Dylan Paiton and Liz Lawler
Monday, November 19, 2018
11:10 am - 12:30 pm
489 Minor Hall
Dylan Paiton's Talk: A Geometric Analysis of Population Interactions Among Primary Sensory Neurons
Neurons compute in high dimensional, non-linear spaces. A given neuron's output is a highly non-linear function of many inputs that carry vastly different information. For example, individual primate layer 4 V1 neurons receive inputs that originate from feed-forward, lateral, and feedback projections. I will discuss my recent research on trying to better understand how the type of non-linearity we assign to our neuron models influences the neuron's information processing capability. I will first present a recent technique for visualizing the response properties of neurons. Next, I will apply the technique to model neurons in order to compare the effect of different non-linearity classes. Finally, I will use the technique to demonstrate unique properties of a novel neural network architecture.
Liz Lawler's Talk: Unconscious learning of simple stimuli leads to perceptual suppression of the expected
Perception relies on making predictions about the environment, and these predictions are informed by prior experiences. I will discuss how various methods of creating a prediction about a stimulus affects perception selection. Next, I will then present a study that aims to examine the effects of stimulus complexity and the method of inducing predictive context on perceptual selection. I used statistical learning to teach observers arbitrary sequences of grating orientations and then used binocular rivalry to measure the effects of statistical learning on perceptual selection. Finally, I will show that exposure to recently acquired, arbitrary sequential structures impacts subsequent visual perception and awareness, causing the visual system to prioritize the unexpected over the expected.