Toward Asymptotic Optimality: Sequential Unsupervised Regression of Density Ratio for Early Classification

TL;DR

This paper introduces two novel SPRT-based algorithms, B2Bsqrt-TANDEM and TANDEMformer, that improve the accuracy of density ratio estimation for early classification of time series, achieving asymptotic optimality.

Contribution

The paper proposes new SPRT-based algorithms that overcome overnormalization issues in density ratio estimation, enhancing early classification performance.

Findings

Significantly reduce density ratio estimation errors.

Achieve better early classification accuracy on multiple datasets.

Demonstrate asymptotic Bayes optimality in theory.

Abstract

Theoretically-inspired sequential density ratio estimation (SDRE) algorithms are proposed for the early classification of time series. Conventional SDRE algorithms can fail to estimate DRs precisely due to the internal overnormalization problem, which prevents the DR-based sequential algorithm, Sequential Probability Ratio Test (SPRT), from reaching its asymptotic Bayes optimality. Two novel SPRT-based algorithms, B2Bsqrt-TANDEM and TANDEMformer, are designed to avoid the overnormalization problem for precise unsupervised regression of SDRs. The two algorithms statistically significantly reduce DR estimation errors and classification errors on an artificial sequential Gaussian dataset and real datasets (SiW, UCF101, and HMDB51), respectively. The code is available at: https://github.com/Akinori-F-Ebihara/LLR_saturation_problem.