Browsing by Subject "Computation"

Now showing 1 - 4 of 4

Camouflage detection & signal discrimination : theory, methods & experiments
(2022-05-05) Das, Abhranil; Geisler, Wilson S.; Reichl, L. E.; Florin, Ernst-Ludwig; Marder, Michael
Camouflage is an amazing feat of evolution, but also impressive is the ability of biological visual systems to detect them. They are the result of an evolutionary arms race that exposes many detection strategies and their limits. In this thesis, we investigate the principles of human detection of maximally-camouflaged objects, i.e. whose texture exactly mimics the background texture. Chapter 1 introduces and contextualizes the problem. In chapter 2, we develop a theory and model that extracts the relevant information in the image, and uses biologically plausible computations on them for detection. In chapter 3, we present a series of experiments which measured human camouflage detection ability along different dimensions of the task, such as across different textures and shapes. Chapter 5 is a reference on some methods and analysis used in the study. Chapter 6 describes mathematical methods and software on statistical signal discrimination that we developed to solve questions in visual detection, but with wider applications in other fields.
A comparative study of cortical computations in the mammalian visual cortex
(2015-05) Scholl, Benjamin Kyle; Priebe, Nicholas; Geisler, Wilson S; Aldrich, Richard W; Pillow, Jonathan W; Hirsch, Judith A
A common feature of all mammals is the cerebral cortex, which is essential for higher-order functions and processing information to generate motor actions. While cortical circuits exhibit a striking uniformity in anatomical organization, it is unknown whether these circuits preform similar computations across mammalian species. In this dissertation I compare the emergence of two computations in the primary visual cortex (V1) of carnivores and rodents. A cortical computation is a transformation in neural representation, such that the spiking output of a cortical neuron exhibits a selectivity not present in the inputs from upstream neurons. Here I explore two computations: orientation selectivity, the preference of neurons for oriented edges in the visual world, and binocularity, the integration of signals from the two eyes. In the first section, I compare the emergence of orientation selectivity in the early visual pathway of mouse and cat. Recordings from thalamic relay cells and V1 neurons in both species reveal orientation selectivity in mouse V1 is not emergent, and could be inherited subcortically. In a second set of experiments, I measure orientation selectivity and the organization of V1 orientation preference in a grasshopper mouse with predatory behavior, compared to the scavenger lab mouse. Here I find the same functional properties. In the second section, I focus on the integration of ocular inputs in V1 of mouse and cat. I first compare disparity selectivity in cats, where convergence of ocular inputs has long been established, with mice, where ocular integration had not previously been investigated. Similar to cats, mouse V1 neurons were sensitive to binocular disparity, albeit to a lesser degree, and could be described by a linear feed-forward model. I next explore the disruption of binocular disparity tuning in both animals. In cats, strabismus induced during development causes increased monocularity in V1 and a loss of disparity selectivity. In mice, monocular deprivation causes increased ocular input, which also manifests as decreased disparity selectivity. Finally, I explore how excitatory and inhibitory neurons in mouse V1 integrate binocular signals. Paravalbumin-expressing inhibitory interneurons are more binocular but less disparity tuned than surrounding cortical neurons, providing a canonical mechanism explaining loss of disparity selectivity in both carnivores and rodents.
Electron temperature gradient mode streamers and the end of the tokamak pedestal
(2020-05-13) Blackmon, Austin Lee; Hazeltine, R. D. (Richard D.)
Fusion experiments still show discrepancies in heat flux from expected results calculated from theory and computation; calculations come up short. As needed, work continues in solving the source of these discrepancies. This thesis seeks to provide part of the answer via electron-temperature-gradient (ETG) fluctuations in the pedestal region of tokamaks, investigated with the Gyrokinetic Electromagnetic Numerical Experiment (GENE). While ETG turbulence is sometimes ignored, as it is very small scale, there may be mechanisms that make it a viable explanation of these discrepancies. In this work, we find a non-negligible heat transport from ETG turbulence in the presence of streamers. Additionally, we find a relation between flux, streamers, and velocity shear. Further analysis is required to confirm these findings, but as it stands, this thesis lays the groundwork
Quantifying the fidelity of a novel methodology for in-core experiment prototyping at the advanced test reactor
(2011-12) Parks, Brian David; Schneider, Erich A.; Deinert, Mark
We have recently developed and tested a new computational method for experiment prototyping at the Advanced Test Reactor (ATR). The method significantly reduces neutronic computation time while maintaining computational accuracy. In this thesis, we present the method and describe the techniques that we used to implement it. We then qualitatively and quantitatively analyze its performance for absorptive and multiplicative experiment perturbations over a single region and across multiple regions of the ATR. We conclude with a discussion of future research that might be conducted on the method.