Toward Trustworthy Symbolic Music Generation in 6G-Enabled IoT Environment via Automatic Neural Architecture Design

The rise of 6G-enabled Internet of Things (IoT) environments opens new frontiers for deploying intelligent music generation systems that must operate in real-time, across heterogeneous devices, and under strict resource constraints. Traditional symbolic music generation (SMG) methods are often manually designed without considering the complex and dynamic requirements of 6G networks, such as ultra-low latency, high connectivity, and resource efficiency. These limitations result in unreliable performance and poor adaptability to diverse edge devices. To address these challenges, we propose a novel evolutionary multi-objective framework for the automatic design of Transformer-based architectures tailored for 6G IoT environments. Unlike previous methods, our approach incorporates trustworthiness as an optimization objective alongside generation quality and inference latency. This ensures that the generated music is both coherent and reliable, even under varying network conditions. Furthermore, we introduce a population memory mechanism that reuses architectural encodings to reduce computational costs and improve search efficiency. We conduct both objective and qualitative evaluations on the POP909 dataset, demonstrating that the discovered architectures produce musically coherent and stylistically faithful compositions. Additionally, real-world deployment on 6G-compatible edge hardware shows that our architectures perform efficiently under constrained resources, exhibiting excellent adaptability across different devices. Our framework bridges the gap between high-quality SMG and trustworthy real-time deployment in 6G IoT environments, providing a scalable path toward next-generation intelligent music services.