藏模式

Unsupervised learning algorithms explore data without any predefined labels. The system tries to identify patterns, similarities or anomalies on its own. This is particularly useful when you have large amounts of data but no clear idea of what groupings might exist.

无监督学习算法在没有任何预定义标签的情况下探索数据。系统自行尝试识别模式、相似性或异常。当你拥有大量数据但不清楚可能存在哪些分组时，这种方法尤其有用。

A typical technique is clustering, where the algorithm groups data points into clusters based on similarity. For example, a streaming service might use clustering to group listeners with similar music tastes without knowing anything about them in advance. Another technique is dimensionality reduction, which simplifies data while retaining its essential structure.

一种典型的技术是聚类，算法根据相似性将数据点划分为不同的簇。例如，流媒体服务可能使用聚类将具有相似音乐品味的听众分组，而无需提前了解他们的任何信息。另一种技术是降维，它在保持数据基本结构的同时简化数据。

6. Reinforcement Learning: Learning from Rewards | 强化学习：从奖励中学习

Reinforcement learning differs from the first two types because the learner, called an agent, interacts with an environment and learns to maximise cumulative reward. The agent takes actions, observes the outcome, and receives feedback in the form of a positive or negative reward. Over many trials, it discovers which actions lead to the best long-term results.

强化学习与前两种类型不同，因为学习者（称为智能体）与环境交互，学习最大化累积奖励。智能体采取行动，观察结果，并以正向或负向奖励的形式获得反馈。经过多次尝试，它会发现哪些行动能带来最好的长期结果。

This approach is used in game-playing AI (like AlphaGo), robotics, and self-driving car navigation. The learning process often involves a trade-off between exploration (trying new moves) and exploitation (using moves known to work well).

这种方法被用于游戏AI（如AlphaGo）、机器人技术和自动驾驶汽车导航。学习过程通常涉及探索（尝试新动作）和利用（使用已知的良好动作）之间的权衡。

7. The Role of Data in Machine Learning | 数据在机器学习中的作用

Data is the foundation of any machine learning project. The quality and quantity of data directly affect how well a model learns. A dataset is typically split into training data, used to build the model, and testing data, used to evaluate its performance on unseen examples.

数据是任何机器学习项目的基础。数据的质量和数量直接影响模型的学习效果。数据集通常分为训练数据（用于构建模型）和测试数据（用于评估其在未见过的示例上的表现）。

If the data is biased or incomplete, the model can inherit those flaws. For instance, a facial recognition system trained mostly on one skin tone may perform poorly on others. Therefore, collecting a representative dataset is crucial.

如果数据存在偏见或不完整，模型就可能继承这些缺陷。例如，主要针对一种肤色训练的面部识别系统可能对其他肤色效果不佳。因此，收集具有代表性的数据集至关重要。

8. Training, Testing and Validation | 训练、测试与验证

To avoid overfitting — where a model memorises the training data instead of generalising — practitioners split the available data. A common split is 70% for training, 15% for validation and 15% for testing. The validation set is used during development to tune hyperparameters, while the test set is only used at the end to measure final performance.

为了避免过拟合（即模型只是记住了训练数据而未能泛化），从业者会将可用数据进行划分。常见的划分比例是70%用于训练，15%用于验证，15%用于测试。验证集在开发过程中用于调整超参数，而测试集仅在最后用来衡量最终性能。

Cross-validation is another technique where the data is divided into multiple folds, and the model is trained and evaluated several times, each time using a different fold as the test set. This gives a more reliable estimate of how well the model will perform on new data.

交叉验证是另一种技术，将数据分为多个折数，模型多次进行训练和评估，每次使用不同的折数作为测试集。这能对模型在新数据上的表现给出更可靠的估计。

9. Key Terminology: Features, Labels and Models | 关键术语：特征、标签和模型

Understanding the language of machine learning is essential. A feature is an individual measurable property of the data, such as the height, weight or age of a person. Labels are the answers we want to predict in supervised learning, for example ‘cat’ or ‘dog’. The model is the mathematical representation that maps features to predictions.

理解机器学习的语言至关重要。特征是数据的单个可测量属性，例如一个人的身高、体重或年龄。标签是在监督学习中我们希望预测的答案，比如“猫”或“狗”。模型是将特征映射到预测的数学表示。

For a house price predictor, features might include number of bedrooms, square footage and location; the label is the sale price. The model learns the relationship between these features and the price from historical data.

对于房价预测器，特征可能包括卧室数量、面积和地段；标签是销售价格。模型从历史数据中学习这些特征与价格之间的关系。

10. Evaluating Model Performance | 评估模型性能

Once a model is trained, we need to measure how good it is. For classification tasks, common metrics include accuracy (percentage of correct predictions), precision (how many of the positive predictions were truly positive), recall (how many actual positives were correctly identified) and the F1 score (harmonised mean of precision and recall). For regression tasks, mean absolute error or root mean squared error are often used.

模型训练完成后，我们需要衡量它的表现如何。对于分类任务，常见的指标包括准确率（正确预测的百分比）、精确率（预测为正的样本中真正为正的比例）、召回率（实际为正的样本中被正确识别出来的比例）以及F1分数（精确率和召回率的调和平均数）。对于回归任务，常使用平均绝对误差或均方根误差。

A confusion matrix is a useful tool for visualising the performance of a classification model. It shows true positives, false positives, true negatives and false negatives, allowing you to see exactly where the model is making mistakes.

混淆矩阵是可视化分类模型性能的有用工具。它展示了真正类、假正类、真负类和假负类，让你能够准确看到模型在哪里犯了错误。

11. Ethical Considerations in Machine Learning | 机器学习中的伦理性考量

As machine learning becomes embedded in everyday life, ethical issues are increasingly important. Algorithmic bias can lead to unfair treatment in areas like hiring, lending and law enforcement if the training data reflects historical inequalities. Transparency is another concern: many complex models act as ‘black boxes’, making it hard to explain why a particular decision was made.

随着机器学习融入日常生活，伦理问题变得越来越重要。算法偏见可能导致招聘、贷款和执法等领域的不公平对待，如果训练数据反映了历史上的不平等。透明度是另一个担忧：许多复杂的模型如同“黑箱”，很难解释为何做出特定决策。

Privacy is also at stake, as machine learning often requires large amounts of personal data. Students should understand the importance of informed consent, data anonymisation, and regulations like the GDPR. Responsible development involves checking for bias, ensuring fairness and maintaining accountability.

隐私也面临风险，因为机器学习通常需要大量的个人数据。学生应理解知情同意、数据匿名化以及GDPR等法规的重要性。负责任的开发包括检查偏见、确保公平和维持问责制。

12. Real-World Applications | 实际应用

Machine learning is not just a theoretical topic; it powers many modern technologies. In healthcare, it helps diagnose diseases from medical images. In finance, it detects fraudulent transactions. In transportation, it enables autonomous vehicles to perceive their surroundings. Other examples include voice assistants, predictive text, and recommendation engines used by social media and e-commerce platforms.

机器学习不只是一个理论课题，它驱动着众多现代技术。在医疗领域，它帮助从医学图像中诊断疾病。在金融领域，它检测欺诈交易。在交通领域，它使自动驾驶汽车能够感知周围环境。其他例子包括语音助手、预测文本以及社交媒体和电商平台使用的推荐引擎。

Understanding these applications helps you appreciate why machine learning is a core topic in CCEA IGCSE Computer Science. You may be asked to discuss examples, identify the type of learning used, or evaluate the benefits and risks of deploying these systems in society.

理解这些应用有助于你认识到为什么机器学习是CCEA IGCSE计算机科学的核心主题。你可能会在考试中被要求讨论实例、辨别所用的学习类型，或评估在社会中部署这些系统的益处与风险。

Published by TutorHao | Computer Science Revision Series | aleveler.com

更多咨询请联系16621398022（同微信）

Introduction to Machine Learning | 机器学习入门