Dating Cool Stars with Neural Networks and Spotted Models. Tracing the Star Formation History in the Local Bubble

Accurately determining the fundamental parameters and ages of pre-main sequence (PMS) cool stars is a challenge, hampered by the complex interplay of extinction, distance, and intense magnetic activity. In this contribution, we demonstratehow young open clusters can serve as ideal laboratories to calibrate stellar ages by explicitly accounting for magnetic activity effects, and how these refined ages can resolvemacroscopic debates in Galactic star formation. We present a novel Deep Learning framework (Tarantino et al. 2025) designedto predict robust effective temperatures for cool stars (<7000 K). Trained on high-qualityGaia-ESO Survey spectroscopy and applied to Gaia DR3 and 2MASS photometry, our Neural Network bypasses the limitations of purely photometric color-Teff relations. By combiningthese temperatures with state-of-the-art evolutionary models parameterized by the starspotcoverage fraction (β), we derive highly precise isochronal ages for a large sample of low-mass stars in young clusters (<100 Myr). We validate our method against model-independent age tracers, such as Lithium depletion, demonstrating that the inclusionof magnetic activity (spots) is strictly required to obtain reliable age estimates and tocorrectly interpret the intrinsic age spreads observed in young stellar populations.Finally, we present the Galactic impact of this calibration by tracing the 3D spatio-temporal architecture of the Solar Neighborhood. Leveraging advancedstatistical tests, we map the age distribution of young stars across the Local Bubble (LB)shell. Our unprecedented age resolution allows us to quantify the chronological gradientacross the expanding shock front, providing a purely empirical, data-driven framework totest whether the LB expansion triggered the birth of nearby stellar associations or theLB is superimposed on, and interacts with, a broader pre-existing hierarchy of recent star-forming structures (Tarantino et al. 2026, submitted).