First Place - Xueying Zhao
Chemical Engineering (PhD, 3rd year)
Chemotactic Response of Marine Bacteria to Hydrocarbons: Experimental and Theoretical Analysis
Microorganisms played a significant role in the degradation of hydrocarbons released into the Gulf of Mexico following the Deepwater Horizon blowout in 2010. Interestingly, some motile marine bacteria (e.g. Halomonas sp.) exhibit chemotaxis toward these hydrocarbons, i.e., they swim preferentially toward chemicals, which they perceive to be beneficial for their survival either as energy or food source. After the Deepwater Horizon oil spill, it was reported that the bacterial density in a plume of dispersed oil was two orders of magnitude greater than the background level surrounding the plume. Processes that concentrate bacteria in the vicinity of dispersed oil affect the biodegradation of the hydrocarbons, which are critical to oil spill cleanup and environmental restoration. The main purpose of our work is to quantify the effect of bacteria chemotaxis to hydrocarbons degradation. To achieve this goal, we exposed a Halomonas strain to hexadecane in a microfluidic device uniquely designed such that the migration of bacteria is solely in response to a linear gradient of hexadecane. This device allows us to readily quantify chemotaxis from the resulting bacterial distribution. Experimental data was used to determine parameters for a theoretical model of bacterial motility and chemotaxis derived from first principles. Specifically, the random motility coefficient was obtained from experiments in a hydrocarbon-free condition, and chemotaxis parameters were determined from experiments over a range of hexadecane concentrations. The parameters we obtained were compared with previously published responses of Pseudomonas putida to the hydrocarbon toluene. As expected, the faster swimming speed of the Halomonas strain allowed its density to reach a steady state in the device in less time than P. putida. This research therefore works toward enabling us to understand the extent that bacterial chemotactic processes can recycle dissolved oil compounds. This can then aid decisions on whether to use other techniques (such as dispersants) to treat oil spills in the future.
Second Place- Vaibhav Verma
Electrical Engineering (PhD, 3rd year)
Energy-efficient deep neural network accelerator with dynamically reconfigurable hardware and processing-in-memory architectures
Deep neural network (DNN) models have surpassed human accuracy in many areas ranging from image classification and face detection to playing a game of Go. But traditional CPU architectures prove inefficient to process compute-intensive DNN workloads. Hence, many specialized hardware accelerators have been proposed but there are still huge gaps in the performance and energy requirements between state-of-the-art software implementations and hardware solutions. To bridge these gaps, we present a novel accelerator designed to maximize the performance and energy-efficiency for a wide variety of DNN workloads. Convolution and fully-connected layers dominate the computation time and resources for all DNN models. However, these different DNN layers exhibit significant heterogeneity in terms of size and computational intensity. Consequently, static accelerators with fixed hardware designed for average case leave massive energy and performance margins unclaimed. Our proposed architecture reconfigures the hardware dynamically during execution of each DNN layer to optimize the metric of interest and achieve pareto-optimal performance. Preliminary results with a 2D systolic array show 15-65% performance improvement (at iso-power) and 25-90% energy improvement (at iso-latency) over the best static configuration for six mainstream DNN workloads. But systolic array architectures which emphasize on data-reuse to improve performance and power are well-suited only for convolution layers. Fully-connected (FC) layers do not reuse any weights and are bottlenecked by huge off-chip memory traffic. Thus, we further optimize the architecture by processing FC layers in-memory to reduce high-cost memory transactions. To the best of our knowledge, this is the first DNN accelerator to implement dynamic reconfiguration in hardware for convolution layers and processing-in-memory for FC layers to achieve best-in-class performance and energy-efficiency. The applications of this accelerator range from super-fast image recognition in battery-constraint low-power environments likes drones to machine-learning applications like speech recognition in IoT devices.
Third Place- Wenjian Jia
Civil & Environmental Engineering (PhD, 2nd year)
Would You Consider a “Green” Vehicle? Investigating Virginian’s Preferences for Electric Vehicles
Electric vehicles (EVs), including plug-in hybrid electric vehicles (PHEVs) and battery electric vehicles (BEVs), have the potential to decrease greenhouse gas emissions, improve air quality, and increase energy security. While EV sales are on the rise, they still account for a small share of all vehicles sold in the U.S. To achieve significant environmental benefits of EV adoption, a larger EV market share is necessary. Thus, it is important to study the influential factors of EV adoption among mainstream consumers. This study presents results of a stated preference (SP) vehicle choice survey across 957 Virginia drivers in 2018. We use choice experiments wherein respondents are asked to choose their most preferred vehicle from four alternatives——internal combustion engine vehicle (ICEV), hybrid electric vehicle (HEV), PHEV, and BEV——under various hypothetical scenarios. Multinomial logit discrete choice models are developed to uncover impacts of socio-economic characteristics, vehicle traits, charging infrastructure, and incentive policies on EV purchase choices. Model results indicate that male, young and middle aged, and college-educated respondents seeking a small car (compared to a mid/full sized car, SUV or Pickup) as an additional vehicle or the first vehicle in their household (compared to replacing an existing vehicle) are more likely to show interest in EVs. Model results also reveal that respondents in the high-income households are less sensitive to purchase price, and respondents with a graduate or professional degree prefer vehicles with lower tailpipe emissions. Regarding charging infrastructure, model results suggest direct current fast charging (DCFC) stations and workplace charging stations increase the utility of BEVs significantly, while respondents without garages at their residences place a significant value on the availability of local public charging stations (e.g., located at shopping center). Results of this study can provide policy and market insights to encourage EV adoption: 1) federal tax credits and state rebates together are critical to offset the high purchase price premiums of EVs; 2) different types of charging infrastructure should be deployed to overcome the range anxiety associated with BEVs; and 3) discriminating marketing strategies should be developed considering the heterogeneous preferences for EVs among population segments.
Benjamin Bowes - Civil & Environmental Engineering (PhD, 3rd year)
Qiyuan Lin - Materials Science (PhD, 3rd year)
William Blades - Materials Science (PhD, 3rd year)
Yuanyuan Ji - Chemical Engineering (PhD, 5th year)
Thanks to our 2019 judges!
- George Prpich
- Rachel Letteri
- Steven Allen
- Geoff Geise
- Amir Chamaani
- Archie Holmes
- Chloe Dedic
- Ahmed Rasin