konst-int-i/remix: Rule Extraction Methods for Interactive eXplainability

Rule Extraction Methods for Interactive eXplainability

This repository implements the computational experiments for the MPhil in Advanced Computer Science project of the title "Global and local interpretability in ML-enabled clinical decision-making tools". Broadly speaking, we build in multiple extensions of the ECLAIRE rule extraction algorithm (lucid/extract_rules/eclaire_extensions.py) and implement a discrete column generation solver to build optimal Boolean Rulesets for a given objective function (Hamming Loss) (lucid/extract_rules/cg_extract.py). We also implement an explainability module, which implements multiple local and global…

saved by

• Denizhan Akar

related reading

• On the Biology of a Large Language Modeltransformer-circuits.pub
• Tracing the thoughts of a large language model \ Anthropicanthropic.com
• Goodfire AIgoodfire.ai
• Topicslearnmechinterp.com
• Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activationstransformer-circuits.pub
• Faithful, Interpretable Model Explanations via Causal Abstraction | SAIL Blogai.stanford.edu
• [2511.08579] Training Language Models to Explain Their Own Computationsarxiv.org
• Parsed | Custom, interpretable AI systems that continuously learnparsed.com
• Datacurve | The data engine for frontier AIdatacurve.ai
• CodaLab Worksheetsworksheets.codalab.org
• Training Language Models to Explain Their Own Computationsarxiv.org
• Neuronpedianeuronpedia.org