When Noise Lowers The Loss: Rethinking Likelihood-Based Evaluation in Music Large Language Models
Georgia Institute of Technology(佐治亚理工学院);ByteDance Inc.(字节跳动公司);Courant Institute of Mathematical Sciences, New York University(纽约大学应用数学科学研究所);Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)(Mohamed bin Zayed人工智能大学)
Xiaosha Li, Chun Liu, Ziyu Wang
备注:
Accepted by IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP) 2026
The rise of music large language models (LLMs) demands robust methods of evaluating output quality, especially in distinguishing high-quality compositions from "garbage music". Curiously, we observe that the standard cross-entropy loss -- a core training metric -- often decrease when models encounter systematically corrupted music, undermining its validity as a standalone quality indicator. To investigate this paradox, we introduce noise injection experiment, where controlled noise signal of varying lengths are injected into musical contexts. We hypothesize that a model's loss reacting positively to these perturbations, specifically a sharp increase ("Peak" area) for short injection, can serve as a proxy for its ability to discern musical integrity. Experiments with MusicGen models in the audio waveform domain confirm that Music LLMs respond more strongly to local, texture-level disruptions than to global semantic corruption. Beyond exposing this bias, our results highlight a new principle: the shape of the loss curve -- rather than its absolute value -- encodes critical information about the quality of the generated content (i.e., model behavior). We envision this profile-based evaluation as a label-free, model-intrinsic framework for assessing musical quality -- opening the door to more principled training objectives and sharper benchmarks.
Rethinking External Slow-Thinking: From Snowball Errors to Probability of Correct Reasoning
Gaoling School of Artificial Intelligence, Renmin University of China, Beijing, China(中国人民大学人工智能学院);Beijing Key Laboratory of Research on Large Models(北京大型模型研究关键实验室);Engineering Research Center of Next-Generation Intelligent Search(下一代智能搜索工程研究中心);College of Artificial Intelligence, Tianjin University of Science(天津科技大学人工智能学院)
MSVAE: Rethinking Musical Semantics in Continuous Latent Spaces
Geely Automobile Research Institute (Ningbo) Company Ltd(吉利汽车研究院(宁波)有限公司);Faculty of Electronic and Information Engineering, Xi’an Jiaotong University(西安交通大学电子与信息工程学院);WeNet Open Source Community(WeNet开源社区)
When 1+1 Backfires: Rethinking Relevance-Diversity Balancing for Sample-Aware Visual Token Reduction
State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences(中国科学院计算技术研究所人工智能安全国家重点实验室);University of Chinese Academy of Sciences(中国科学院大学);Beijing Academy of Artificial Intelligence(北京智源人工智能研究院)
Rethinking Repetition Problems of LLMs in Code Generation
Key Laboratory of High Confidence Software Technologies (Peking University), Ministry of Education(高可信软件技术教育部重点实验室(北京大学));School of Computer Science, Peking University(北京大学计算机学院)