How to Write a Computer Science PhD Proposal (Expose) for German Universities: Problem, Novelty and Feasibility

1. Defining the Research Problem

The research problem should emerge from an identifiable limitation in existing knowledge or technology. Applicants can begin by examining recent journal articles, conference papers, systematic reviews, and relevant doctoral research. They should compare the stated objectives, methods, datasets, and limitations of relevant studies rather than relying on titles or abstracts alone.

A useful research problem answers three questions: What limitation exists? Why does it matter? What remains insufficiently investigated?

A practical way to narrow the problem is to record the limitation reported by each relevant study, the conditions under which it was observed, and the method used to evaluate it. This comparison helps distinguish a recurring research gap from an isolated implementation issue.

For example, existing machine-learning algorithms may perform effectively on test datasets but show reduced accuracy when they are applied to changing real-world datasets. A focused investigation could examine which types of distribution shift affect performance, how selected adaptive learning techniques respond, and which evaluation measures can demonstrate whether performance has improved.

The research problem in a Computer Science PhD proposal should be clearly defined so that the research questions and objectives follow directly from the limitation being investigated.

2. Establishing Research Novelty

Novelty refers to the contribution that distinguishes the proposed research from existing work. It does not necessarily require inventing an entirely new technology.

For example, a proposal may retain an established model architecture but contribute a new evaluation framework for testing its performance under a defined data condition. The novelty claim should describe that specific contribution rather than suggesting that the entire technology is new.

A contribution may involve a new algorithm, an improved architecture, a different evaluation framework, an original dataset, or a new understanding of an existing computational problem. The proposal should state what will be different and why that difference matters.

Applicants should compare their proposed approach with relevant studies and explain the remaining limitation. A useful comparison identifies what existing approaches do, where their limitations remain, and how the proposed work will address a defined part of that limitation. Claims such as “the first study ever conducted” should be avoided unless supported by a comprehensive and defensible literature review.

PhD research novelty becomes convincing when the proposed contribution follows directly from the identified research gap and can be evaluated against relevant existing approaches.

3. Demonstrating Research Feasibility

A research idea may be interesting but unsuitable for a doctoral project if its implementation is unrealistic.

A feasibility assessment can make these decisions explicit by mapping each research objective to its required data, tools, computational resources, evaluation procedure, and estimated time. Where a resource is uncertain, the proposal should identify a realistic alternative.

Feasibility requires applicants to consider available datasets, computational infrastructure, software tools, research skills, evaluation methods, and project duration. They should also identify access restrictions, licensing conditions, hardware constraints, and possible alternatives if a planned resource is unavailable.

For example, a proposal involving deep learning should explain whether suitable data are available, how model performance will be evaluated, which baseline approaches will be used for comparison, and what computational resources are required.

A realistic Computer Science PhD research methodology should connect each objective with an appropriate research activity, an evaluation method, and a measurable outcome.