Jet origin identification and measurement of rare hadronic decays of Higgs boson at $e^+e^-$ collider

We propose to identify the jet origin using deep learning tools for experiments at the high energy frontier, where jet origins are categorized into 5 species of quarks, i.e., $b,c,s,u,d$, 5 species of anti-quarks, i.e., $\bar{b},\bar{c},\bar{s},\bar{u},\bar{d}$, and gluons. Using simulated physics events at the Circular Electron Positron Collider and the ParticleNet algorithm, we quantify the performance of jet origin identification using an 11-dimensional confusion matrix. This matrix exhibits flavor tagging efficiencies of 91% for $b$ and $\bar{b}$, 80% for $c$ and $\bar{c}$, and 64% for $s$ and $\bar{s}$ quarks, as well as jet charge misidentification rates of 18% for $b$ and $\bar{b}$, 7% for $c$ and $\bar{c}$, and 16% for $s$ and $\bar{s}$ quarks, respectively. We use this method to determine the upper limits on branching ratios of Higgs rare hadronic decays, specifically for $s\bar{s}$, $u\bar{u}$, and $d\bar{d}$, as well as for decays via flavor-changing neutral current, such as $sb$, $sd$, $db$, $cu$. We conclude that these Higgs decay branching ratios could be measured with typical upper limits of 0.02%-0.1% at 95% confidence level at CEPC nominal parameters. For the $H\rightarrow s \bar{s}$ decay, this upper limit corresponds to three times the standard model prediction.