ResearchI’m a PhD candidate working in the labs of Ari Rosenberg and Bas Rokers. My thesis focuses on understanding how the brain extracts, weighs, and combines multiple different sensory signals to create our perception of 3D motion. I approach this problem using three different methodologies: traditional psychophysics, electrophysiology, and neural network modeling.
|
Psychophysics
A large portion of my current research endeavors involves evaluating human and non-human primate behavior pertinent to 3D motion perception. To do so, I make use of classic psychophysical research techniques and signal processing models. Stimuli for these tasks are generated using various 3D display technologies, MATLAB, and OpenGL. I selectively manipulate cues to 3D motion that a single eye can process (monocular cues) as well as those requiring binocular integration (binocular cues). The scope of this research is threefold: First, we have already evaluated the sensitivity of observers to individual cues throughout the visual field. Secondly, we have modeled behavior when provided both sets of cues (natural viewing) using a maximum likelihood estimation based on the individual cue sensitivities. Lastly, work is currently ongoing to determine whether sensitivity to these cues throughout the visual field is gravity-centered or retinotopic.
Electrophysiology
Our behavioral results inform our investigations into the underlying neural mechanisms of 3D motion processing. The same behavioral tasks we use to quantify behavioral sensitivity are performed by non-human primates while we collect single unit and local field potential measurements using electrophysiology in cortical regions of interest.
Neural Network Modeling
One of the goals of my Ph.D. research via my involvement in the LUCID training program, is to develop a performance optimized neural network model of the dorsal visual stream. Specifically, a binocular neural network model undergoes unsupervised learning, where two parallel convolutional neural networks (CNNs) receive images with proper eye-specific vantage points of natural scenes. These parallel streams are later integrated into a single ("cyclopean") CNN stream - much like we expect happens in the brain. After undergoing this stage of unsupervised learning, we use the same 3D stimuli tested in both human and non-human primates to directly compare the neuronal activity in our observers, to that of the CNN. Such a model can provide insight into the underlying computations that are occurring in higher-order cortical areas which are often difficult to interpret with neural data alone.
Conference Proceedings
- Ji, M., Thompson, L., Rokers, B., Rosenberg, A. The contribution of monocular and binocular cues to the perception of 3D motion. Vision Sciences Society Seventeenth Annual Meeting, St. Pete’s Beach, FL, May 20th, 2017 (Poster).
- Ji, M., Thompson, L., Rokers, B., Rosenberg, A. The neural basis of visual agnosia: Developing a model system for cortical blindness. 8th Annual McPherson ERI Vision Science Poster Session, Madison, WI, October 4th, 2016 (Poster).
- Thompson, L., O’Brien, A. Embracing the whole spectrum: An investigation of females with tetrachromacy. Midwestern Psychological Association, Chicago, IL, May 2015 (Poster).
- Thompson, L., O’Brien, A. Embracing the whole spectrum: An investigation of females with tetrachromacy. Celebration of Undergraduate Research, La Crosse, WI, April 2015 (Poster).
- Thompson, L., O’Brien, A. The effect of fatigued photoreceptors on color perception Celebration of Undergraduate Research, La Crosse, WI, April 2014 (Poster).