Music Minion Guided Evolutionary Multitasking: A Case Study in Piano Fingering Estimation

Piano fingering estimation is an important task in piano performance and pedagogy, aiming to assign effective fingering sequences that support fluent and expressive performance. Musical pieces often share recurring motifs and structural patterns, creating opportunities to solve multiple fingering estimation tasks jointly through shared information. In this work, we introduce an evolutionary multitask optimization framework to address multiple fingering estimation tasks simultaneously by enabling knowledge transfer across them. To further enhance the multitask search process, we incorporate auxiliary optimization tasks derived from classical technical exercises into the multitask framework, which help guide each main task toward promising regions of its respective solution space. These auxiliary tasks are assigned to each main task based on their mutual musical relevance. To support future research, we also present the Piano Scale Dataset, a collection of annotated exercises covering a comprehensive range of musical keys. Experimental results show that the proposed framework improves convergence and fingering accuracy compared to baseline methods, highlighting the potential of multitask optimization for music-related estimation problems.