Dr Koorosh Aslansefat¶
Assistant Professor in Computer Science, University of Hull
I work on AI safety, trustworthy machine learning, explainable AI, and dependability for safety-critical and autonomous systems. My research develops practical methods, tools, and assurance workflows for using AI responsibly in settings where failure matters.
"Trustworthy AI is not only about better models. It is about evidence, monitoring, explanation, and responsible deployment."
Looking for a PhD or MSc by Research?¶
I welcome enquiries from motivated PhD and MSc by Research candidates interested in AI safety, responsible LLMs, multimodal alignment, explainable AI, safety-critical systems, autonomous systems, digital twins, and dependable machine learning.
Good fits are students who enjoy both theory and implementation: building methods, evaluating them carefully, and turning research into tools that other engineers and researchers can use.
Research Interests¶
-
AI safety and alignment
Safety monitoring for large language models, multimodal AI systems, responsible AI frameworks, and collaborative alignment workflows.
-
Machine learning dependability
Runtime monitoring, uncertainty quantification, robustness evaluation, and statistical safety measures for data-driven systems.
-
Explainable AI
Model-agnostic interpretation, robust local explanations, SMILE/XWhy, and evidence that humans can use during assurance.
-
Safety-critical and autonomous systems
Dependability assessment, runtime assurance, drones, multi-robot systems, fault diagnosis, and model-based safety engineering.
Latest Papers¶
- 10 Jul 2026 - New preprint: ConceptSMILE: Auditing the Trustworthiness of Concept-Based Explainable AI in arXiv (Cornell University).
- 03 Jul 2026 - New conference paper: When Words Move Markets: Interpretable Behavioural and Robustness Analysis of LLMs for Financial Sentiment Reasoning via Local Perturbation Explanations in Lecture notes in computer science.
- 01 Jul 2026 - New preprint: Bayesian Uncertainty Propagation for Agentic RAG Pipelines: A Proof-of-Concept Study on Multi-Hop Question Answering in arXiv (Cornell University).
- 28 Apr 2026 - New preprint: Risk Assessments for Evasive Emergency Maneuvers in Autonomous Vehicles in arXiv (Cornell University).
- 10 Apr 2026 - New data paper: A Multi-Modal Dataset for Ground Reaction Force Estimation Using Consumer Wearable Sensors in Scientific Data.
Latest News¶
2025 - Academic Lead for an Innovate UK project on trustworthy AI agents for incident management with Veracity Healthcare.
2024-2025 - TrustLLM project launched to investigate responsible and trustworthy LLM use for planning law.
2024 - SafeLLM work on domain-specific safety monitoring for large language models appeared in IOP Journal of Physics.
2024 - Ongoing collaborations include QinetiQ, Google, Microsoft, Walton & Co Ltd, Connexin, and IIT Madras.
2023 - SafeML was recommended in German Industry Standard DIN SPEC 92005 for ML uncertainty quantification.
Selected Highlights¶
| Area | Highlights |
|---|---|
| Research output | 65+ publications across AI safety, dependability, explainability, and safety-critical systems. |
| Impact | SafeML recommended in German Industry Standard DIN SPEC 92005. |
| Funding | Approximately £1.237M funded portfolio, including £285K as PI, Academic Lead, or Co-PI and £952K through wider Co-I collaborations. |
| Supervision | Supervision across responsible LLMs, medical AI safety, fairness, wind-energy AI, generative AI explainability, and EdgeAI. |
| Open source | SafeML, SafeDrones, XWhy/SMILE, and related dependability tooling. |
Contact¶
For research collaboration, PhD or MSc by Research enquiries, invited talks, and academic service, use the links below.