Coupled Scaling: A Representational Accessibility Framework for Neural Scaling Laws
本文提出耦合缩放框架，通过任务结构与架构-优化系统可访问几何关系解决神经缩放问题。
0 citationsRead paper
Selected work
Representative Papers
The Voiceprint Fallacy: Why Voices Are Not Unique Biometric Imprints
This study critiques the “voiceprint” metaphor, which misleadingly treats the human voice as a stable and unique biometric akin to a fingerprint, and exposes its problematic implications in both technological applications and policy. Integrating insights from phonetics, forensic voice comparison, automatic speaker recognition, deepfake detection, and probabilistic reasoning, the paper systematically examines vocal variability, courtroom practices of voice identification, and challenges in identity verification. It argues against the assumption of a deterministic, fixed vocal signature and advocates instead for a calibrated probabilistic framework that accounts for the dynamic, context-dependent nature of speech. The work calls for a paradigm shift in evaluating voice evidence—one that rigorously incorporates variability, uncertainty, and alternative explanations.
0 citationsRead paper
Recent publications
Latest Papers
Coupled Scaling: A Representational Accessibility Framework for Neural Scaling Laws
本文提出耦合缩放框架，通过任务结构与架构-优化系统可访问几何关系解决神经缩放问题。
0 citationsRead paper
The Voiceprint Fallacy: Why Voices Are Not Unique Biometric Imprints
This study critiques the “voiceprint” metaphor, which misleadingly treats the human voice as a stable and unique biometric akin to a fingerprint, and exposes its problematic implications in both technological applications and policy. Integrating insights from phonetics, forensic voice comparison, automatic speaker recognition, deepfake detection, and probabilistic reasoning, the paper systematically examines vocal variability, courtroom practices of voice identification, and challenges in identity verification. It argues against the assumption of a deterministic, fixed vocal signature and advocates instead for a calibrated probabilistic framework that accounts for the dynamic, context-dependent nature of speech. The work calls for a paradigm shift in evaluating voice evidence—one that rigorously incorporates variability, uncertainty, and alternative explanations.
0 citationsRead paper