Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Causal Multiscale Models of Temporally Resolved Auditory Processing in Human Cortex

Michele Esposito¹, Tonio Weidler², Christian Ferreyra¹, Bruno L. Giordano³, Elia Formisano²; ¹Université d'Aix-Marseille, ²Maastricht University, ³CNRS

Presenter: Tonio Weidler

Most existing neuroconnectionist models of sound recognition assume access to the full spectrogram and predict a single class for the duration of the sound. However, human auditory processing is inherently online and causal, i.e., decisions about categories can only depend on past information and happen in real-time. To better align with these constraints, we propose a time-resolved multiscale convolutional recurrent neural network model of auditory cortex, explicitly designed to process acoustic information sequentially, generating temporally evolving predictions. We show that by processing the sound along multiple temporal scales, a design choice motivated by neuroanatomical evidence, the model can overcome performance limitations introduced by the causality constraint.

Topic Area: Auditory, Speech & Language Processing