diff --git a/docs/research/blogs.md b/docs/research/blogs.md new file mode 100644 index 00000000..ecad8a0c --- /dev/null +++ b/docs/research/blogs.md @@ -0,0 +1,26 @@ +--- +title: Research Blogs and Explainers +description: Published SMILE and responsible-AI blog posts, plus XWhy guides about local explanations, LLMs, and evaluating attribution. +--- + +# Research blogs + +Explore published articles from the University of Hull research group and longer XWhy explainers. For research citations, use the [publication catalogue](publications.md). + +## Published research group blog posts + +- **[Mapping the Mind of an Instruction-Based Image Editing Using SMILE](https://www.responsibleaihull.com/post/mapping-the-mind-of-an-instruction-based-image-editing-using-smile)** — Koorosh Aslansefat explains how changes to words in an editing instruction can be connected to changes in the generated image. +- **[Pint of Science: Can we trust AI for skin cancer? Are they fair?](https://www.responsibleaihull.com/post/pint-of-science-trustworthy-ai-can-we-trust-ai-for-skin-cancer-are-they-fair)** — A public discussion of fairness and reliability in medical image classification, featuring Koorosh Aslansefat and Kuniko Paxton. +- **[Research Visits of Responsible AI Team to Chennai, India](https://www.responsibleaihull.com/post/research-visits-of-responsible-ai-team-to-chennai-india)** — A report on research exchanges about trustworthy AI in healthcare. +- **[Responsible Digital Twin Research Theme in Chennai, India](https://www.responsibleaihull.com/post/responsible-digital-twin-research-theme-in-chennai-india)** — A related research visit on digital twins and dependable intelligent systems. + +[Browse all posts from the Centre for Responsible AI at Hull](https://www.responsibleaihull.com/research-blog). + +## XWhy explainers and practical reading + +- **[How SMILE works](../concepts/smile.md)** — The perturbation, weighting, and local surrogate workflow behind an explanation. +- **[What a local explanation tells you](../concepts/local-explanations.md)** — How to interpret results for a particular input and model output. +- **[Explaining LLM responses](../explainers/llm/index.md)** — Prompt perturbations and token contributions in language-model workflows. +- **[Image generation and editing examples](../explainers/image-generation/image-generation-editing-examples.md)** — Examples and scope for visual generation workflows. +- **[Evaluating explanation fidelity](../evaluation/attribution-fidelity.md)** — Ways to test whether an attribution reflects the model's behaviour. +- **[Limitations and responsible interpretation](../concepts/limitations.md)** — Where local explanations need additional checks. diff --git a/docs/research/index.md b/docs/research/index.md index d039383f..0408f86a 100644 --- a/docs/research/index.md +++ b/docs/research/index.md @@ -17,6 +17,7 @@ SMILE—**Statistical Model-Agnostic Interpretability with Local Explanations** | Language and generative AI | gSMILE | Large language models | arXiv preprint | | Retrieval-augmented generation | KG-SMILE | Knowledge-graph and GraphRAG systems | arXiv preprint | | Concept-based explainability | ConceptSMILE | Concept-based explainable-AI methods | arXiv preprint | +| Concept-based explainability | ConceptSMILE thesis | Retinal-image AI and trustworthy visual concepts | MSc thesis | | Financial language analysis | Local perturbation explanations derived from gSMILE | LLM financial-sentiment reasoning | Peer-reviewed conference chapter | | Consolidated academic study | Generative-AI SMILE thesis | LLMs and instruction-based image editing | MSc thesis | @@ -45,6 +46,7 @@ ConceptSMILE audits whether human-understandable concepts provide explanations t ## Research resources - [Browse the complete publication catalogue](publications.md) +- [Read research blogs and explainers](blogs.md) - [Choose the correct citation for your use case](citation.md) - [Follow the reproducibility guidance](../how-to/reproducibility.md) diff --git a/docs/research/publications.md b/docs/research/publications.md index 8467333d..2fca9c6d 100644 --- a/docs/research/publications.md +++ b/docs/research/publications.md @@ -17,6 +17,7 @@ This catalogue connects the SMILE research family to the systems each publicatio | 2025 | gSMILE | Large language models | arXiv preprint | [arXiv:2505.21657](https://arxiv.org/abs/2505.21657) | | 2025 | KG-SMILE | Knowledge-graph and GraphRAG systems | arXiv preprint | [arXiv:2509.03626](https://arxiv.org/abs/2509.03626) | | 2026 | ConceptSMILE | Concept-based explainable AI | arXiv preprint | [arXiv:2607.09649](https://arxiv.org/abs/2607.09649) | +| 2026 | ConceptSMILE thesis | Retinal-image explainability and trustworthy visual concepts | MSc thesis | [University of Hull repository](https://hull-repository.worktribe.com/output/5763337) | | 2026/2027 | Financial-sentiment local perturbation study | LLM financial-sentiment reasoning | NLDB 2026 conference chapter; 2027 volume | [DOI](https://doi.org/10.1007/978-3-032-29532-3_19) | | 2025 | Generative-AI SMILE thesis | LLM and image-editing explainability | MSc thesis | [arXiv:2602.01206](https://arxiv.org/abs/2602.01206) | @@ -226,6 +227,28 @@ ConceptSMILE audits whether higher-level, human-understandable concepts provide ## Related thesis +### ConceptSMILE: Towards A Framework for Trustworthy Concept-Based Explainable Artificial Intelligence using SMILE + +**Mohadeseh Mollapour.** “ConceptSMILE: Towards A Framework for Trustworthy Concept-Based Explainable Artificial Intelligence using SMILE.” MSc thesis, University of Hull, 2026. Supervised by Koorosh Aslansefat and Bhupesh Mishra. + +This thesis studies concept-based explanations for retinal-image AI and evaluates their trustworthiness. It is a separate academic record from the [ConceptSMILE arXiv preprint](https://arxiv.org/abs/2607.09649) listed above. + +[Thesis record](https://hull-repository.worktribe.com/output/5763337) + +??? note "BibTeX" + + ```bibtex + @mastersthesis{mollapour2026conceptsmilethesis, + author = {Mohadeseh Mollapour}, + title = {{ConceptSMILE}: Towards A Framework for Trustworthy + Concept-Based Explainable Artificial Intelligence using {SMILE}}, + school = {University of Hull}, + type = {{MSc} thesis}, + year = {2026}, + url = {https://hull-repository.worktribe.com/output/5763337} + } + ``` + ### Addressing Explainability of Generative AI Using SMILE **Zeinab Dehghani.** “Addressing Explainability of Generative AI Using SMILE: Statistical Model-Agnostic Interpretability with Local Explanations.” MSc thesis, University of Hull, September 2025. Also available as arXiv:2602.01206. diff --git a/properdocs.yml b/properdocs.yml index 2710b75d..a22bd3ab 100644 --- a/properdocs.yml +++ b/properdocs.yml @@ -92,6 +92,7 @@ plugins: - evaluation/attribution-faithfulness.md: ATT faithfulness under interventions Research: - research/index.md: Research and reproducibility entry point + - research/blogs.md: Research blogs and XWhy explainers - research/citation.md: How to cite XWhy Releases: - release-notes.md: Version history and detailed release notes @@ -159,6 +160,7 @@ nav: - Research: - Overview: research/index.md - Publications: research/publications.md + - Blogs: research/blogs.md - Citation: research/citation.md - API Reference: reference/index.md - Release Notes: release-notes.md