Reproduce “Formally Explaining Decision Tree Models with Answer Set Programming” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Akihiro Takemura, Masayuki Otani, Katsumi Inoue
Researchers from the National Institute of Informatics and Tokyo Institute of Technology developed an Answer Set Programming (ASP) framework for generating various types of formal, logically guaranteed explanations for decision tree models, including random forests and boosted trees. The ASP method demonstrated superior capability in enumerating a larger quantity of contrastive and majority explanations compared to SAT-based methods when successful, achieving 100% success for decision tree explanations and competitive rates for single explanations in random forests.
Key insights
- ASP's declarative nature and combinatorial reasoning capabilities offer a flexible and extensible framework for formally explaining diverse model behaviors, overcoming some limitations of existing SAT-based approaches.
- By encoding model structures and explanation types, ASP maintains logical guarantees for explanations while supporting complex reasoning tasks, such as comprehensive explanation enumeration within a single solver call.
- The approach effectively captures the extremal behavior of individual trees in boosted ensembles and can identify minimal sets of conditions under best-case or worst-case scenarios for predictions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.