One example at a time一次看一個例子
The computer checks two fruits. For each one, read what it found and see whether the wall moved.
電腦檢查兩顆水果。每一顆都讀一讀它發現了什麼，看看隔板有沒有動。
How Computers Work · Unit 4 · Lesson 15 · 電腦概論 單元四 第十五課
How Does a Computer Learn?電腦怎麼「學習」？
Nobody writes the rule this time. On a table in 3D lie apples and oranges, and a wall that starts in the wrong place. The computer looks at one example after another, and moves the wall a little whenever it is wrong. Watch it find the border by itself. Then give it too few examples, give it wrong answers, and be the teacher yourself.
這一次沒有人寫規則。3D 的桌面上擺著蘋果和橘子，還有一片一開始放錯位置的隔板。電腦一個例子接一個例子地看，每分錯一次，就把隔板挪一點。看它自己把界線找出來。接著只給它很少的例子、給它錯的答案，再換你當老師。
Big idea · 一句話看懂
In machine learning, nobody writes the rule for the computer. Instead it is given many examples with the right answers, and it adjusts itself a little each time it gets one wrong, until it gets most of them right. What it ends up with can only be as good as the examples it was given.
機器學習的時候，沒有人替電腦把規則寫好。電腦拿到的是很多附有正確答案的例子；每分錯一個，它就把自己調整一點，直到大部分都分對。它最後學到的東西，最多只能和它拿到的例子一樣好。
3D Model · 3D 模型
A Wall That Moves Itself 一片會自己挪動的隔板
Each ball on the table is one fruit: red for an apple, orange for an orange. A fruit sits farther right if it is heavier and farther back if it is redder. The upright wall is the computer’s rule: everything on one side it calls an apple, everything on the other an orange. Press “One example” to watch the computer check a single fruit, or “Keep learning” to let it run. The small fruits with white rings are new ones it never studies; they show how well it does on fruit it has not seen.
桌上每一顆球是一顆水果：紅的是蘋果，橘的是橘子。越重的水果越靠右，越紅的越靠後。那片立著的隔板就是電腦的規則：它把一邊的全部叫做蘋果，另一邊的全部叫做橘子。按「看一個例子」，看電腦檢查一顆水果；或按「一直學」讓它跑下去。那些小一號、底下有白圈的是新水果，電腦從來不拿它們來學；它們用來看電腦面對沒看過的水果時表現如何。
Drag to turn · 拖曳旋轉
This 3D model needs WebGL, which this browser does not support. The buttons and the counters below still work, and so do the reading and the games.
這個瀏覽器不支援 WebGL，無法顯示 3D 模型；下面的按鈕和計數照樣能用，課文和小遊戲也都能用。
- Red ball: a fruit labeled “apple”
紅球：標著「蘋果」的水果 - Orange ball: a fruit labeled “orange”
橘球：標著「橘子」的水果 - The wall: the computer’s rule at this moment
隔板：電腦此刻的規則 - Small ball with a white ring: a new fruit, used only for checking. The ring turns red when the computer gets it wrong
底下有白圈的小球：新水果，只用來檢查。電腦分錯時白圈會變紅 - Dark ring: a fruit that was given the wrong label
深色圈：答案被標錯的水果
A teaching model. The fruits and their numbers are made up for this lesson; nobody measured them. The method is one of the earliest and simplest, called a perceptron: one straight wall and three numbers to adjust. Real systems that recognize faces or speech are neural networks, which have a great many adjustable numbers in place of our three, and learn from far more examples. The word “learn” needs care too. The computer is not understanding fruit. It is following a rule for adjusting numbers, exactly as written.
這是教學用的示意模型。這些水果和它們的數字是為這一課編的，沒有人真的去量。這裡的方法是最早、最簡單的一種，叫感知器（perceptron）：一片直的隔板，要調的只有三個數。能認出人臉或語音的真實系統是類神經網路，要調的數多得多，學的例子也多得多。「學習」這兩個字也要小心。電腦並沒有在了解水果；它是在一絲不差地照一條「怎麼調整數字」的規則做。
Model and code are drawn and programmed by My Culture Connect. Every count on this page is checked in tools/computers/test/learn.test.mjs.
模型與程式由人師教育協會自行繪製、撰寫；這一頁出現的每一個數字都有自動測試（tools/computers/test/learn.test.mjs）。
Reading · 英文閱讀
15
Unit 15: How Does a Computer Learn?
Every program so far in this course followed rules that a person wrote. The robot turned left because someone wrote “turn left.” But some jobs are very hard to write rules for. How would you write the rule for telling an apple from an orange in a photo, or for reading messy handwriting? There are too many cases to list. For jobs like these, there is another way.
到目前為止，這門課裡每一個程式照的都是人寫好的規則。機器人會左轉，是因為有人寫了「左轉」。可是有些工作，很難替它寫規則。要怎麼寫出「從照片裡分辨蘋果和橘子」的規則？怎麼寫出「讀懂潦草字跡」的規則？情況多到列不完。這一類的工作，有另一種做法。
Instead of rules, give the computer examples. Each example comes with the right answer: this one is an apple, that one is an orange. Picture the fruits lying on a table, placed by how heavy and how red they are, and an upright wall across the table. The wall is the computer’s rule: apples on one side, oranges on the other. At the start the wall is in a bad place.
不給規則，改給電腦例子。每個例子都附上正確答案：這個是蘋果，那個是橘子。想像這些水果擺在一張桌上，依照有多重、有多紅來放，桌上還立著一片隔板。隔板就是電腦的規則：蘋果在一邊，橘子在另一邊。一開始，隔板擺在很差的位置。
Now the computer looks at one example. If the wall already has that fruit on the correct side, nothing changes. If the fruit is on the wrong side, the computer moves the wall a little, toward getting that one right. Then it looks at the next example. In our model there are 24 fruits. After looking at 40 examples, the wall has moved 5 times, and every fruit is on the correct side. Nobody told the computer where the wall should go.
現在電腦看一個例子。如果隔板已經把那顆水果分在對的一邊，什麼都不改。如果那顆水果在錯的那一邊，電腦就把隔板挪一點，往「把這一顆分對」的方向。然後看下一個例子。我們的模型裡有 24 顆水果。看了 40 個例子之後，隔板挪了 5 次，每一顆水果都在對的那一邊。沒有人告訴電腦隔板該放在哪裡。
This is called machine learning, and the steps of looking and adjusting are called training. Notice that the computer is still doing exactly what a program says, as in Lesson 9. The program just does not say “this is an apple.” It says “when you are wrong, adjust a little.” That is why the word “learn” is in quotation marks. The computer is adjusting numbers. It does not understand what a fruit is.
這叫做機器學習（machine learning），那些「看一看、調一調」的步驟叫做訓練（training）。請注意，電腦還是在一絲不差地照程式說的做，和第九課一樣。只是這個程式寫的不是「這是蘋果」，而是「分錯的時候，調整一點」。所以「學習」這兩個字要加引號。電腦是在調整數字，它並不了解水果是什麼。
Because everything comes from the examples, the examples matter a great deal. Give the computer only four, and it is satisfied too soon: the wall fits those four, but it gets some new fruits wrong. Give it examples with wrong answers, and the wall never settles, because no wall can agree with all of them. The computer cannot tell which answers to doubt.
既然一切都來自例子，例子就非常重要。只給電腦四個例子，它太早就滿意了：隔板符合那四顆，可是有些新水果它會分錯。給它答案錯誤的例子，隔板就永遠定不下來，因為沒有任何一片隔板能同時符合全部。電腦分不出該懷疑哪幾個答案。
Real systems do not use one wall. They use neural networks, with a great many numbers to adjust, and they are used to recognize handwriting, faces, and speech. What such a network knows is spread through those numbers. It is not written out anywhere in the program. For a large network, it can be extremely difficult to follow exactly how an answer was reached. So people test it on new cases it has never seen, and they keep checking what it says.
真實的系統用的不是一片隔板，而是類神經網路（neural network），要調整的數非常多；它們被用來辨認字跡、人臉和語音。這樣一個網路「知道」的東西，分散在那些數裡面，並沒有寫在程式的任何一個地方。網路很大的時候，要弄清楚某個答案到底是怎麼得出來的，可能極為困難。所以人們會拿它沒看過的新例子來測試它，並且持續檢查它說的話。
Read aloud · 課文朗讀
Questions · 閱讀理解
How is machine learning different from the programs in Lesson 9?
In Lesson 9 a person wrote the rule. In machine learning nobody writes the rule; the computer is given examples with answers and adjusts itself until it gets them right.
What does the computer do when the wall has a fruit on the wrong side? What does it do when the fruit is on the correct side?
When the fruit is on the wrong side, it moves the wall a little toward getting that one right. When the fruit is on the correct side, it changes nothing.
Why do the examples matter so much? Give two ways they can go wrong.
Everything the computer ends up with comes from the examples. With too few, it is satisfied too soon and gets new cases wrong. With wrong answers among them, it can never fit them all, and it cannot tell which ones to doubt.
Words & Phrases · 生字及片語
example(n.)例子Show me an example, and I will understand.給我看一個例子，我就懂了。train(v.)訓練The team trained the program with thousands of photos.這個團隊用幾千張照片訓練那個程式。adjust(v.)調整Adjust the mirror a little to the left.把鏡子往左調整一點。mistake(n.)錯誤It learns from each mistake.它從每一次錯誤裡學到東西。improve(v.)進步；改善Your writing will improve with practice.多練習，你的寫作就會進步。recognize(v.)辨認；認出The phone can recognize your face.手機認得出你的臉。label(n.)標籤；標示The label on this box is wrong.這個箱子上的標籤是錯的。test(v.)測試Test it on something it has never seen.拿它從來沒看過的東西來測試它。
Quick Check · 小測驗
1. In machine learning, what is the computer given?
2. In the model, what happens when the wall has a fruit on the correct side?
3. Why is “learn” in quotation marks in this lesson?
4. What goes wrong when the computer is given only four examples?
5. According to the reading, why do people test a neural network on new cases?
Try It in the Model · 在模型裡試試
Four things to watch 四件值得看的事
Turn the model with your finger or mouse to look straight down on the table. From above, the wall is a line.
用手指或滑鼠轉動模型，從正上方往下看桌面。從上面看，隔板就是一條線。
24 good examples24 個好例子
At the start 17 of the 24 fruits are on the wrong side. Watch the “wrong now” counter. Does it go down smoothly?
一開始 24 顆裡有 17 顆在錯的那一邊。注意看「現在分錯幾顆」。它是平順地往下降嗎？
Only 4 examples只有 4 個例子
The computer stops because nothing is wrong any more. Then look at the small fruits with rings. Which rings are red?
電腦停下來了，因為沒有東西分錯。接著看那些有白圈的小水果：哪幾個圈變紅了？
Four wrong labels四個標錯的答案
Four fruits carry the wrong answer. Watch how often the wall moves now. Does it ever come to rest?
有四顆水果的答案是錯的。注意隔板現在多常移動。它有停下來的時候嗎？
Try It · 自己試試
Be the Teacher 換你當老師
Here the examples come from you. Choose a fruit and tap the square to put it down: across is weight, up is redness. Put apples in one area and oranges in another, then press Learn and see where the computer draws its line. Then try to fool it.
這一次，例子由你來給。選一種水果，點方框把它放下去：橫的是重量，直的是有多紅。把蘋果放在一區、橘子放在另一區，按「學」，看電腦把線畫在哪裡。然後試著騙騙它。
heavier → 越重redder ↑ 越紅
Which fruit to put down · 要放哪一種
- Fruits on the table · 桌上幾顆
- 0
- Examples looked at · 看了幾個例子
- —
- Still wrong · 還分錯幾顆
- —
The computer here can only draw one straight line, and it stops after 400 examples. A fruit with a dark outline is one the line gets wrong. Nothing you put down here is sent anywhere.這裡的電腦只會畫一條直線，而且看了 400 個例子就停。外框變深的水果，就是那條線分錯的。你在這裡放的東西不會送到任何地方。
- Put down just one apple and one orange, and press Learn. The line is “right,” but would you trust it?只放一顆蘋果、一顆橘子，按「學」。線是「對」的，可是你信得過它嗎？
- Put one orange in the middle of the apples. What happens to the line, and to the count of fruits still wrong?在一堆蘋果中間放一顆橘子。線怎麼了？「還分錯幾顆」又怎麼了？
- Put apples in a ring with oranges in the center. No straight line can separate them. A real system needs something more than one wall.把蘋果排成一圈，橘子放在中間。沒有任何一條直線分得開它們。真的系統需要的不只是一片隔板。
Game · 小遊戲
Written Rule, or Learned from Examples? 人寫的規則，還是從例子學的？
Eight jobs a computer can do. For each one, decide: did a person write the rule step by step, or did the computer adjust itself by looking at examples?
電腦會做的八件事。每一件都想一想：是人一步一步把規則寫好的，還是電腦看著例子自己調整出來的？
Right on the first try · 第一次就答對 0 / 0
The four “learned” answers follow Britannica’s articles on neural networks and machine learning. The four “rule” answers are from earlier lessons of this course.四個「從例子學的」答案依照大英百科「類神經網路」與「機器學習」條目；四個「人寫的規則」來自這門課前面的課。
Stories Behind Machine Learning · 機器學習的故事
An idea from the brain, the perceptron, teaching with answers, and a box that is hard to see into 來自大腦的想法、感知器、附答案的教法、一個很難看進去的盒子
Four short stories about machines that adjust themselves.
關於「會自己調整的機器」的四個小故事。
1943, 1954
An idea borrowed from the brain向大腦借來的想法
Britannica describes a neural network as a computer program that works in a way inspired by the natural neural network in the brain. The theory was developed in 1943 by the neurophysiologist Warren McCulloch and the mathematician Walter Pitts. In 1954 Belmont Farley and Wesley Clark of the Massachusetts Institute of Technology succeeded in running the first simple neural network.
大英百科說，類神經網路是一種電腦程式，運作的方式受到大腦裡天然神經網路的啟發。它的理論基礎在 1943 年由神經生理學家麥卡洛克（Warren McCulloch）和數學家皮茨（Walter Pitts）提出。1954 年，麻省理工學院的法利（Belmont Farley）和克拉克（Wesley Clark）成功執行了第一個簡單的類神經網路。
1957
The perceptron感知器
The moving wall in this lesson is a perceptron. According to Britannica, perceptrons are a type of artificial neural network investigated by Frank Rosenblatt, beginning in 1957, at the Cornell Aeronautical Laboratory. Rosenblatt and his followers called their approach connectionism, to stress how important the making and changing of connections is in learning. Researchers still use that word.
這一課那片會動的隔板，就是一個感知器。根據大英百科，感知器是一種人工類神經網路，由羅森布拉特（Frank Rosenblatt）從 1957 年起在康乃爾航空實驗室研究。羅森布拉特和追隨他的人把這種做法叫做連結主義（connectionism），強調「建立和改變連結」在學習中有多重要。這個詞今天的研究者還在用。
Training · 訓練
Teaching with the answers attached附著答案的教法
What you watched has a name: supervised learning. Britannica explains that each training example contains both the input data and the desired output, just as each fruit on the table came with its label. As soon as the network performs well enough on additional test cases, it can be applied to new ones. The small ringed fruits in the model are test cases of exactly that kind.
你剛才看到的做法有個名字：監督式學習（supervised learning）。大英百科解釋，每一個訓練用的例子都同時包含輸入的資料和想要的輸出——就像桌上每一顆水果都附著它的標籤。等到網路在另外的測試案例上表現得夠好，就可以用在新的案例上。模型裡那些有白圈的小水果，正是這樣的測試案例。
Hard to follow · 很難追
Knowing without a written rule沒有寫下來的規則
In an ordinary program you can read the rule. In a neural network, Britannica notes, knowledge is distributed throughout the network itself rather than being explicitly written into the program. It adds that for large networks it can be extremely difficult to follow exactly how an output was determined. With one wall you can see the whole rule at a glance. With a great many numbers, nobody can.
一般的程式，規則讀得出來。大英百科指出，在類神經網路裡，知識是分散在整個網路之中，而不是明白地寫進程式。它還說，網路很大的時候，要弄清楚某個輸出到底是怎麼決定的，可能極為困難。只有一片隔板的時候，整條規則一眼就看完；數多到不行的時候，誰都看不完。
Myth vs. Fact · 常見迷思
Four things people get wrong 四個常見的誤會
✗ Myth · 迷思A computer that learns understands what it has learned.會學習的電腦，了解它學到的東西。
✓ Fact · 事實In the model it adjusts three numbers by a fixed rule. It can separate apples from oranges without any idea of what a fruit is.在模型裡，它是照一條固定的規則調整三個數。它分得開蘋果和橘子，卻完全不知道水果是什麼。
✗ Myth · 迷思With machine learning, nobody has to write a program.有了機器學習，就不必有人寫程式。
✓ Fact · 事實Someone still writes the program that says how to adjust. What nobody writes is the final rule.還是得有人寫出「怎麼調整」的程式。沒有人寫的，是最後得到的那條規則。
✗ Myth · 迷思If it gets every example right, it has learned well.每一個例子都分對了，就表示學得好。
✓ Fact · 事實With only 4 examples the model gets all 4 right and then gets 2 of 12 new fruits wrong. It has to be tested on cases it never saw.只給 4 個例子的時候，模型 4 個都分對，卻把 12 顆新水果裡的 2 顆分錯。一定要拿它沒看過的案例來測試。
✗ Myth · 迷思A learning computer can work out which of its examples are wrong.會學習的電腦，看得出哪些例子是錯的。
✓ Fact · 事實In the model it treats every label as true. Four wrong labels keep the wall moving, and it never finds out which four they are.在模型裡，它把每一個標籤都當成真的。四個標錯的答案讓隔板動個不停，而它始終不知道是哪四個。
Remember It · 記憶口訣
Machine learning in a sentence 一句話記住機器學習
Examples, not rules給例子，不給規則
The computer finds the rule itself.
規則由電腦自己找。
Wrong? Move a little錯了？挪一點
Right? Change nothing.
對了？什麼都不改。
Only as good as the examples最多和例子一樣好
Too few, or mislabeled, and it goes wrong.
例子太少或標錯，它就學歪。
Test on what it has not seen拿沒看過的來測
Getting the examples right is not enough.
例子都分對還不夠。
Before You Start · 動手之前
A few reminders 幾個提醒
- In the rope activity, lay the rope flat on the floor and step over it carefully. Nobody pulls it tight while others are walking.玩繩子活動的時候，繩子平放在地上，跨過去時要小心。有人在走動的時候，誰都不要把繩子拉緊。
- An answer from a system that learned from examples can be wrong, even when it sounds sure. Check it against something you trust.從例子學出來的系統給的答案可能是錯的，即使它說得很肯定。要和你信得過的東西核對。
- Do not give photos of yourself or your classmates, or other private information, to a program just to see what it does.不要為了看看會怎樣，就把自己或同學的照片、其他私人資訊交給某個程式。
- Use a school or family computer only with permission.用學校或家裡的電腦要先得到同意。
Classroom Activity 1 · 課堂活動一
The Rope That Learns 會學習的繩子
You need · 準備材料
- An open floor, with tape marking a large square一塊空地，用膠帶標出一個大正方形
- Red cards and orange cards, one for each student紅色卡片和橘色卡片，每位同學一張
- A long rope, and two students to hold its ends一條長繩，和兩位負責拿繩子兩端的同學
Steps · 步驟
- Students with red cards stand mostly toward one corner of the square; students with orange cards stand mostly toward the opposite corner. Each holds the card up.拿紅卡的同學大多站在正方形的一個角落附近，拿橘卡的大多站在對角。每個人把卡片舉起來。
- The rope holders lay the rope across the square in a bad place and name the two sides: “red side” and “orange side.”拿繩子的兩位同學先把繩子橫放在一個很差的位置，並說出兩邊的名字：「red side」和「orange side」。
- One student is the computer and may say only this, about one classmate at a time: “Right side: no change,” or “Wrong side: move the rope a little.” The holders then shift the rope slightly toward putting that classmate on the correct side.一位同學當電腦，一次只看一位同學，而且只能說這兩句話之一：「Right side: no change」或「Wrong side: move the rope a little」。拿繩子的人就把繩子稍微挪動，往「讓那位同學站到對的一邊」的方向。
- Keep going around the class. Count how many classmates the computer looks at before a whole round passes with no change.一輪一輪地看下去。數一數電腦看了幾位同學之後，才有一整輪都不用再動。
Then swap two students’ cards without telling the computer and run it again. Ask what the computer would need in order to notice the swap. In Taiwan’s curriculum, machine learning is not named in the junior high Information Technology content. It appears in senior high: 資 D-V-2, basic concepts of data mining and machine learning.
接著偷偷把兩位同學的卡片對調，不告訴電腦，再跑一次。問問大家：電腦需要什麼，才發現得了有人被對調？在十二年國教課綱裡，國中「資訊科技」的學習內容沒有列出機器學習；它出現在高中：資 D-V-2 資料探勘與機器學習的基本概念。
Classroom Activity 2 · 課堂活動二
Check It with the Page 用這一頁來驗證
You need · 準備材料
- A computer, tablet, or phone with this page open一台開著這一頁的電腦、平板或手機
- The “Be the Teacher” tool on this page這一頁的「換你當老師」小工具
- Paper and a pencil紙和鉛筆
Steps · 步驟
- Put down six apples and six oranges in two clear groups. Press Learn. Write down how many examples the computer looked at.放六顆蘋果和六顆橘子，分成清楚的兩堆。按「學」。記下電腦看了幾個例子。
- Clear the table. Put down the same two groups, but closer together. Does it need more examples or fewer?清空桌面。放同樣的兩堆，但靠近一點。它需要看更多例子，還是更少？
- Now add one orange deep inside the apples and press Learn. Write down what the counters say.現在在蘋果堆的深處加一顆橘子，按「學」。把計數顯示的數字記下來。
- Explain in one sentence, in English, why the computer could not finish.用一句英文說明，為什麼電腦沒辦法學完。
There is no single right count for the first two steps; it depends on where you put the fruit. That is the point: what the computer ends up with depends on its examples. For the last step, a good sentence is: “No straight line can put every fruit on the correct side.”
前兩步沒有唯一的正確數字，要看你把水果放在哪裡。重點正在這裡：電腦最後學到什麼，取決於它的例子。最後一步，一個好的句子是：「No straight line can put every fruit on the correct side.」
Go Further · 延伸閱讀
Rules written by people, clear methods, and what comes next 人寫的規則、清楚的方法，以及下一課
Previous Lesson · Lesson 14 · 上一課・第十四課How Does a Computer Keep a Secret?電腦怎麼保守祕密？There the rule was public and only a number was secret. Here even the rule is not written by anyone.那一課的規則是公開的，保密的只有一個數。這一課連規則都不是誰寫的。Go to the lesson · 前往這一課 →Lesson 9 · 第九課What Is a Program?程式是什麼？The computer does exactly what the program says. That is still true here: the program says how to adjust.電腦完全照程式說的做。這在這一課仍然成立：程式說的是「怎麼調整」。Go to the lesson · 前往這一課 →Lesson 10 · 第十課What Is an Algorithm?演算法是什麼？“Look, compare, adjust, repeat” is an algorithm too. It is a method for finding a rule.「看、比、調、再來一次」也是一個演算法。它是一個用來找規則的方法。Go to the lesson · 前往這一課 →Next Lesson · Lesson 16 · 下一課・第十六課How Did Computers Get So Small?電腦怎麼越變越小？The last lesson. Everything in this course ran on switches. Next: how the switches shrank from the size of a room to the size of a fingernail.最後一課。這門課裡的一切都跑在開關上。下一課：這些開關怎麼從一個房間那麼大，縮到指甲那麼小。Go to the lesson · 前往這一課 →
Sources · 資料出處
Facts on this page were checked against these sources (October 2026); the fruits and their numbers are made up for the example, and every count is tested · 本頁的事實依下列資料查證（2026 年 10 月）；水果和它們的數字是為了舉例編的，每一個數字都有測試
- Machine learning is a “discipline concerned with the implementation of computer software that can learn autonomously”; it is a subject within artificial intelligence; among the most common approaches is the use of artificial neural networks機器學習是一門研究如何實作出能自主學習的電腦軟體的學問；它屬於人工智慧；最常見的做法之一是使用人工類神經網路William L. Hosch, “machine learning,” Encyclopaedia Britannica ↗
- Machine learning “is particularly useful in situations where it is impractical to write detailed instructions for every possible scenario”; many image classification systems are trained to recognize and categorize images在不可能替每一種情況都寫出詳細指令的場合，機器學習特別有用；許多影像分類系統經過訓練來辨認並分類影像The Editors of Encyclopaedia Britannica, “What Is Machine Learning?” ↗
- A neural network is “a computer program that operates in a manner inspired by the natural neural network in the brain”; theory by McCulloch and Pitts in 1943; first simple network run by Farley and Clark in 1954; used to recognize handwriting; “knowledge of its domain is distributed throughout the network itself rather than being explicitly written into the program”; it learns by adjusting the weights of connections; facial recognition; in supervised learning “each training example contains the values of both the input data and the desired output,” and the network is then checked on additional test cases; “for large networks it can be extremely difficult to follow exactly how an output was determined”; speech recognition類神經網路是一種運作方式受大腦天然神經網路啟發的電腦程式；1943 年麥卡洛克與皮茨提出理論；1954 年法利與克拉克執行了第一個簡單的網路；用於辨認手寫字跡；它對領域的知識分散在整個網路之中，而不是明白地寫進程式；它靠調整連結的權重來學習；人臉辨識；監督式學習中每個訓練例子都包含輸入資料與想要的輸出，之後再用另外的測試案例檢查；網路很大時要弄清楚輸出怎麼決定的可能極為困難；語音辨識Vladimir Zwass, “neural network,” Encyclopaedia Britannica ↗
- Perceptrons are “a type of artificial neural network investigated by Frank Rosenblatt, beginning in 1957, at the Cornell Aeronautical Laboratory”; Rosenblatt and his followers called their approach connectionism感知器是一種人工類神經網路，由羅森布拉特從 1957 年起在康乃爾航空實驗室研究；他和追隨者把這種做法稱為連結主義B.J. Copeland, “perceptrons,” Encyclopaedia Britannica ↗
- 資 D-V-2, basic concepts of data mining and machine learning, in the senior high content (p. 11; notes on p. 45). The junior high Information Technology content does not name machine learning or artificial intelligence資 D-V-2 資料探勘與機器學習的基本概念，屬於高中階段的學習內容（第 11 頁；附錄三的說明在第 45 頁）。國中「資訊科技」的學習內容沒有列出機器學習或人工智慧教育部《十二年國民基本教育課程綱要 國民中學暨普通型高級中等學校 科技領域》（2018） ↗
- Taiwan terms: artificial intelligence 人工智慧, neural network 類神經網路, perceptron 感知器; 機器學習 is the curriculum’s own word台灣用語：artificial intelligence 人工智慧、neural network 類神經網路、perceptron 感知器；「機器學習」是課綱的用詞國家教育研究院《樂詞網》 ↗
- The model: 24 made-up fruits, 12 apples and 12 oranges, and 12 new ones for checking. At the start 17 of 24 are on the wrong side. With the 24 good examples, the wall settles after 40 examples and 5 moves, with 0 wrong and 12 of 12 new fruits right. With only 4 examples it settles after 14 examples and 4 moves, and gets 10 of 12 new fruits right. With 4 wrong labels it is stopped at 240 examples after 92 moves, still wrong on 5, and it never settles模型：24 顆編出來的水果（蘋果、橘子各 12 顆），另有 12 顆新水果用來檢查。一開始 24 顆裡有 17 顆在錯的那一邊。24 個好例子：看了 40 個例子、挪了 5 次就定下來，0 顆錯，新水果 12 顆全對。只有 4 個例子：看了 14 個、挪了 4 次就定下來，新水果 12 顆對 10 顆。4 個答案標錯：看到第 240 個例子被停下來，挪了 92 次，還錯 5 顆，而且永遠定不下來Our own illustration, checked by tools/computers/test/learn.test.mjs ↗