Level 4
Separate input, output and data, and lay out the order in which you would build.
Who can apply
Open to applicants aged 14 and over.
Required programming language
Required at every level, from Level 5 to Level 1.
Shell Scripting
The single most important language for working with AI.
The is useful for running , managing files, and automating development. can also run through , apps, notebooks, and . A is one useful interface, not a requirement for every AI user.
Learn shell programmingThe one question, at every level
You want to build an artificial intelligence completely from scratch. Design how you would build it.
Every level from 5 to 1 answers this same question. What changes is the depth expected of the answer, and the constraints attached to it.
What to learn
- Decide what goes in and what must come out.
- Tell apart what you solve with rules and what you learn from data.
- Break the build into steps and put them in order.
How you are marked
- Are the input and the output stated clearly?
- Did you say where the data would come from?
- Do the steps follow one another in a workable order?
Exam format
Stage 1 is a written paper at the venue; Stage 2 is an interview followed by a practical build. Both stages use the same question.
Held at a designated exam venue. You write your answers on the exam paper by hand. This stage tests the design you can put into words.
An examiner interviews you about your design, then you build the for real. This stage tests whether the plan on paper actually works in your hands.
Practice questions
The official exam gives you just the one common question above. The 20 questions below are practice questions that check the words, names and facts you need to know before you write that design. Every one is a short-answer question with a single correct answer: read the description, write your answer, then open the answer to check. They are not official questions.
Data and features
-
1
When you split your data, which data do you keep out of training and use only at the end to check how good the model really is?
Show answer
Answer Test data
If you test on the data you trained on, you cannot tell whether the model learned or just memorized. So you keep data it has never seen.
-
2
In a model that predicts house prices, what do you call each property of the input that the model uses to decide, such as size, number of rooms or distance to the station?
Show answer
Answer Feature
In table data, one column is one feature. Choosing which features to use is the first step of a design.
-
3
What is the way of learning that finds structure in data by itself, such as grouping similar items, without any answer labels?
Show answer
Answer Unsupervised learning
It is the partner of supervised learning, where the model learns to match labeled answers.
-
4
What do you call grouping similar data together, such as splitting customers into groups with similar buying habits?
Show answer
Answer Clustering
It is a typical unsupervised learning task. People name the groups afterwards.
-
5
What do you call data stored in a fixed frame of rows and columns, like a spreadsheet? (It is the opposite of text, photos and sound, which have no fixed frame.)
Show answer
Answer Structured data
Text, photos and sound without a fixed frame are called unstructured data.
Rules and ways of learning
-
6
What was the early AI approach where people wrote rules like "if this, then do that" by hand so the computer could judge like an expert?
Show answer
Answer Expert system
It was widely used in the 1970s and 80s. People had to write every rule, so the more rules there were, the harder it was to build and fix.
-
7
What do you call a machine learning problem whose answer is one of a few fixed categories, like "spam / not spam"?
Show answer
Answer Classification
With two possible answers it is binary classification; with three or more it is multi-class classification.
-
8
What do you call a machine learning problem whose answer is a continuous number, like a house price or tomorrow's temperature?
Show answer
Answer Regression
Classification predicts a category; regression predicts a number. Decide the input and output first and you will see which one you need.
-
9
What is the way of learning where the model tries actions, looks at the reward (score) it gets, and learns better actions by itself? AlphaGo grew stronger this way by playing against itself.
Show answer
Answer Reinforcement learning
It differs from supervised learning because it gets rewards instead of answer labels.
-
10
What do you call it when a model gets almost everything right on its training data but often gets new data wrong, because it memorized the training data?
Show answer
Answer Overfitting
The opposite, when the model cannot even fit its training data well, is called underfitting.
Steps of training and evaluation
-
11
What do you call the function that turns how wrong the model is into a single number, so that training can work to make that number smaller?
Show answer
Answer Loss function
Training is really changing the weights so that the loss goes down.
-
12
What do you call one full pass in which the model trains on the whole training set once, from start to finish?
Show answer
Answer Epoch
You usually train for several epochs and watch whether the loss keeps going down.
-
13
What is the step before training where you fix missing, duplicated or wrong values, match units and so on, to put the data into a shape fit for training?
Show answer
Answer Data preprocessing
Messy data makes a messy model. A large share of real development time goes into this step.
-
14
Like getting 90% when you answer 90 out of 100 questions correctly, what is the most basic metric that rates a model by the share of answers it gets right?
Show answer
Answer Accuracy
It is simple and handy, but it can make a model look better than it is when one answer makes up most of the data.
-
15
What do you call the data kept apart from both the training and test data, used during training to choose settings or decide when to stop?
Show answer
Answer Validation data
Test data is used only once at the very end; choosing settings is done with validation data.
The shell and tools
-
16
In the shell, write the command that shows the path of the folder you are in now (the current working directory).
Show answer
Answer pwd
It is short for print working directory.
-
17
In the shell, write the command that prints the contents of a file to the screen. It is also used to join several files together.
Show answer
Answer cat
The name comes from concatenate (to join together). For example: cat data.csv
-
18
In the shell, write the command that finds and shows only the lines of a file that contain certain text.
Show answer
Answer grep
For example, grep error log.txt shows only the lines in log.txt that contain error.
-
19
Write the three-letter name of the text file format that stores table data with values separated by commas — the most common format for machine learning data.
Show answer
Answer CSV (Comma-Separated Values)
Each line is one row, and each comma-separated value is one column. You can even open it in a plain text editor.
-
20
Write the name of the programming language most widely used for data analysis and machine learning, which marks blocks of code with indentation instead of curly braces.
Show answer
Answer Python
Most AI tools, such as NumPy, scikit-learn and PyTorch, are used from Python.