Jarrod Hicks (McDermott Lab)

Segregation from background noise as outlier detection

How do we detect auditory events amid clutter? We are exploring the idea that listeners fit distributions to an environment's background texture and register outliers of these distributions as new events. In this talk, we investigate human abilities to detect outliers embedded in textures, propose a simple model for outlier detection, and suggest future directions and applications of this research.

Mark Saddler (McDermott Lab)

Neural Network Models of Pitch Perception

The perceived pitch of a sound depends on both spectral and temporal information available in the auditory periphery, but the relative contribution of the various cues and the reasons for their varying importance are poorly understood. We investigated whether the well-characterized properties of human pitch perception would emerge simply from optimizing a convolutional neural network to estimate fundamental frequency from cochlear representations of natural sounds. When tested on classic stimulus manipulations from the psychophysics literature, the trained network exhibited many of the known dependences of human pitch discrimination on stimulus parameters such as harmonic composition and relative phase. To better understand how the dependencies of pitch perception arise from either constraints of the peripheral auditory system or from statistics of sounds in the world, we independently manipulated parameters of the peripheral model and the training corpus. The results collectively suggest that human pitch perception can be understood as having been optimized to estimate the fundamental frequency of natural sounds heard through a human cochlea.

Heather Kosakowski (Saxe and Kanwisher Labs)

Development of high-level visual areas in human infants

Virtually every healthy adult has areas of the brain that are selectively involved in processing visual categories such as faces, bodies, and scenes. How do these regions of cortex develop in healthy humans? I address this question by scanning awake infants using functional magnetic resonance imaging (fMRI). I have scanned 31 infants and obtained usable data from 21 of those infants. I have validated these data by replicating previous results demonstrating that the organization of face and scene responses in infants and adults is similar. My future work will use representational similarity analysis (RSA) to compare the representational structure of infant cortical responses to adults as well as various models of development, including convolutional neural networks.