The query complexity of certification
From MaRDI portal
Abstract: We study the problem of {sl certification}: given queries to a function with certificate complexity and an input , output a size- certificate for 's value on . This abstractly models a central problem in explainable machine learning, where we think of as a blackbox model that we seek to explain the predictions of. For monotone functions, a classic local search algorithm of Angluin accomplishes this task with queries, which we show is optimal for local search algorithms. Our main result is a new algorithm for certifying monotone functions with queries, which comes close to matching the information-theoretic lower bound of . The design and analysis of our algorithm are based on a new connection to threshold phenomena in monotone functions. We further prove exponential-in- lower bounds when is non-monotone, and when is monotone but the algorithm is only given random examples of . These lower bounds show that assumptions on the structure of and query access to it are both necessary for the polynomial dependence on that we achieve.
Recommendations
- The quantum query complexity of certification
- Certified complexity (CerCo)
- scientific article; zbMATH DE number 2044532
- Query Learning and Certificates in Lattices
- Certification of complexity proofs using CeTA
- The complexity of higher-order queries
- On the complexity of database queries
- Quantum certificate complexity
- The complexity of evaluating relational queries
- From query complexity to computational complexity
This page was built for publication: The query complexity of certification
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6083517)