AI RESEARCH
Latent Space Disentanglement via Activation Steering for Interpretable Attribute Control in Symbolic Music Generation
arXiv CS.AI
•
ArXi:2605.31295v1 Announce Type: cross Transformer-based architectures have significantly advanced the generation of complex symbolic sequences, yet a significant gap remains in achieving fine-grained, interpretable control over discrete signal attributes. This paper investigates the mechanistic interpretability of the Multitrack Music Transformer (MMT) and proposes a framework for deterministic attribute modulation without re