Property Testing with Incomplete or Manipulated Inputs
summary
Introduces an online erasure/corruption oracle that, after each query, can erase or corrupt up to t input locations; uses Fourier analysis and affine-subspace sampling to prove concrete complexity changes. Concretely, linearity remains testable with the usual 1/ε dependence plus a small (provably necessary) overhead in t (polylogarithmic), low-degree testing requires polylog(t) overhead, while sortedness becomes untestable and Lipschitz arrays are likewise impossible even for t=1. Shows separations (an artificial property testable online but not offline), phase transitions with batched queries, and links to Maker–Breaker games.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 1:47Sublinear algorithms motivation
- 3:11Erasure and corruption oracles
- 6:02Property testing model overview
- 7:53Adversarial models comparison
- 8:35Talk plan and extreme properties
- 9:02Linearity and low-degree results
- 12:03Why linearity testing works
- 17:29Low-degree testing generalization
- 27:50Impossibility of online sortedness
- 30:01Online versus offline separations
- 32:14Query complexity separations
- 33:26Randomness separation
- 34:55Batch queries and strategies
- 36:22Local properties and transitions
- 39:02Maker-Breaker connection
- 41:10Open questions
- 42:58Closing
