Undergraduate Course: Neurosymbolic AI (INFR11303)
Course Outline
| School | School of Informatics |
College | College of Science and Engineering |
| Credit level (Normal year taken) | SCQF Level 11 (Year 4 Undergraduate) |
Availability | Not available to visiting students |
| SCQF Credits | 20 |
ECTS Credits | 10 |
| Summary | This course is an introduction to neurosymbolic AI, which combines neural networks with symbolic representation (e.g. using rules and constraints) and reasoning (e.g. deduction). While neural networks bring the capability to learn and generalise from data and from experimentation, symbolic techniques robustly incorporate knowledge and support safe and transparent reasoning. Neurosymbolic AI combines the virtues of both approaches. Neural models, for instance, give symbolic systems the ability to learn or adapt symbolic information from data (e.g. extracting rules or guiding search). Symbolic constraints, for their part, can steer the training of neural components and adapt their inference-time behaviour to ensure robustness. For example, this is vital in the application of neural systems to safety-critical autonomous systems such as self-driving cars. |
| Course description |
The course presents examples of popular neurosymbolic approaches, frameworks and applications. Topics will vary from year to year. Potential topics include:
- Neurosymbolic AI: motivation, tasks, and system-level failure modes
- Neural learning with logical constraints
- Neural components inside search
- Knowledge graph embedding and neural induction of rules
- Neurosymbolic architectures for code generation and verification
- Neural network verification for safety and robustness
- Neurosymbolic approaches for mathematics e.g. auto formalization
- Symbolic approaches for ensuring robust LLM use of tools
- Evaluating and comparing neural and neurosymbolic systems, addressing safety, trustworthiness, accuracy and explainability.
Depending on the neurosymbolic topics chosen, the course also covers essential foundations in symbolic AI such as:
- Propositional and first-order logic
- Probabilistic logics and reasoning with uncertainty
- Deterministic inference: forward/backward chaining and rule-based inference
- Constraint solving techniques e.g. CSP, SAT and SMT
The course is delivered through a combination of lectures and unassessed practical lab sessions, providing hands-on experience with key example tools. Opportunities for interaction with course staff include labs, an online discussion forum and office hours.
|
Entry Requirements (not applicable to Visiting Students)
| Pre-requisites |
It is RECOMMENDED that students have passed
Machine Learning (INFR10086) OR
Applied Machine Learning (INFR11211)
|
Co-requisites | |
| Prohibited Combinations | |
Other requirements | MSc students must register for this course, while Undergraduate and Visiting Undergraduate students must register for INFR11306 instead.
This course is open to all Informatics students including those on joint degrees.
Students are expected to have background in maths, programming and machine learning. Specifically:
1. Maths: linear algebra, calculus and probability at undergraduate level.
2. Programming: familiarity with a modern programming language such as Python.
3. Machine Learning: Neural network-related concepts e.g. loss functions, hyperparameter tuning, regularisation, gradient descent, etc. More generally, we expect that students have taken a basic introduction to machine learning involving neural networks that covers data processing and exploration, and how to apply and compare baseline methods.
4. Evaluation of machine learning models via e.g. accuracy, precision, recall and ROC AUC.
External students whose DPT does not list this course should seek permission from the course organiser. |
Course Delivery Information
|
| Academic year 2026/27, Not available to visiting students (SS1)
|
Quota: None |
| Course Start |
Semester 2 |
Timetable |
Timetable |
| Learning and Teaching activities (Further Info) |
Total Hours:
200
(
Lecture Hours 30,
Supervised Practical/Workshop/Studio Hours 12,
Summative Assessment Hours 2,
Revision Session Hours 2,
Programme Level Learning and Teaching Hours 4,
Directed Learning and Independent Learning Hours
150 )
|
| Assessment (Further Info) |
Written Exam
100 %,
Coursework
0 %,
Practical Exam
0 %
|
| Feedback |
Sample solutions will be provided to hands-on non-assessed practical exercises. Through the semester, support and feedback will be provided for these assessments and for the material covered in lectures. |
| Exam Information |
| Exam Diet |
Paper Name |
Minutes |
|
| Main Exam Diet S2 (April/May) | Neurosymbolic AI (INFR11303 and INFR11306) | 120 | |
Learning Outcomes
On completion of this course, the student will be able to:
- judge the effectiveness of core symbolic reasoning approaches in neural network contexts
- describe key neurosymbolic architectures for integrating symbolic and neural approaches
- evaluate and compare neural and neurosymbolic architectures, considering features such as failure modes, safety, trustworthiness, accuracy and explainability
|
Reading List
Russell and Norvig, AI: A Modern Approach, 4ed (Global Edition). Pearson. 2022.
Shakarian et al. Neuro Symbolic Reasoning and Learning, Springer, 2023. https://doi.org/10.1007/978-3-031-39179-8
D'Avila Garcez et al. Neural-Symbolic Cognitive Reasoning, Springer, 2009. https://doi.org/10.1007/978-3-540-73246-4
Selected papers will also be part of the reading list. |
Additional Information
| Graduate Attributes and Skills |
Not entered |
| Keywords | Neurosymbolic AI,Hybrid AI,Neural networks,Machine learning,Automated reasoning |
Contacts
| Course organiser | Dr Jacques Fleuriot
Tel: (0131 6)50 9342
Email: Jacques.Fleuriot@ed.ac.uk |
Course secretary | Mr Lachlan Boyd
Tel:
Email: lboyd@ed.ac.uk |
|
|
University Homepage