Project Type: Machine Learning
Overview
An ML-based project may focus either on developing, improving or investigating machine learning techniques, or on applying existing techniques to a particular problem domain. Project elements may include one or more of:
- Design of novel, general ML techniques as a theoretical contribution.
- Design of ML algorithms for a specific problem domain or data.
- Design of techniques in subfields of ML, such as domain adaptation, meta learning, federated learning, continual learning or reinforcement learning.
- Application and evaluation of an existing ML pipeline in a novel domain.
Suggested report/dissertation content
| Section | Content |
|---|---|
| Title page | See general guidance. |
| Abstract | Ditto |
| Declaration | Ditto |
| Contents page | Ditto |
| Introduction | Ditto |
| Background | Ditto |
| Context survey | Ditto |
| Methodology | Explain why the selected techniques were chosen, what adaptation or customisation was needed, and how this relates to the problem domain. |
| Experiments | Explain and justify evaluation methodology, metrics, protocols and baselines for comparison. Present results. |
| Analysis and discussion | Explain the results in relation to the problem domain and what we can learn from them. |
| Evaluation and critical appraisal | See general guidance. |
| Conclusions and further work | Ditto |
| References | Ditto |
| Appendix: ethics | Ditto |
| Appendix: optional further material | Ditto |
This structure can be varied with supervisor approval.
Mark descriptors
- 1-3: Little evidence of any attempt to complete the project.
- 4-6: Little evidence of any acceptable attempt to complete the project, with no substantial relevant material submitted.
- 7: Minimal application of ML to a problem with some attempt at evaluation.
- 8-10: Basic application of ML to a problem with little evaluation or demonstration of significant understanding.
- 11-13: Reasonable design of ML, evaluation and discussion of results.
- 14-16: Reasonable design of ML and good presentation of results, but justification of design decisions lacking clarity and discussion of results somewhat shallow.
- 17-18: Innovation in ML design, solid evaluation with in-depth discussion of results.
- 19-20: Innovative algorithm design with a deep understanding of problem, ML, and results. Code and results are reproducible. Potential, with some additional work, for a publication at a top-tier conference.