Artificial Intelligence: Computer Science 9618 (Cambridge International AS & A Level)

A video streaming service holds a record for each of 5,000 past subscribers. Each record already states the genres that subscriber watched and a label showing whether that subscriber went on to cancel their subscription or not. The company wants to train a machine learning model on this labelled data so that it can predict whether a new subscriber is likely to cancel.

Which category of machine learning is being described?

A warehouse uses an automated delivery robot that moves between five junctions, A to E, connected by fixed two-way paths. The table below gives the distance, in metres, of each path.

The robot starts at junction A and must reach the packing station at junction E, using Dijkstra's algorithm to find the shortest total distance.

(a) Describe, in the context of this scenario, what the nodes and the edges of a graph represent. [2]

(b) Trace Dijkstra's algorithm starting at A. Show the order in which junctions are settled (given a final, shortest distance from A), and state the shortest distance from A to each junction. [4]

(c) State the shortest path from A to E, and its total distance. [2]

A photo-sharing app uses an artificial neural network to decide whether an uploaded photo shows a cat or does not show a cat. The network is trained on a large set of existing photos, each already labelled cat or no cat.

(a) Describe the three types of layer found in a typical artificial neural network, stating what each one does in this scenario. [2]

(b) State what a "weight" is in an artificial neural network, and explain what changing a weight allows the network to do. [2]

(c) Describe how back propagation of errors is used, during training, to improve the accuracy of the network's future predictions. [2]

A ride-hailing app models the road network of a district as a graph: junctions are nodes, and roads are edges weighted with the typical drive time, in minutes, between two junctions. A driver currently at junction S needs the fastest route to a single passenger waiting at junction G. The app can search this graph using either Dijkstra's algorithm or the A* algorithm.

(a) Describe one key difference between how Dijkstra's algorithm and the A* algorithm decide which node to explore next when searching a graph. [2]

(b) Explain why, for this single-destination journey from S to G, the A* algorithm would typically need to explore fewer junctions than Dijkstra's algorithm before finding the driver's route. [2]

(c) The app also has a separate feature that must work out the fastest drive time from junction S to every other junction in the district, not just one destination. State one reason why Dijkstra's algorithm may be more suitable than A* for this separate feature. [2]

An agricultural technology company is building three different AI systems. For each system, state which category of machine learning is being used, supervised learning, unsupervised learning or reinforcement learning, and give one reason for your choice.

(a) System 1 predicts a crop's expected yield, in tonnes per hectare, from historical records of rainfall and soil quality, where every past record already states the yield that was actually achieved. [2]

(b) System 2 sorts thousands of unlabelled satellite images of fields into groups of similar-looking images, without being told in advance what the groups should be or how many groups to form. [2]

(c) System 3 controls a crop-spraying drone that chooses its own flight actions. It receives a positive reward when it applies fertiliser evenly across a field, and a penalty when it wastes fertiliser or misses part of the field, without being told the correct action for any given situation in advance. [2]

A speech-recognition company builds a system that converts spoken audio into written text. The system uses an artificial neural network with 40 hidden layers, trained on millions of labelled audio clips paired with their correct text transcriptions. Each successive hidden layer learns to recognise increasingly complex patterns in the audio, building on the simpler patterns learned by the layer before it.

Which term most precisely describes this system, given that it uses an artificial neural network with many hidden layers to learn increasingly complex patterns from the data?

A search-and-rescue drone models its flying area as a graph. The nodes S, P, Q, R and T are waypoints, where S is the drone's launch point and T is a stranded hiker's last known location. The edges are direct flight paths between waypoints, weighted with the flight time, in minutes, needed to fly directly between them.

The drone's onboard sensor also estimates, for each waypoint, the remaining flight time to reach T in a straight line. These heuristic estimates, h(n), are:

The drone uses the A* algorithm, starting at S, to find its route to T.

(a) State the two values that A* combines to calculate f(n) for a node n, and explain what each one represents in this scenario. [2]

(b) Trace the A* algorithm from S. For each node expanded, state the order of expansion and the values of g(n), h(n) and f(n) used to expand it. [5]

(c) State the route the drone follows from S to T, and its total flight time. [1]

A fitness-tracking company wants to predict a user's resting heart rate, in beats per minute, from their age. It trains a regression method on a large set of past records, each already stating a user's age and their measured resting heart rate.

(a) Describe what a regression method does with this training data in order to build a model that can make predictions. [2]

(b) Once trained, the model is given the age of a new user whose resting heart rate has not been measured. Explain how the trained model uses this to produce its predicted resting heart rate for the new user. [2]

(c) A separate team at the company instead wants to predict whether a user should be labelled "active" or "inactive", not a numeric value. Explain why a regression method, on its own, would not be an appropriate choice for this second task. [2]

Two artificial neural networks are built for the same handwriting-recognition task, and both are trained using back propagation of errors on the same set of labelled handwriting samples.

• Network 1 has a single hidden layer.
• Network 2 has twenty hidden layers, and is described by its developers as using deep learning.

(a) State the specific structural feature that makes Network 2 an example of deep learning, which Network 1 does not have. [1]

(b) Explain one advantage the many hidden layers in Network 2 can give it, compared with Network 1, when learning to recognise complex handwriting patterns. [2]

(c) Explain how training Network 2 differs from training Network 1 in terms of how far the error found at the output must be propagated back through the network. [2]

A phone manufacturer records, for thousands of batteries, the number of charge cycles each battery has completed and its remaining capacity, as a percentage of its original capacity. It trains a model on these paired records to predict the remaining capacity of a new battery from its number of completed charge cycles.

Which method is being used by this model to make its prediction?