Abstract DGP2026-63

In 2026, the search for extrasolar planets has become a well established part of astronomical research but although it incorporates various computational methods, it still heavily relies on manual analysis of light curves (LCs), a process that is both time-intensive and demanding. Our research in the EXOWORLD project addresses these challenges by integrating advanced  machine learning (ML) techniques, including convolutional neural networks, into the transit search process, combining them with recurrent networks to create a fully integrated ML-based transit detection and characterization pipeline. This approach reimagines transit search as a pattern recognition task, employing self-learning algorithms to efficiently process vast amounts of astronomical data. We aim to explore and apply a range of ML methods, establishing a foundation for comparison not only among these methods, but also against traditional transit search techniques. Although still in the early stages, our research aims to significantly enhance exoplanet detection methods, streamlining the process, and building a framework for making new discoveries through light curve (LC) analysis.
In this context, we present TRANSCENDENCE, our ML-based pipeline, which has demonstrated the ability to identify exoplanets larger than 2 Earth radii consistently. Moreover, the pipeline is capable of detecting smaller planets, albeit with lower detection probabilities. One of TRANSCENDENCE's key strengths lies in its remarkably low false positive alarm rate, which ranges between 5\% and 10\% of all identified transits on a balanced data set. By significantly reducing the need for manual intervention and minimizing false positives, this pipeline has the potential to significantly improve the efficiency of exoplanet detection and characterization. 
We attribute the observed success of our pipeline to the use of state of the art ML model architecture which was specifically tailored to the problem as well as the use of a more realistic training data that includes real star parameters and their corresponding stellar activity instead of simulated ones.