Researchers have employed deep neural networks to enhance the search for heavy Majorana neutrinos ($N_R$) and $W_R$ bosons at the Large Hadron Collider (LHC). These components are predicted by the Left-Right Symmetric Model, an extension of the Standard Model that could explain neutrino mass and the matter-antimatter asymmetry. Right-handed lepton flavor mixing, a phenomenon analogous to neutrino mixing in the Standard Model, directly influences the production and decay of these $N_R$, and its impact on collider experiments has been less explored until now.

The study focused on the Keung-Senjanović process ($pp \to W_R \to \ell_\alpha N_R \to \ell_\alpha \ell_\beta jj$) with leptons $\ell_{\alpha,\beta}=e,\mu$, analyzing both same-charge and opposite-charge dilepton channels. Three flavor mixing scenarios were adopted: no mixing, maximal mixing, and a PMNS (Pontecorvo-Maki-Nakagawa-Sakata) matrix-like mixing. The application of deep neural networks (DNNs) significantly improved the expected sensitivity compared to traditional cut-based analyses, such as those performed by the ATLAS experiment, allowing for stricter exclusion limits on the masses of $W_R$ and $N_R$.

For the combined dilepton analysis, the High-Luminosity LHC (HL-LHC) could exclude $m_{W_R}$ and $m_{N_R}$ masses up to 6.7 TeV and 4.4 TeV, respectively, under the maximal mixing scenario, and 6.3 TeV and 4.1 TeV for PMNS-like mixing. Run 2 LHC data has already excluded a considerable portion of the $|V_{e1}|-|V_{\mu1}|$ parameter plane, and the HL-LHC will probe even smaller mixing values, potentially ruling out maximal and PMNS-like mixing patterns. Furthermore, complementarities with low-energy charged lepton flavor violation processes were investigated, where future searches could overlap or even surpass the LHC's reach.