Paper detail

Phase 4: DCL System Using Deep Learning Approaches for Land-Based or Ship-Based Real-Time Recognition and Localization of Marine Mammals - Distributed Processing and Big Data Applications

While the animal bioacoustics community at large is collecting huge amounts of acoustic data at an unprecedented pace, processing these data is problematic. Currently in bioacoustics, there is no effective way to achieve high performance computing using commericial off the shelf (COTS) or government off the shelf (GOTS) tools. Although several advances have been made in the open source and commercial software community, these offerings either support specific applications that do not integrate well with data formats in bioacoustics or they are too general. Furthermore, complex algorithms that use deep learning strategies require special considerations, such as very large libraiers of exemplars (whale sounds) readily available for algorithm training and testing. Detection-classification for passive acoustics is a data-mining strategy and our goals are aligned with best practices that appeal to the general data mining and machine learning communities where the problem of processing large data is common. Therefore, the objective of this work is to advance the state-of-the art for data-mining large passive acoustic datasets as they pertain to bioacoustics. With this basic deficiency recognized at the forefront, portions of the grant were dedicated to fostering deep-learning by way of international competitions (kaggle.com) meant to attract deep-learning solutions. The focus of this early work was targeted to make significant progress in addressing big data systems and advanced algorithms over the duration of the grant from 2012 to 2015. This early work provided simulataneous advances in systems-algorithms research while supporting various collaborations and projects.