01 - Berkenalan dengan Machine Learning

Machine learning is best understood not as magic but as a math-based branch of artificial intelligence in which computers learn from data to recognize patterns, and it is already embedded in everyday life through tools like spam filters, facial recognition, voice assistants, traffic cameras, and self-driving cars. The core workflow involves collecting training data, feeding it into an algorithm to produce a model, and then testing that model on separate data to ensure it performs accurately—for example, a decision tree trained to classify emails as spam or non-spam, or a neural network trained to recognize handwritten digits. Within AI, "learning" is one of four fundamental techniques and is broken down into supervised, unsupervised, and reinforcement learning, with supervised learning further divided into classification—predicting categories such as spam vs. non-spam, ideal vs. not-ideal posture, or COVID-positive vs. COVID-negative from breath—and regression, which predicts continuous numeric values like temperature, gold prices, or age. Ultimately, the central message is that data is the crucial ingredient: the quality of what a machine learns depends entirely on the examples it is given, and after learning and testing, the model becomes a smart, useful tool for making everyday decisions.

• ▶ 0:21 Machine learning is best understood through everyday applications we already use, such as spam filtering, facial recognition, and voice assistants.
• ▶ 2:45 Machine learning powers real-world systems like traffic cameras that detect violations and read license plates, as well as self-driving cars that interpret their surroundings.
• ▶ 3:44 All these examples show that machine learning is already embedded in daily life, making routine activities easier and less noticeable.
• ▶ 4:23 Machine learning is a field within artificial intelligence (AI) that lets computers learn from data to gain patterns and knowledge — it is a math-based technique, not magic or sorcery.
• ▶ 7:40 Data is a crucial ingredient: you need data to teach the computer, and once it learns from that data, it becomes smart and useful.
• ▶ 8:02 Within AI, there are four basic techniques — searching, reasoning, planning, and learning — and the "learning" technique is what is now called machine learning, covering supervised, unsupervised, and reinforcement learning.
• ▶ 10:55 The machine learning workflow starts with collecting training data, which the computer studies using a specific algorithm to produce a machine learning model (the "knowledge" gained from the data).
• ▶ 12:33 The model is then evaluated using separate testing data—analogous to actual exam questions versus try-out questions—to check if the model is smart enough.
• ▶ 13:01 The model's output on the testing data is compared against the answer key; if it doesn't match, the learning process must be repeated or extended until results align.
• ▶ 14:24 The first step in the workflow is collecting training data—example spam and legitimate emails—which is then combined into one dataset.
• ▶ 15:04 The training data is fed into a machine learning algorithm (here, a decision tree), which learns patterns from the data by repeatedly studying it.
• ▶ 15:52 After learning, the algorithm produces a final decision tree model, ready to be used for classifying emails as spam or not spam.
• ▶ 16:19 The type of model produced depends on the chosen machine learning algorithm, so different algorithms yield different models.
• ▶ 16:38 Before trusting a model, it must be tested for accuracy using a separate testing dataset that is different from the training data.
• ▶ 16:49 The model is tested on unseen examples to check whether it correctly produces the expected output from the new test data.
• ▶ 17:10 A known spam email is used as test data and fed into the trained Decision Tree model.
• ▶ 17:25 The model correctly outputs a prediction that the email is spam, matching the expected result.
• ▶ 17:44 The model is considered "smart" because its prediction matches the actual test label, successfully classifying spam vs. non-spam.

• ▶ 18:01 The example goal is to recognize handwritten digits 0–9, using as many collected image examples as possible to improve model performance.

• ▶ 18:38 An artificial neural network is trained on the handwritten digit data, learning to recognize patterns like shapes for "1" and round forms for "8".

• ▶ 19:14 Training produces an artificial neural network model (unlike the decision tree from the spam example), which is then tested with new data and retrained if predictions are incorrect.

• ▶ 20:06 Machine learning is fundamentally about the data studied by the algorithm; the data is the central input for learning.
• ▶ 21:23 There are three learning paradigms—supervised, unsupervised, and reinforcement learning—but the section focuses only on supervised and unsupervised.
• ▶ 22:35 Unsupervised learning is generally more challenging than supervised learning because supervised learning has a teacher to correct errors, while unsupervised learning has no external guidance or correction.
• ▶ 23:03 Supervised learning types will be discussed one by one in upcoming videos.
• ▶ 23:06 Under supervised learning, there are two key variants: classification and regression.
• ▶ 23:15 The next videos will explain what classification is in more detail, as part of supervised learning.
• ▶ 23:25 Classification is a core supervised learning concept where a computer separates one type of object or instance from another by distinguishing categories/classes.
• ▶ 23:37 Real-world classification examples include spam detection in emails and the GeNose C19 device, which uses breath to distinguish COVID-positive from COVID-negative individuals without invasive nasal swabs.
• ▶ 24:10 Classification also powers face recognition and smartphone Face Unlock, as well as speech and speaker recognition, differentiating between faces, spoken words, and voices.
• ▶ 25:02 Regression is introduced as a supervised learning technique that outputs a numeric value, unlike classification which predicts a class or category.
• ▶ 25:30 Regression is used for prediction/forecasting, with the output always in the form of a number; examples include age-estimation filters, production forecasting, gold price prediction, and weather temperature forecasting.
• ▶ 27:20 The key distinction is that classification assigns labels/groups, while regression predicts continuous numeric quantities.
• ▶ 27:38 Classification and regression examples are introduced to illustrate supervised learning concepts.
• ▶ 27:49 Core principle: a computer learns from data, specifically example data provided to the system.
• ▶ 28:05 Body posture classification example: system classifies posture as ideal or not ideal, using labeled/color-coded data.
• ▶ 28:14 The speaker introduces a concrete example: labeling data for body posture classification, aiming to determine if a person's posture is "ideal" or "not ideal" based on weight and height.
• ▶ 28:17 Data is collected from people in Sewon, recording each person's weight and height as raw measurements.
• ▶ 28:39 Experts (e.g., bodybuilders or fitness enthusiasts) review the collected data and label each individual's posture as "ideal" or "not ideal," producing labeled data for supervised learning.
• ▶ 29:20 The dataset includes a label column representing the expected outcome, which is essential for supervised learning.
• ▶ 29:24 Data is labeled by domain experts, making it reliable training data for the machine learning algorithm.
• ▶ 29:34 The labeled data is fed into the algorithm so it learns and remembers patterns, such as weight/height combinations mapping to "ideal" or "not ideal" labels.
• ▶ 30:04 The system produces a decision tree model from the training data, not a manually designed tree.
• ▶ 30:15 The learned rules classify body type as ideal or not ideal by checking weight first (e.g., weight < 50 leads to a height check for ideal vs. not ideal).
• ▶ 30:44 The resulting branching structure is a simple, interpretable model learned from labeled examples.
• ▶ 31:01 The section revisits the earlier learning-from-data example to show how a classification result is produced.
• ▶ 31:07 The example task is classifying body posture as “ideal” or “not ideal” from physical measurements.
• ▶ 31:16 Using a decision tree produces one form of result, but using a different algorithm changes the result's shape and representation.
• ▶ 31:28 Regression differs from classification: classification uses category labels (e.g., "ideal" vs. "not ideal"), while regression uses numeric/continuous labels (e.g., price).
• ▶ 31:56 Regression is illustrated with house price prediction: input features like land area, building size, and bathrooms are used to train a model on real estate data, producing patterns such as "if land area is this, price is this."
• ▶ 33:53 The trained regression model becomes a mathematical formula (e.g., price = 6.19 × land area + 2.79 × building area + 36.09 × bedrooms), created by an algorithm called linear regression.
• ▶ 34:09 Review of machine learning: training data is processed by a learning algorithm to create a model, which is then tested against testing data to check accuracy.

• ▶ 0:00 Introduction and Applications (0:00 - 4:09) - - Greets viewers, opens the lecture, and shows everyday ML examples like spam filtering.
• ▶ 4:09 Machine Learning Foundations (4:09 - 10:55) - - Defines machine learning as a math-based technique for learning from data, uses the toddler analogy, and places ML within artificial intelligence.
• ▶ 10:55 Machine Learning Workflow and Examples (10:55 - 20:08) - - Introduces the ML workflow and demonstrates it with spam classification and handwritten digit recognition.
• ▶ 20:08 Data and Learning Paradigms (20:08 - 23:06) - - Emphasizes data as the main object and contrasts supervised learning with unsupervised learning.
• ▶ 23:06 Supervised Learning: Classification and Regression (23:06 - 31:30) - - Explains classification vs regression, labeled data, training patterns, and decision tree models.
• ▶ 31:30 Regression Recap, Review, and Closing (31:30 - 35:09) - - Covers house price prediction, reviews how ML learns from patterns, and closes with a summary.