Merging cutting-edge technologies for data manipulation on dateseries, high-performing APIs, and standard astrophysics data transporation methods, we've created a high-performing distributor for the future of seismic data analysis, supporting a great variety of data sources with adaptability.
Our model was developed under milimetric tuning with Neural Architecture Search (NAS) with multi-objective optimization. Physics informed neural networks were embedded to increase explainability and decrease model robustness, optimizing power consumption. It consists on a multiheaded like algorithm, with one binary classifier that accounts for P-wave and S-wave arrival, and the reconstructor that proposes a different representation of the data that increases interpretability and decreases its resolution to send it directly to the earth.
Energy efficiency, Deep Learning supremacy, and Open Science.
