Greedy Volume Maximization of Gradient Embeddings for Long-Tailed Frame-Level Bioacoustic Active Learning

2026-07-15

AI Summary

This work addresses the high annotation cost inherent in bioacoustic call classification tasks characterized by sparse, long-tailed distributions. To efficiently select the most informative rare samples under a limited labeling budget, the authors propose BADGE-Greedy-DPP, which integrates greedy volume maximization via Determinantal Point Processes (DPPs) into the BADGE framework for the first time. Leveraging the submodularity of DPPs, this approach provides theoretical guarantees of near-optimality while enhancing diversity and informativeness in sample selection. Furthermore, a frame-level pseudo-gradient aggregation mechanism weighted by prediction residuals ensures that critical rare frames dominate segment selection. Experiments on a hyena vocalization dataset demonstrate that the proposed method significantly outperforms baseline approaches—including MFFT and the original BADGE—both overall and specifically on rare classes.

Abstract

Bioacoustic call-type classification relies on costly expert annotation. Active learning can reduce this burden by selecting a small batch of segments for expert annotation and using the labeled segments for training the classifier. The setting is hard: the target calls are extremely sparse and the call-type distribution is long-tailed, so a tight budget must be spent on the few rare, informative segments. We propose BADGE-Greedy-DPP, a deterministic batch selector that greedily adds the segment whose BADGE gradient embedding most enlarges the volume spanned by the batch; because this log-volume objective is submodular, the greedy rule guarantees a batch value at least a (1-1/e) fraction of the optimum of this objective, a guarantee not provided by BADGE's existing k-means++ and MCMC DPP sampling heuristics. There is also a temporal granularity mismatch in the task. The acquisition function scores whole segments, yet the informative frames inside them are few. Uniform averaging therefore washes them out. We show that the BADGE construction naturally addresses this mismatch when applied frame-wise, as prediction residuals weight the aggregated pseudo-gradient, so confidently predicted no-call frames contribute little while a single uncertain rare-call frame can still set the segment's direction. Across 10 runs on a sparse, imbalanced hyena call-type dataset, BADGE-Greedy-DPP achieves the best overall and rare-call-type performance among all compared query strategies, including MFFT, the strongest non-BADGE baseline, and the two vanilla BADGE traversals.

Problem

• active learning
• long-tailed distribution
• bioacoustic classification
• frame-level selection
• annotation efficiency

Innovation

• active learning
• submodular optimization
• gradient embedding
• long-tailed classification
• bioacoustic call-type