Active learning for ecological monitoring

Purpose and goal

In this project, we will develop novel active learning algorithms tailored to applications within ecological monitoring. By incorporating active learning techniques, we aim to enhance the efficiency, accuracy, and effectiveness of data collection and analysis, with the long-term goal of developing automated monitoring devices. The resulting techniques will be benchmarked, compared to existing techniques, and tested using real-world problems at the Swedish University of Agricultural Sciences (SLU).

Expected effects and result

The project will develop solutions that adapts and extends current state of the art for active learning, tailoring these to soundscape analysis and ecological monitoring. At least two scientific publications will be produced and submitted to high impact conferences or journals. The developed techniques will be demonstrated and benchmarked on real-world data from SLU.

Planned approach and implementation

This project is organized around four work packages which will all run for the duration of the project. WP1: Project administration 97h, WP2: Hierarchical acquisition functions 445h, WP3: Multi-oracle solutions 445h, and WP4: Pathway to impact 134h. In WP3 we will develop and implement our ideas on hierarchical acquisition functions of increasing model complexity selecting smaller and smaller subsets of the unlabeled data for annotation will be developed. WP4 will explore the impact of multi-oracle solutions, leveraging different strengths from different annotators.