Master 2017 2018
Stages de la spécialité SAR
Predicting Neural Responses to Music and Speech with Structured Neural Networks

Much remains unknown about the functional organization and computational properties of human auditory cortex, despite its critical role in speech understanding, music perception, and environmental sound recognition. Traditional models of the auditory cortex have relied on relatively simple hand-engineered feature to predict cortical responses. These models explain some of the responses in primary auditory cortex, where sound is first processed by the cortex, but fail in higher-order non-primary regions, which are critical to representing abstract properties of sounds (e.g. speech phonemes and musical melodies). The goal of this internship is to use tools from machine learning (e.g. convolutional neural networks) to learn features that are better able to predict responses in higher-order brain regions, and to then investigate the properties of the learned features. These analyses will take advantage of neural responses to a wide variety of natural sounds (i.e. speech, music, mechanical sounds, etc.) measured directly from the surface of the human brain using electrocorticography (ECoG), which provides the best temporal and spatial resolution of any human neuroscience technique.