EXIN BCS Machine Learning Award (AIMLA) Exam Syllabus

• Indicative content
  a. Machine learning is a subset of artificial intelligence (AI)
  b. “Learning from experience”
  c. Tom Mitchell – definition (Academic) iterative, continuous learning (Machine Learning 1997, first publication, 2013)
  d. Requirement for talent for learning/mathematics (i.e. data scientist)
  e. Application of algorithms to given data to derive insight
• Guidance
  ​- It is important for candidates to understand that machine learning is a subset of AI. AI itself is not a new concept; machine learning is another step in the evolution of AI. Machine learning is used within data science and is the application of algorithms to derive insight from data and big data.

- explain different applications of machine learning.

• Indicative content
  a. Prediction
  b. Object recognition
  c. Classification
  d. Clustering
  e. Recommendations (e.g. Netflix, Spotify)
  f. Generative AI (e.g. ChatGPT, Copilot)
• ​Guidance
  - Machine learning can be used in a number of contexts to complete different types of tasks. Candidates should be encouraged to explore different examples and applications of machine learning.

- describe the role of a learning agent.

• Indicative content
  a. Data
  b. Single task
  c. Learning from experience
• ​Guidance
  - Learning agents are commonly used in machine learning. Each agent is designed to undertake a specific task using a given amount of data, which they undertake autonomously. Through the repetition of undertaking this task they learn to improve each time. Examples include chatbots, driverless cars, facial recognition.

- explain the concept of deep learning.

• Indicative content
  a. Universal technique to solve a larger set of problems
  b. Neural networks combined with large data sets
• Guidance
  - The application of deep learning (a subset of machine learning) involves the training of large neural networks to process and analyze vast amounts of data to derive greater insight and to solve more complex problems.

- describe the purpose of a neural network.

• Indicative content
  a. Input > identify patterns in data > output
  b. Decision making
• Guidance
  - Neural networks are commonly used in machine learning, particularly in the analysis of unstructured or unlabeled data (e.g. images, handwritten documents), whereby the input data is analyzed to determine any recognizable or similar patterns against other learned bits of data in order to determine the output. Candidates may wish to explore the concept of a neural network by considering technologies that use machine learning such as voice recognition software where the input (captured user’s voice) is analyzed and compared against stored patterns (data) to identify the output (a specific action, acceptance of voice command, text-to-speech).

- illustrate how machine learning compliments knowledge-based systems.

• Indicative content
  a. Knowledge-based systems
  b. Complimentary AI technologies
• Guidance
  - A knowledge-based system is a form of AI designed to capture human expertise/knowledge (within a knowledge base) and apply a set of rules to identify an outcome (through an inference engine). Machine learning is data based and can derive outcomes through the use of algorithms e.g. a neural network. Technologies such as driverless cars may use a combination of different AI applications to perform different tasks. It may include a knowledge based system to make informed decisions or identity the probable cause of a fault, and it may use a neural network for image recognition for navigation using the car’s camera.

- explain the process through which machine learning works with data.

• Indicative content
  a. The machine learning process
  b. Analyze the problem
  c. Data selection
  d. Data pre-processing
  - Cleaning
  - Integration
  - Transformation
  - Reduction
  - Wrangling
  e. Data visualization
  f. Select a machine learning model (algorithm)
  - Train the model
  - Test the model
  - Repeat (learning from experience to improve results)
  g. Review
  - Peer review
  - Learning from multiple algorithms
  - Identify best machine learning model
• Guidance 
  - The machine learning process allows us to define the solution based on the problem that has been identified through the process of data selection, pre processing, visualization and testing of data with specific algorithms. Once we are happy that both the data and the algorithms we use, are performing well we can deploy our model. The machine learning process is explored in detail by Google director Aurélien Géron; recognize the problem, define data, check algorithms, improve results, present results.
  - There is no de facto method within machine learning, learning through experience is vitally important. Testing involves creating the correct test data,creating bins to learn from and bins for what you wish to test.

• Indicative content
  a. Object-oriented programming languages
  - Python
  - R
  - C++
  - Java
  b. Libraries/templates
• ​Guidance
  - Candidates should be familiar with common programming languages and their use, although it is not expected that they are fluent in using them. Python is a very popular language used in machine learning and data science. Libraries are used to bundle functions into templates that include the use of different programming languages e.g. Python.

- identify common open source and proprietary software used in coding for machine learning.

• Indicative content
  a. TensorFlow
  b. R Studio
  c. Cuda
  d. Scikit-Learn
  e. MATLAB
• Guidance
  - Candidates should be encouraged to explore some of the known software and programming environments used in programming machine learning. It is not expected that they are proficient in their use however they should be familiar with at least one software.

• Indicative content
  a. Probability (Bayes’ theorem)
  b. Statistics
  - Descriptive statistics
  - Inferential statistics
  c. Linear algebra
• Guidance
  - It is important for candidates to have a basic understanding of the mathematics used within machine learning, regardless of whether the software they go on to use handles this automatically. Bayes’ theorem is a method which can be used to calculate probability where other probabilities are known. Understanding the basic principles of linear algebra will provide them with the foundation on which to better understand machine learning and in implementing algorithms.

- list and describe typical algorithms used in machine learning.

• Indicative content
  a. Regression algorithms, e.g.:
  - Linear regression
  - Polynomial regression
  b. Classification algorithms, e.g.:
  - K-nearest neighbors
  - Decision trees
  - Logistic regression
  c. Clustering algorithms, e.g.:
  - K-means
  - Hierarchical
• Guidance
  - Candidates should have a basic understanding of some of the common algorithms used in machine learning and where they may be used in supervised or unsupervised learning. It is not essential at this level for them to understand the specific formulas used within each algorithm, however it is certainly advantageous to have a basic understanding of the mathematics involved in order to make it easier to program machine learning. You may wish to further challenge candidates by looking into the use of boosting, decision forests, and ensembles.

​- describe supervised, unsupervised and semi-supervised learning.

• Indicative content
  a. Supervised learning
  b. Unsupervised learning
  c. Semi-supervised learning
• Guidance
  - It is useful for candidates to have a basic understanding of the different types of approaches to machine learning to understand how it can be used to work with different types of data and where different algorithms are best used. Supervised learning involves the application of an algorithm to labeled data to solve a problem, for example classification, where we know what the output will be.
  - Unsupervised learning involves the application of an algorithm to unlabeled data to solve a problem, for example clustering (grouping data based on similarities).
  - Semi-supervised learning involves the application of an algorithm where during the training of the algorithm we begin with a small amount of labeled data and then introduce a larger amount of unlabeled data.
  - Candidates may be encouraged to also consider reinforcement learning which is commonly used in gaming. 

The candidate can…
- describe a particular problem that can be addressed through the use of machine learning.

• Indicative content
  a. Problem identification
  b. Requirements for data collection
  c. Proposing the machine learning solution
• Guidance
  - Candidates should be encouraged to identify a specific problem which could be solved through implementing machine learning.

- outline typical tasks required in the preparation of data for developing a particular application of machine learning.

• Indicative content
  a. Data pre-processing
  b. Data transformation
  c. Importing/loading data
• Guidance
  - Candidates should be able to outline the tasks they would need to undertake to prepare the data for use within an application of machine learning. This may include steps such as cleaning the data, data validation, and data transformation to ensure it is in a suitable format for using within a chosen software.

- explain the process of training a machine learning model.

• Indicative content
  a. Requirements for training
  b. Setting up training bins for data
  c. Selecting an algorithm
  d. Rules
  e. Supervised, unsupervised, semi-supervised
• Guidance
  - Candidates should be able explain the process of training a particular algorithm using their prepared data

- explain the process of testing a machine learning model.

• Indicative content
  a. Testing
  b. Tuning
  c. Ensembles
  d. Statistical testing
  e. Review
• Guidance
  - Candidates should be able to explain the process through which they tested a particular algorithm using their prepared data and how they identified whether it was performing well. They may use a number of methods to test their algorithm, and they may wish to test and compare multiple algorithms.

- discuss how to evaluate the results of testing in order to identify the information to be shared with key stakeholders.

• Indicative content
  a. Evaluating findings
  b. Identifying relevant information for your stakeholders/context
  - What have we learned?
  - Have we been able to address the problem?
  - What next?
  - Learning from experience
  c. Drawing conclusions
  d. Communication techniques/methods
• Guidance
  - Candidates should be able to explain how they would go about identifying the key pieces of information to share with their stakeholders. They should also explain key considerations for sharing information with stakeholders e.g. type of information, presentation, language and use of technical terms, being prepared to answer questions.