Status of the Machine Learning Efforts at the International Data Centre of the Ctbto

Machine learning projects were conceived in March 2009 as part of the International Scientific Studies Project initiative at the Provisional Technical Secretariat of the Comprehensive Nuclear-Test-Ban Treaty Organization (CTBTO) and initiated a few months later. Some of the projects are intended to aim at short to medium term operational applications. These include the identification of seismic and hydroacoustic phase names using a large number of features extracted from the waveforms, and the labeling of automatic events as real or false depending again on a large number of features from the automatic events. Concrete research results using International Data Centre (IDC) data are available for these two sets of projects. Seismic phase identification is shown to have the potential to improve its accuracy by 23 %, and the software developed for the project on false events identification has been tested at the IDC and shown to correctly label 80% of the false alarms. Some projects are aimed at the longer term. This is the case of a Bayesian approach to the automatic seismic network processing problem and a distributed database approach to the waveform cross-correlation problem. The first project is well under way and has surpassed the current operational system by 14 % in accuracy for the same false alarm rate. The second has shown the potential of distributed systems to solve efficiency issues. 2010 Monitoring Research Review: Ground-Based Nuclear Explosion Monitoring Technologies