LandscapeSHAP: Which Persistent Homology Class Gets the Credit?
LandscapeSHAP assigns prediction credit back to individual persistence diagram points, with an exact closed form for linear landscape models.
The paper says Shapley values have not previously been used to explain models trained on topological data analysis features. Its method handles the hard part: persistence landscape coordinates are rank statistics, so credit does not map cleanly back to a single homology class. For nonlinear models, the exact value still requires averaging over all `2^N` coalitions, so the author adds Monte Carlo approximation and convergence bounds. The paper also proves stability results for the credit allocation.
ArXiv · AI/CL/LG's note
The paper says Shapley values have not previously been used to explain models trained on topological data analysis features. Its method handles the hard part: persistence landscape coordinates are rank statistics, so credit does not map cleanly back to a single homology class. For nonlinear models, the exact value still requires averaging over all `2^N` coalitions, so the author adds Monte Carlo approximation and convergence bounds. The paper also proves stability results for the credit allocation.
ArXiv · AI/CL/LG's note
score 3