Educerie · IB Diploma · Computer Science

Theme A — Concepts of computer science

Theme A is the "how does the machine work" half of the course, examined in Paper 1. It runs from individual bits up to a trained model, and Section B of the paper applies it all to the pre-seen case study.

A1 Computer fundamentals

173 → 10101101      (128 + 32 + 8 + 4 + 1)
10101101 → AD       (1010 = A, 1101 = D)

A2 Networks

A network shares resources between connected devices. LAN is local; WAN spans large distances; the internet is a network of networks.

Topologies — star (all devices to a central switch: robust, but the switch is a single point of failure), mesh (many redundant paths: resilient, expensive), bus and ring (largely historic).

Layered models matter because each layer can change without disturbing the others. In the TCP/IP model: application, transport, internet, link. Two protocols to distinguish:

• TCP — connection-oriented, guarantees delivery and ordering by acknowledging packets and resending lost ones. Used where correctness matters: web pages, email, file transfer.
• UDP — connectionless, no guarantee of delivery or order, but far less overhead. Used where timeliness beats completeness: live video, voice, games.

Packet switching breaks data into packets, each with a header carrying source and destination addresses and a sequence number. Packets may travel different routes and arrive out of order; the receiver reassembles them by sequence number. This is what makes the internet resilient — a failed link reroutes rather than disconnects.

Security. Encryption makes intercepted data unreadable without the key: symmetric uses one shared key (fast, but the key must be distributed safely), asymmetric uses a public key to encrypt and a private key to decrypt (solves distribution, but is slower). In practice the two are combined — asymmetric encryption is used to exchange a symmetric key, which then does the bulk work. A firewall filters traffic by rules; authentication confirms identity.

A3 Databases

A relational database stores data in tables of rows (records) and columns (fields).

• Primary key — uniquely identifies each record in a table.
• Foreign key — a field referencing another table's primary key, creating the relationship.
• Referential integrity — a foreign key must match an existing primary key, so you cannot reference a record that does not exist.

Normalisation removes redundancy, which is what prevents update anomalies:

• 1NF — atomic values only; no repeating groups.
• 2NF — 1NF, plus every non-key field depends on the whole primary key.
• 3NF — 2NF, plus no non-key field depends on another non-key field.

The reason to normalise is not tidiness. Storing a customer's address in every order means changing it requires updating every row, and missing one leaves the database contradicting itself.

SQL basics: SELECT fields FROM table WHERE condition, with JOIN to combine tables on a key, and GROUP BY with aggregates such as COUNT and SUM.

Transactions must be ACID — atomic (all or nothing), consistent, isolated from other transactions, and durable once committed. The standard illustration is a bank transfer: debiting one account and crediting another must both happen or neither.

A4 Machine learning

A machine learning system learns patterns from data rather than following rules written by a programmer.

• Supervised learning uses labelled data — inputs paired with correct outputs — to learn a mapping. Classification predicts a category, regression a continuous value.
• Unsupervised learning uses unlabelled data and finds structure in it, chiefly by clustering.
• Reinforcement learning learns through reward and penalty from interacting with an environment.

Data is split into a training set, used to fit the model, and a test set, held back to measure performance on data the model has never seen. Evaluating on the training set tells you nothing, because the model has already seen the answers.

Overfitting is when a model learns the noise in the training data as though it were signal: it performs excellently on training data and poorly on new data. Underfitting is a model too simple to capture the real pattern, performing poorly on both.

Bias enters through the data. If the training set under-represents a group, the model performs worse for that group — not because of a coding error, but because the data taught it to. This is examinable as an ethical issue, and the strongest answers identify the source of the bias rather than merely deploring the outcome.

What actually loses marks in this theme

1. Reading division remainders downwards instead of upwards in denary-to-binary conversion.
2. Giving the fetch–decode–execute cycle out of order, or omitting the PC increment.
3. Confusing MAR and MDR.
4. Describing TCP and UDP separately when the question said distinguish.
5. Saying packets "arrive in order" — they do not; they are reassembled in order.
6. Confusing primary and foreign keys.
7. Justifying normalisation as "tidier" rather than as preventing update anomalies.
8. Defining supervised learning without the word labelled.
9. Explaining overfitting as "the model is wrong" rather than as excellent training performance with poor generalisation.

Educerie · written from the published IB syllabus structure for Computer Science Theme A, first assessment 2027. Original text. Last reviewed 5 September 2026.