Request for Concepts 2018

The Green Light SONATA—Demonstrating Human Artistic Capabilities in Data and Decision Science and Engineering

 

  • Monty Abbas, Ph.D., P.E., Professor, Civil Engineering
  • Charles Nichols, Ph.D., Assistant Professor, School of Performing Arts
  • Anne Elise Thomas, Ph.D., Research Fellow at Moss Arts Center

 

Abstract
Because of the difficulty/infeasibility of optimizing traffic operation with traditional methods ($160 billion of wasted time and fuel in 2014 alone), several researchers turned to heuristic methods (e.g., supervised learning) for the solution. However, supervised learning needs known optimal outputs corresponding to each input in the training set, whereas the optimality of a given control policy in traffic operation is only known at the end of the analysis period.

To fill this gap, our research team conducted preliminary research to investigate a novel idea (the Green Light SONATA) to translate traffic data into sounds that can be heard by musicians, and translate musicians’ real-time improvisation response to signal indications that control the traffic. It was very interesting to find that two of the three team members have outperformed the Webster’s optimal control (by 22.3% and 17.6%, respectively).

In this work, we propose three major activities: (1) design a demonstration of the Green Light SONATA that can draw public attention to VT Data and Decision SGA capabilities, (2) showcase the system by controlling staged pedestrians flows on the Drill Field with musically controlled temporary traffic lights, and (3) collect and analyze data from the showcase for further NSF proposal development.

 

 

Deep Learning Infrastructure for Precision Medicine with Multi-Omics Data

 

  • Zhang, Liqing, Associate Professor, Department of Computer Science
  • Bert Huang, Assistant Professor, Department of Computer Science
  • Zhi Sheng, Assistant Professor, Virginia Tech Carilion Research Institute and Virginia Tech Carilion School of Medicine

 

Abstract
The core motivation of this research is to provide computational infrastructure to accelerate precision medicine. Omics data today, from genomics to metagenomics, is becoming a basis enabling medical decisions to be tailored to individual variability in genes as wells as environments. Therefore, it is crucial to develop an advanced infrastructure that fully leverages different omics data with cutting edge machine intelligence to ensure reliable medical predictions and decisions. We propose to develop a deep learning infrastructure that can scale to the high-dimensionality of multi-omics data and enable individualized medical decisions, with the goal of dramatically reducing the cost of deploying deep models in clinical practice. This work can be applied to various medical and biological domains that require data-driven decision models based on multi-omics data. As a case study with the proposed deep learning infrastructure, we will focus on solving problems in cancer treatment, especially, distinguishing responder and non-responder of cancer immunotherapy agents based on patients’ omics data. To successfully achieve this interdisciplinary research, we have established a research team that consists of experts in computational biology, machine learning, and cancer medicine. The funding will be used to garner preliminary results that serve as a proof-of-concept for the upcoming external federal grants.

 

 

Attention and Judgement Aggregation: Theory and Experiments

 

  • Matthew Kovach, Collegiate Assistant Professor, Department of Economics
  • Gerelt Tserenjigmid, Assistant Professor, Department of Economics

 

Abstract
Our goal is to understand the roles of attention and judgement aggregation in decision making through two experiments. The first experiment (attention) studies the role of attention in strategic environments. In particular, we want to understand the role of attention in strategy selection and why certain strategies command more attention than others. The second experiment (aggregation) is about how conclusion vs premise based aggregation rules effect decision making. For example, consider a 3-member appellate court determining liability in a contract dispute. Suppose defendant (D) is liable only if there was a valid contract and D failed to satisfy its requirements. Suppose judge one finds that contract was valid and that D breached, judge two finds that the contract was valid but D did not breach, and judge three finds that D would have been in breach if it were valid but that this contract is not. If we aggregate their conclusions, we find that that majority find D not liable. If we aggregate their premises, we find that a majority of judges find that the contract was valid (judges one and two) and that D was in breach (judges one and three), and thus we conclude that D is liable.