CSIT940 / CSIT440 Research Methodology – Assignment 2 Literature Review

Unit and Assessment Overview

Subject codes: CSIT940 / CSIT440 – Research Methodology
Assessment: Assignment 2 – Systematic Literature Review
Weighting: 20% of final subject mark
Indicative length: Typically 3,000 to 4,000 words with no explicit word limit given. Students should write a full structured review that accommodates at least 20 research papers.
Due date: Mid to late April session in most offerings. Students must check the current subject outline for the exact deadline applicable to their session.

Purpose

The assignment develops core research skills including searching digital libraries, identifying and reading seminal and milestone papers, synthesising a focused body of work into a coherent literature review, and presenting findings with correct academic writing and referencing. Students demonstrate the ability to move from a broad research area to a clear perspective and to map out the main contributions, trends, gaps, and open problems within a specialised niche.

Topic Selection

Select one topic from the list below, determined by the last digit of your student number.

• 0 or 5: Smart contracts

• 1 or 6: Confidential computing

• 2 or 7: AI security

• 3 or 8: Differential privacy

• 4 or 9: Multi-party computation

Within the allocated topic, students must narrow their focus to a specific and defensible perspective. Examples include confidential computing for machine-learning workloads, membership-inference attacks in AI security, differential privacy in deep learning, multi-party computation for privacy-preserving federated learning, or smart contract vulnerabilities in decentralised finance.

Core Task

Write a structured literature review on the chosen topic and perspective that:

• Is grounded in at least 20 peer-reviewed papers.

• Includes seminal foundational work and key milestone developments such as original definitions, widely cited system designs, major attacks and defences, or influential surveys.

• Draws from both journals and conferences, recognising that top conferences are essential sources in computing research.

• Summarises and analyses each paper’s motivation, problem, methods, main results, and conclusions while offering brief critical comments.

• Organises the literature into clear themes rather than presenting only paper-by-paper summaries.

• Includes at least one comparison table that highlights key similarities and differences across a subset of papers, for example assumptions, techniques, datasets, performance, and limitations.

Technical and Formatting Requirements

• Number of references: At least 20 peer-reviewed papers. Blogs and non-archival sources should be avoided except for minor contextual information.

• Coverage: Must include seminal and milestone papers, not only very recent publications.

• Venue mix: Both journal articles and conference papers must be used.

• Reference style: References must be listed in plain style. Either numeric or author–year plain style is acceptable as long as it follows the subject template.

• Typesetting: Times New Roman, 11-point font, single column layout.

• Similarity: Overall similarity rate must be under 40 percent including references, or under 15 percent if references are excluded. Excess similarity due to copied text or templates will result in penalties.

Suggested Structure

Use the sample PDF and LaTeX or Word template provided on the subject site as the baseline. A typical structure is outlined below.

1. Title and abstract

   • Provide a concise and precise title reflecting the chosen perspective.

   • Include a 150 to 250 word abstract summarising focus, scope, main themes, and key observations.

2. Introduction

   • Introduce the broader topic and explain why it is important in computing or security.

   • Define the specific perspective of the review.

   • Outline what the review will cover, how prior work is organised, and the value added by the synthesis.

3. Background and definitions

   • Provide clear definitions of core concepts, threat models, and system components.

   • Briefly discuss any necessary mathematical or systems background with citations to classic texts or surveys.

4. Thematic literature review

   • Organise papers into three to five major themes such as trusted execution environments for machine learning, limitations and side-channel attacks, or privacy mechanisms in deep learning.

   • Discuss seminal and milestone papers first within each theme, followed by more recent work.

   • For each paper, describe motivation, problem, techniques, results, and limitations, and explain how it fits within the theme.

5. Comparison and synthesis

   • Include at least one comparison table with rows representing papers and columns representing factors such as threat model, technique, guarantees, overhead, datasets, evaluation metrics, or limitations.

   • Discuss the table in the text to highlight patterns, trade-offs, and research gaps.

6. Open problems and future directions

   • Identify unresolved challenges, shortcomings of current approaches, and promising directions for future research based on the literature and analysis.

7. Conclusion

   • Summarise the main themes and provide a high-level assessment of the maturity of the field.

   • Reiterate the key contributions of the review.

8. References

   • List all cited works in plain style consistent with the provided template.

Process Tips

• Use digital libraries such as IEEE Xplore, ACM Digital Library, SpringerLink, and ScienceDirect to locate high-quality research papers.

• Begin with influential surveys and follow their reference lists to identify seminal works.

• Ensure the review is analytical rather than purely descriptive by comparing, critiquing, and synthesising findings.

• Keep detailed notes on each paper’s key ideas and limitations while reading to avoid accidental plagiarism.

Marking Criteria

Content – 60%

• Appropriateness and focus of the chosen perspective within the assigned topic.

• Coverage and selection of literature with at least 20 papers and inclusion of seminal and milestone work from journals and conferences.

• Depth and accuracy of descriptions, analysis, and critical comments.

• Quality of thematic organisation and synthesis.

• Usefulness and clarity of comparison tables.

Format – 20%

• Compliance with formatting rules regarding font, size, column layout, and reference style.

• Correct and consistent typesetting following the provided sample or template.

• Professional structure and layout of sections, figures, and tables.

Grammar and Writing – 20%

• Clarity and coherence of writing.

• Correct grammar, punctuation, and spelling.

• Appropriate academic tone and precise technical language.

Additional Academic Guidance Paragraph

Producing a high-quality literature review requires more than simply summarising a collection of research papers. Effective reviewers must evaluate how individual contributions relate to one another and how they collectively shape the development of a research field. This involves identifying common assumptions, methodological differences, and variations in evaluation practices across studies. A well-constructed review also highlights where existing approaches fail to address important practical or theoretical challenges. By carefully analysing these aspects, students can demonstrate critical thinking and an ability to synthesise complex technical information into a coherent narrative. Such analytical synthesis is essential for guiding future research and for understanding the broader significance of technological advances in computing and security (Webster & Watson 2002).

A strong literature review on differential privacy in deep learning, for example, usually starts with a clear definition of epsilon-differential privacy and an intuitive explanation of how adding calibrated noise can limit an adversary’s ability to infer the presence of any individual record. The background section then separates privacy and security concepts and outlines typical attack models such as membership inference, model inversion, and reconstruction attacks before introducing the main differential privacy mechanisms used in deep learning. In the thematic review, high-quality work tends to group papers according to where the noise is applied, such as at the data level, during gradient updates, or on the final model outputs, and it clearly explains the trade-offs between privacy loss, accuracy, and computational cost. A well-constructed comparison table highlights which methods scale to large models, support distributed training, or offer formal composition guarantees. The synthesis section then draws out patterns, such as the tendency for tighter privacy guarantees to degrade model performance on complex tasks, and it notes recurring open problems like tuning privacy budgets or defending against adaptive adversaries. Finally, an effective conclusion pulls these insights together and points to specific gaps where new methods or theoretical tools are still needed.

Recent References

• Zhu, T, Gu, L, Xiong, P & Li, J 2024, ‘Differential privacy in deep learning: A literature survey’, Neurocomputing, vol. 588, 127740, doi:10.1016/j.neucom.2024.127740.

• Zhang, T, Chen, S, Zhu, H & Li, J 2023, ‘A survey on AI security: Threats and defenses’, ACM Computing Surveys, vol. 55, no. 12, pp. 1–37, doi:10.1145/3547159.

• Hanzlik, L, Tople, S, Koruyeh, E & Bindschaedler, V 2022, ‘Machine learning with confidential computing: A systematization of knowledge’, arXiv preprint, arXiv:2208.10134.

• Abadi, M et al. 2016, ‘Deep learning with differential privacy’, in Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security, ACM, New York, pp. 308–318, doi:10.1145/2976749.2978318.

• Shokri, R et al. 2017, ‘Membership inference attacks against machine learning models’, in Proceedings of the 2017 IEEE Symposium on Security and Privacy, IEEE, Los Alamitos, pp. 3–18, doi:10.1109/SP.2017.41.

• Webster, J & Watson, RT 2002, ‘Analyzing the past to prepare for the future: Writing a literature review’, MIS Quarterly, vol. 26, no. 2, pp. xiii–xxiii.