Questions and answers that come up while studying theory, papers, mathematics, and implementations.
Sequential recommender systems use the order of user interactions to predict the next item, but it is difficult to explain why a particular item was recommended.
TRACE studies ranking-aware counterfactual explanation. The search is divided into influence-guided position selection, plausibility-aware candidate retrieval, and actual-margin beam pruning.
Two algorithms that reduce repeated computation while preserving exact TreeSHAP values.
Why exact SHAP computation becomes tractable with Tensor Train structure.
An efficient method for computing Shapley interactions in tree ensembles.
The difference between finding an answer quickly and verifying a candidate answer quickly.
Existence, uniqueness, and stability in linear systems and inverse problems.
Shapley Interaction Index (in progress)
Tensor-Train Decomposition (in progress)
First research note