IBDP Computer Science A4.1 Machine learning fundamentals HL Paper 2 - New Syllabus

IBDP Computer Science HL Paper 2 - All Chapters

Question 

A genetic algorithm will be used to develop a self-driving car’s antenna to communicate with traffic lights and other vehicles.

(a) Describe the steps a genetic algorithm would use to develop an antenna for a self-driving car. [3]

Some aspects of self-driving cars are designed using neural networks.

(b) Sketch a simple block diagram, with labels, of a neural network. [2]

Self-driving cars use computer vision to identify objects.

(c) Outline two differences in the way supervised and unsupervised learning are used to identify objects. [4]

Self-driving cars communicate to their passengers using natural language processing and voice recognition.

For example, a passenger could say, “Take me home via the scenic route,” and the natural language processing should correctly interpret this instruction.

(d) (i) Identify a noun in this instruction. [1]
(ii) Explain the semantics of this instruction. [4]

The passenger could also use natural language to check the grammar of messages sent from the self-driving car. The model would suggest and include missing words.

For example, “I late. Don’t wait me,” might be interpreted as, “I am late. Don’t wait for me.”

(e) Compare voice recognition with natural language grammar checking using neural network modelling. [6]

Most-appropriate topic code

• A4.3: Machine learning approaches [HL only] — part (a)
• A4.1: Machine learning fundamentals — parts (b) and (c)
• A4.3: Machine learning approaches [HL only] — parts (d) and (e)
▶️ Answer/Explanation

(a)
Award [3 max]

  • Create the initial population of antenna designs.
  • Evaluate the fitness of each design by testing each design against the goal.
  • Apply selection, crossover and mutation.
  • Iterate against a stopping condition or until the population approaches convergence.

For example, the goal could be to maximise signal range, minimise interference or optimise bandwidth.

(b)
Award [2 max]

The diagram should show at least:

  • Input layer with at least 2 input nodes.
  • At least 3 hidden nodes/layer.
  • Output layer with at least 2 output nodes.
  • Uni-directional arrows showing the flow of information.

A simple structure is:

Input layer          Hidden layer          Output layer

   ○ ───────────────→   ○ ───────────────→   ○
   ○ ───────────────→   ○ ───────────────→   ○
                     →   ○   →

(c)
Award [4 max]

Training:

  • Supervised learning trains using a labelled dataset.
  • Unsupervised learning trains without labelled outputs.

Approach:

  • Supervised learning learns to classify objects using the provided labels.
  • Unsupervised learning clusters similar objects together or detects anomalies in visual data.

Environment:

  • Supervised learning works well when the objects are known and defined in advance.
  • Unsupervised learning is advantageous when the types of objects or patterns to be identified are not predefined.

(d) (i)
Award [1 max]

Either route or home.

(d) (ii)
Award [4 max]

  • Semantics is used to identify the intent of the speaker.
  • The interpretation is based on contextual understanding.
  • For example, the system understands that “home” refers to the passenger’s place of residence.
  • The system uses cultural or situational awareness to understand the intended meaning.
  • The passenger wants to travel home while taking a slight detour through a picturesque or scenic route.

(e)
Award [6 max]

Award [2] for each valid comparison.

Architectural complexity: Voice recognition can use a combination of neural network types, such as CNNs for audio feature extraction and RNNs, LSTMs or transformers for speech sequencing, whereas grammar checking commonly uses advanced NLP models such as transformers.

Data processing: Voice recognition processes audio data and requires conversion from voice to text, whereas grammar checking primarily processes text data.

Preprocessing: Voice recognition may use noise reduction and normalization of audio signals, whereas text-based grammar checking can use tokenization, part-of-speech tagging and parsing for syntactic analysis.

Training data: Voice recognition requires diverse spoken-language datasets containing different accents and background noises, whereas grammar checking requires large text corpora containing grammatically correct and incorrect sentences.

Feature extraction: Voice recognition can convert sound waves into spectrograms or MFCCs, whereas grammar checking converts words into numerical representations or vectors.

Sequential processing: Both tasks rely heavily on sequential data processing and can detect anomalies in sequence patterns.

Real-time processing: Voice recognition generally requires real-time processing of incoming speech, whereas grammar checking is usually less time-sensitive.

Post-processing: Voice recognition can use language modelling to improve word choice and fluency, whereas grammar checking involves identifying errors and suggesting corrections.

Computational resources: Both can be resource intensive. Voice recognition requires significant computational resources for real-time processing of audio, whereas grammar checking focuses on computationally intensive linguistic analysis.

Types of errors: Errors in voice recognition can result from misinterpretation of sounds or speech nuances, whereas grammar-checking errors relate to the misuse of language rules.

More IBDP Computer Science HL Paper 2 Questions..........
Scroll to Top