Logic-Based Learning

Module aims

In this module you will have the opportunity to:
- appraise, compare and contrast foundational principles of different logic-based machine learning frameworks
- describe and discriminate foundational semantics and methods for logic-based machine learning 
- distinguish and analyse keys steps of  state-of-the-art learning algorithms and heuristics
- specify state-of-the-art machine learning methods for Answer Set Programming (ASP)
- specify logic-based learning tasks capable of solving real-world problems  
- generalise logic-based learning to probabilistic inference and learning 

Learning outcomes

Upon successful completion of this module you will be able to:
- compare and contrast the semantics of different logic-based learning frameworks
- appraise the core principles of a logic-based learning task
- analyse and deploy algorithms to solve learning tasks in different semantic contexts
- devise answer set programs to solve real-word problems
- generalise logic-based learning algorithms to the learning of Answer Set programs
- evaluate current forms of probabilistic logic-based inference and learning
- outline and explain state of the art algorithms for probabilistic learning

Module syllabus

  • Deductive, abductive and inductive reasoning
  • Bayesian reasoning techniques 
  • Answer Set Programming 
  • Top-down and bottom-up approaches to learning
  • Meta-level logic-based learning
  • Monotonic and non-monotonic learning
  • Enhanced logic programming systems (Progol, TopLog, Metagol, TAL, Imparo, ASPAL and ILASP)
  • Soundness and completeness of learning algorithms
  • Probabilistic inductive learning
  • Probabilistic logic programming    

Teaching methods

The material will be taught through traditional lectures, backed up by unassessed, formative exercises designed to reinforce the material as it is taught. The lectures will introduce the theoretical foundations and algorithms for the various learning approaches and provide examples on how a learning problem can be specified and solved in each of the learning tasks.  You will also be able to access an ILP Framework platform where various ILP systems will be made available. Some of the tutorial hours will be run as hands-on sessions where you will solve the exercises in class using the ILP Framework. 

An online service will be used as an open discussion forum for the module. 

Assessments

There will be two courseworks that collectively contribute 20% of the mark for the module—10% each. The first coursework will focus on formulating, illustrating and assess different learning algorithms and their respective semantics. The second coursework will focus on specifying an agent-based learning problem using different logic-based learning frameworks, implement and solve the problem using appropriate learning algorithms,  and evaluate, compare and contrast their performance in terms of consistency, completeness and expressivity of their learned solutions. There will be a final written exam, which counts for the remaining 80% of the marks. Formative assessments include unassessed exercises.                
Feedback on the formative exercises will be given in class. The assessed courseworks will be accompanied by individual written feedback. Class-wide feedback on the assessed courseworks will also be given.

Module leaders

Professor Alessandra Russo