C4DM Seminar: Pre-ISMIR Presentations
QMUL, School of Electronic Engineering and Computer Science
Centre for Digital Music Seminar Series
Seminar by: Zixun (Nicolas) Guo, Jinwen Zhou, Yinghao Ma, Weixiong Chen
Date/time: Wednesday, 4th November 2026, 14:00 - 15:00
Location: GC601, Graduate Center, Mile End Campus, Queen Mary University of London
Teams link: https://teams.microsoft.com/meet/3263075600728523?p=9S1B0S8hlMZrtfDM3j
Title: Pre-ISMIR Presentations
Paper: Equivariant Music Transformer
Presenter: Zixun (Nicolas) Guo (https://guozixunnicolas.github.io/)
Abstract: Humans recognize a musical passage even when it is shifted in time or transposed in pitch, indicating a notion of equivariance in the representation space. Our analysis, however, shows that standard music transformers map such time-shifted or pitch-transposed inputs onto uncorrelated representations: these models become progressively less equivariant as they scale in size or train longer. This suggests that in standard music transformers, additional model capacity is allocated to memorizing absolute patterns rather than capturing shared musical structures. In this paper, we propose the Equivariant Music Transformer (EMT), which enforces equivariance through self-distillation by jointly optimizing a next-token-prediction and an auxiliary equivariance regularization loss. We find that the additional equivariance loss acts as a beneficial regularizer, simultaneously improving next-token prediction and producing equivariant latent representations. Through both objective and subjective evaluations, EMT demonstrates superior equivariance and generative capability compared to data augmentation, feature engineering, and state-of-the-art baselines. More broadly, our findings reveal that standard language modeling methods alone do not capture music’s translational symmetries, and dedicated inductive biases are required to produce better music representations. The code, weights and demos are available online.
Paper: MAJEPPA: Morphing and Assessing in a Unified Piano Performance Space
Presenter: Jinwen Zhou
Abstract: We present MAJEPPA, a self-supervised framework to learn piano performance representations that span the full skill spectrum, from beginner practice sessions to virtuoso concert recordings. We curate the MAJEPPA dataset, comprising ~4,000 annotated recordings across six expertise levels and six recording contexts. We adapt a single pre-trained MIDI autoregressive model with a joint objective: next-token prediction learns score-conditioned performance generation at various skill levels, while InfoNCE and supervised contrastive losses align abstract score and performance representations in a joint embedding space. The proposed model both generates and understands performances in a unified framework. By introducing the EVPMR benchmark, a suite of downstream tasks spanning quality assessment, competition ranking, mistake and technique classification, we evaluate the learnt representations, demonstrating progress towards a real-world model for the piano performance space.
Tutorial: Evaluating Music Foundation Models: From MIREX To Multimodal, Preference-Aware Benchmarking
Presenter: Yinghao Ma, Weixiong Chen
Abstract: Recent progress in music foundation models, including music representation models, audio language models, multi- modal instruction-following systems, and long-form music generation models, has created an urgent need for new evaluation methodologies. Traditional MIR evaluation has largely focused on task-specific metrics and benchmark datasets, while modern music foundation models require broader evaluation across representation quality, instruction following, multimodal alignment, musicality, and human preference. This tutorial introduces a unified view of music evaluation in the foundation model era. We begin by revisit- ing the role of evaluation in MIR through the lens of MIREX and related benchmark traditions, and discuss how these ideas can be extended to support faster iteration of music foundation models. We then cover evaluation for represen- tation learning, using MARBLE and its later developments as examples of benchmark design for universal music audio representations. Next, we examine supervised fine-tuning and instruction-following evaluation for audio-LLMs, fo- cusing on the design principles behind CMI-Bench 2.0 and the limitations of current generation-based evaluation for music. We further introduce evaluation metrics and reward- model benchmarks, including CMI-RewardBench, and dis- cuss how such frameworks may inspire new MIREX-style tasks for music generation and multimodal systems. Finally, we outline future directions on how evaluation can support reinforcement-learning-based post-training, preference op- timization, and tokenizer design for next-generation music models. This tutorial is intended for MIR researchers, ma- chine learning practitioners, and newcomers interested in how evaluation can serve as a bridge between traditional mu- sic information retrieval and emerging foundation-model- based music systems. It aligns strongly with the ISMIR 2026 theme of “Crossroads” by connecting established MIR evaluation culture with new paradigms in generative and multimodal AI.
