End-to-end Topographic Auditory Models Replicate Signatures of Human Auditory Cortex
Abstract
The human auditory cortex is topographically organized. Neurons with similar response properties are spatially clustered, forming smooth maps for acoustic features such as frequency in early auditory areas, and modular regions selective for music and speech in higher-order cortex. Yet, evaluations for current computational models of auditory perception do not measure whether such topographic structure is present in a candidate model. Here, we show that cortical topography is not present in the previous best-performing models at predicting human auditory fMRI responses. To encourage the emergence of topographic organization, we adapt a cortical wiring-constraint loss originally designed for visual perception. The new class of topographic auditory models, TopoAudio, are trained to classify speech, and environmental sounds from cochleagram inputs, with an added constraint that nearby units on a 2D cortical sheet develop similar tuning. Despite these additional constraints, TopoAudio achieves high accuracy on benchmark tasks comparable to the unconstrained non-topographic baseline models. Further, TopoAudio predicts the fMRI responses in the brain as well as standard models, but unlike standard models, TopoAudio develops smooth, topographic maps for tonotopy and amplitude modulation (common properties of early auditory representation, as well as clustered response modules for music and speech (higher-order selectivity observed in the human auditory cortex). TopoAudio is the first end-to-end biologically grounded auditory model to exhibit emergent topography, and our results emphasize that a wiring-length constraint can serve as a general-purpose regularization tool to achieve biologically aligned representations.
The following analysis is generated by AI and may contain errors. It is separate from the authors’ abstract and is not peer review.
Research summary
The main contribution of this paper is the introduction of TopoAudio, a biologically grounded auditory model that successfully replicates the topographic organization of the human auditory cortex while maintaining high performance on auditory tasks. This work significantly advances the understanding of how topographic structures can be integrated into machine learning models, providing a framework for future research in both AI and neuroscience.
Methodology
The paper introduces TopoAudio, a novel class of auditory models that incorporates a cortical wiring-constraint loss to encourage topographic organization in neural networks. The methodology is well-grounded in biological principles and effectively adapts existing techniques from visual perception to the auditory domain. The use of two distinct architectures (CochResNet50 and AST) enhances the robustness of the findings, and the evaluation of smoothness and selectivity metrics provides a comprehensive assessment of the model's performance.
Results & evaluation
The experiments are thorough, utilizing multiple benchmark datasets for auditory classification and fMRI response prediction. The results demonstrate that TopoAudio maintains high accuracy while developing biologically relevant topographic structures. The comparison with baseline models is well-executed, and the quantitative and qualitative analyses of tonotopic and amplitude modulation maps substantiate the claims made about the model's performance.
Reproducibility
The paper provides detailed descriptions of the methodologies, datasets, and evaluation metrics used, which facilitates reproducibility. However, the absence of a publicly accessible code repository or demo limits the ability for others to directly replicate the findings.
Limitations
The paper acknowledges that the current work is a starting point and suggests that further exploration of different architectures and datasets could yield additional insights into auditory cortical organization. Additionally, the qualitative nature of some findings may require more quantitative validation.
Broader impact
The work has significant implications for both machine learning and neuroscience, as it bridges the gap between computational models and biological systems. By demonstrating that topographic organization can emerge from task-driven optimization, the research opens avenues for more biologically inspired AI systems that could lead to advancements in auditory processing applications. The main contribution of this paper is the introduction of TopoAudio, a biologically grounded auditory model that successfully replicates the topographic organization of the human auditory cortex while maintaining high performance on auditory tasks. This work significantly advances the understanding of how topographic structures can be integrated into machine learning models, providing a framework for future research in both AI and neuroscience.
Author of this paper? Send a correction.