Advancements in 2D and 3D imaging have led to their widespread consideration as tools for the non-destructive evaluation (NDE) of engineered components. Modern imaging can capture sub-millimeter flaws and defects in materials, but the resulting scans do not provide inherent information on how captured flaws are impacting mechanical performance, limiting their value. By leveraging a combination of computer vision and machine learning techniques, this challenge can be overcome, enabling the expanded use of 3D images in a broad range of engineering disciplines. This seminar will focus on two current avenues of work in this domain. The first is the use of machine learning to establish a statistical relationship between images and mechanical performance,with applications in post-disaster assessment. The second avenue explores how computer vision techniques can be used to model 3D image information in a time-variant context in order to allow for the dynamic analysis of image information.]]>
Modeling the bio-mechanical behavior of soft-tissues in their service configuration is often challenging because of their complicated geometry, material heterogeneity, nonlinear behavior under finite strains and the associated fluid-structure interaction problem. Efficient solutions to such complex coupled biological processes are still a challenging problem in computational sciences. Direct numerical simulation of the associated non-linear equations, governing even the most simplified model depends on the convergence of iterative solvers which in turn rely heavily on the properties of the coupled system. In this talk, we will review mathematical and experimental approaches for quantifying the multiaxial mechanical properties of hyperelastic membranes interacting with fluid dynamics. Some numerical findings on the dynamic stability and the influence of contact constraints on these lesions will also be presented.
We will also describe some novel research, training and education programs that this multidisciplinary research has helped to enhance ongoing interaction among communities of people, including students and faculty. Participants will also have the opportunity to learn about new and ongoing funding programs that can be catalysts to help reinforce and drive reform across an institution.
Trust is a key component that shapes inter-personal relationship and is known to vary with social contexts. Previous evidence has shown the power of ascribed identity (e.g., ethnicity, gender) upon trust behaviors of human beings. However, few studies have investigated the neural mechanisms underlying how acquired identity (e.g., political party) may influence one’s trust-related decision making. To address this issue, we enrolled 58 healthy adults who share different political identities, defined by their presidential choices in 2012 Taiwan presidential election (i.e., KMT vs. DPP supporters), to participate in a repeated binary trust game experiment while undergoing fMRI scan. For each trial of the game, participants (investor) could choose to invest (“trust”) their partner or not (“keep”) in the first round and their partner would reply with either a “reciprocate” or “defect” feedback decision. Participants were informed that they would play the game with partners with the same, a different or no political identity. At the behavioral level, participants showed significantly higher probability of trust decisions when a partner shared the same political identity, suggesting that political identities indeed modulate their cooperative decisions. At the neural level, we found that identities defined by political preferences have different neurophysiological effects on decision outcomes. When playing with partners with the same political identity, functional contrasts between trials in which a partner defected participants’ trust and trials in which a partner reciprocated participants’ trust showed significant hemodynamic signal changes in brain regions including anterior insula (emotional processing), the temporoparietal junction (mentalizing), and the dorsolateral prefrontal cortex (self-regulatory control and/or working memory). In contrast, when playing with partners with a different political identity, participants exhibited greater activation in the striatum (reward learning) in response to trials in which a partner reciprocated as compared with trials in which a partner defected. More interestingly, increased activation in the anterior insula significantly correlated to closer perceived social distances between participants and their partners. In summary, these findings provide the first evidence on the neural foundations for the modulation effects of political identities upon trust behaviors, and indicate that studies of decision making should account for the role of social identity in altering behavior and brain response.]]>
Heterogeneity of neural attributes has recently gained a lot of attention and is increasing recognized as a crucial feature in neural processing. Recent experimental recordings in electric fish indicate that the heterogeneous network input can mediate response heterogeneity of superficial pyramidal cells in a cortical area (Marsat Lab, WVU). These data motivated us to theoretically study how heterogeneity of neural attributes in general alter firing rate heterogeneity. We ask how 2 sources of heterogeneity: network (synaptic heterogeneity) and intrinsic heterogeneity alter response heterogeneity.
First we address this in a canonical recurrent spiking network model with random connectivity (Erdos-Renyi graph). The relationship between intrinsic and network heterogeneity can lead to amplification or attenuation of firing rate heterogeneity, and these effects depend on whether the recurrent network is firing asynchronously or rhythmically. We analyze the system and derive compact analytic formulas to precisely describe the phenomena.
Second, we adapt our theory to a delayed feedforward neural network to better model the electric fish system. The theory is used to demonstrate that a feedforward network with structured connectivity rules exhibit qualitatively similar statistics as the experimental data. Specifically, the stimulus tuning of particular cells is related to the network architecture, i.e., the number of synaptic connections. Thus, the model demonstrates that intrinsic and network attributes do not interact in a linear manner but rather in a complex stimulus-dependent fashion to increase or decrease response heterogeneity and thus shape population codes.
This is joint work with Gary Marsat (West Virginia University).]]>