This paper presents the notion of meta-learner (NML), an innovative meta-learning methodology designed to enhance the performance of few-shot learning (FSL) regarding the recognition of music genres. Current FSL techniques frequently encounter difficulties due to the absence of organized representations and low capacity for generalization, which impede their efficacy in practical scenarios. The NML meta-learner overcomes these obstacles by acquiring the ability to learn across notion dimensions that humans can understand, thus improving its capacity for generalization and interpretability. Instead of gaining knowledge in a combined and disorganized metric space, the notion meta-learner acquires knowledge by mapping high-level notions into partially organized metric spaces. This technique allows for the efficient integration of several notion learners. We assessed the performance of NMLFSL by utilizing the GTZAN dataset and comparing employing seven different benchmarks. The experimental outcomes show that the NML performs superior to current FSL approaches in tasks that include recognizing music genres with only one or five examples, thereby demonstrating its potential to improve the current state of the art in this field. In addition, ablation experiments assess the influence of essential variables, offering valuable information about the effectiveness of the suggested method. NMLFSL is a notable advancement in using meta-learning to enhance the reliability and precision of music genre recognition (MGR) systems.
Except where otherwise noted, this item's license is described as CC0 1.0 Universal