02 / Research topics
Potential PhD Research Topics in Computing
Explore suggested topics and the methods, tools and expertise that could support each investigation.
1
Trustworthy Generative Artificial Intelligence for Reliable and Explainable Decision Support
Generative models can produce persuasive but incorrect, biased, or unverifiable outputs. The gap lies in integrating factuality, uncertainty estimation, explanation, and accountability without severely reducing usefulness or efficiency. This project could develop a theoretically grounded framework combining retrieval, calibrated generation, provenance tracking, and human oversight. Objectives include designing new methods, establishing evaluation benchmarks, and testing performance in high-stakes decision settings against contemporary language models.
Applications and use cases
2
Quantum Machine Learning Algorithms for Practical Hybrid Classical Quantum Computing
Quantum machine learning remains constrained by noisy hardware, limited qubit counts, costly data encoding, and uncertain advantages over classical methods. A valuable PhD could close the gap between theoretical algorithms and realistic hybrid deployment by developing noise-aware models, resource-efficient encodings, and rigorous comparison protocols. Objectives should include proving complexity or convergence properties, implementing algorithms on simulators and quantum processors, and identifying application regimes where measurable benefits are credible.
Applications and use cases
3
Privacy Preserving Federated Learning for Resource Constrained Edge Intelligence
Federated learning avoids centralizing data, yet model updates can leak information, communication can be expensive, and edge devices vary widely in capability. This research should develop adaptive privacy and aggregation mechanisms that remain accurate under heterogeneous, unreliable, and adversarial conditions. Novelty may come from jointly optimizing differential privacy, secure aggregation, personalization, and energy use. Evaluation should measure privacy leakage, accuracy, robustness, latency, communication, and battery consumption across realistic deployments.
Applications and use cases
4
Post Quantum Cybersecurity Architectures for Long Lived Critical Infrastructure
Quantum algorithms threaten widely deployed public key cryptography, while critical infrastructure often contains devices and protocols that cannot be rapidly replaced. The research gap concerns migration strategies that combine post quantum algorithms with legacy systems under strict performance, safety, and interoperability constraints. A PhD could create discovery and inventory methods, hybrid cryptographic architectures, automated protocol transformation, and lifecycle governance. Objectives should include formal security analysis and evaluations in representative operational technology environments.
Applications and use cases
5
Embodied Multimodal Artificial Intelligence for Safe Adaptive Human Robot Collaboration
Robots still struggle to interpret ambiguous human instructions, changing environments, social cues, and safety constraints together. The missing contribution is an integrated approach linking vision, language, action planning, uncertainty, and continual adaptation in physical settings. This project could create multimodal world models with verifiable safety policies and efficient learning from demonstrations. Objectives include improving task generalization, reducing interaction data requirements, and validating behavior through laboratory and real-world collaborative experiments.
Applications and use cases
6
Energy Aware Computing Architectures for Sustainable Artificial Intelligence Infrastructure
AI growth increases electricity consumption, carbon emissions, water use, and hardware waste, yet existing performance metrics rarely capture complete environmental costs. The research opportunity is to co-design algorithms, compilers, hardware scheduling, and data center operations around lifecycle sustainability rather than speed alone. A strong proposal could develop energy and carbon models, adaptive workload placement, and training efficiency techniques. Objectives should quantify tradeoffs across accuracy, latency, cost, emissions, and resource longevity.
Applications and use cases
7
Neuro Symbolic Artificial Intelligence for Interpretable Scientific Knowledge Discovery
Purely data-driven models often discover correlations without producing scientific explanations, while symbolic systems struggle with noisy, high-dimensional observations. This proposal could bridge both paradigms by learning representations from data and imposing mathematical, causal, or logical structure during discovery. Novel contributions may include differentiable hypothesis generation, automated theory refinement, and uncertainty-aware explanations. Objectives include recovering known laws, discovering useful candidates, and validating findings with domain scientists and independent experiments.
Applications and use cases
8
Digital Twins for Resilient Climate Adaptive Urban Infrastructure Management
Urban infrastructure data is fragmented across sensors, agencies, models, and communities, limiting timely responses to floods, heat, outages, and transport disruptions. The gap is a trustworthy digital twin that combines real-time observations, simulation, uncertainty, and stakeholder decisions rather than offering visualization alone. Research could develop interoperable twin architectures, predictive models, and intervention optimization. Objectives should assess resilience improvements, fairness, data quality, computational scalability, and decision value in realistic city scenarios.
Applications and use cases
9
Autonomous Software Engineering Agents for Verified Secure and Maintainable Systems
AI coding agents can accelerate development but may introduce vulnerabilities, hidden dependencies, poor tests, and difficult-to-maintain designs. Current tools lack dependable guarantees across requirements, implementation, verification, and long-term maintenance. This research could create agent teams that use repository context, formal checks, security analysis, and human approval strategically. Objectives include measuring productivity and defect reduction, developing new agent coordination methods, and establishing reproducible evaluation protocols across diverse software projects.
Applications and use cases
10
Privacy Preserving Artificial Intelligence for Fair Clinical Decision Support Systems
Clinical AI can improve diagnosis and resource allocation, but patient data are sensitive, institutions differ in practice, and models may disadvantage underrepresented groups. The research gap involves achieving privacy, fairness, calibration, interpretability, and clinical usefulness simultaneously under distributed data constraints. A PhD could develop federated and causal methods with subgroup-aware uncertainty and clinician interaction. Objectives should include prospective workflow evaluation, external validation, privacy auditing, and evidence that the system improves decisions without increasing disparities.
Applications and use cases