Statistical Classification via Robust Hypothesis Testing: Non-Asymptotic and Simple Bounds

We consider Bayesian multiple statistical classification problem in the case where unknown source distributions are estimated from labeled training sequences, then estimates used as nominal a robust hypothesis test. Specifically, we employ DGL test due to Devroye et al. and provide non-asymptotic, exponential upper bounds on error probability of classification. The proposed simple evaluate reveal effects length alphabet size numbers exponent. method can also be for large sources when grows sub-quadratically sequence. simulations indicate that performance gets close optimal testing sequences increases.