Spring 2019 Research Incubation Award Recipients

About the Research Incubation Awards (RIA)

The Institute supports a portfolio of ambitious research projects and activities that align with its mission to catalyze new directions and collaborations in interdisciplinary research that applies the computational 3and data-driven approaches. By supporting projects in fields from social sciences, humanities, law, business, and medicine – to engineering, computer science, and statistics, the Research Incubation Awards advance the Institute’s goal of bringing together and building a community of researchers across schools and colleges at BU. In addition to the main RIA program, the Hariri Institute for Computing is happy to offer four special tracks for projects, including the Artificial Intelligence Research (AIR) Initiative, the Digital Health Initiative (DHI), Red Hat Collaboratory and the Urban Climate Initiative (UCI).

Pre-Cancer AutoHist: A Software Application to Classify Lung Pre-Cancer Histologic Features

PI: Jennifer Beane, Assistant Professor of Medicine, MED
Track: Digital Health Initiative

Lung cancer is the leading cause of cancer death, and the pre-cancer lesion is an important clinical indicator of lung cancer risk to help therapies intercept the cancer development process in the early stage to increase survival. In this case, the team proposes a software application to accurately and automatically identify histologic features of lung pre-cancerous lesions based on digitized whole-slide images. The Human in the Loop (HITL) machine will be used to automate the analyze and generate user-friendly web interface-based application (App).

Novel Greenness Metrics Using Remote Sensing and Street Imagery for Climate Change Mitigation and Environmental Health Research

PI: M. Patricia Fabian, Research Assistant Professor of Environment Health, SPH
Co-PI: Lucy Hutyra, Associate Professor of Earth and Environment, CAS
Track: HIC & Urban Climate Initiative

Urban green space can impact communities by mitigating the effects of heatwaves, reducing the rates of urban runoff, and moderating heat islands. Exposure to green space will also lead to various types of diseases. However, little research has been done to leverage the extensive knowledge and computational advances from other fields to characterize greenness. The team will construct novel metrics of greenness exposure to be used in climate and health studies by using high-resolution remote sensing, urban tree biomass, and processed street view image databases and to yield a comprehensive database that can be used in the future studies of climate, health, and environmental health. The team is planning on publishing at least one manuscript of the work.

Accelerating Multi-Party Computation with FPGAs

PI: Martin Herbordt, Professor of Electrical and Computer Engineering, ENG
Co-PI: Mayank Varia, Research Associate Professor of Computer Science, CAS
Track: HIC & Red Hat Collaboratory

Datasets can tell a great story and provide critical insights in almost every aspect of society. Unfortunately, guaranteeing privacy can slow this process, and critical insights otherwise unavailable can be gained by sharing data. The team aims to accelerate these secure computations without losing any guarantees of privacy and enabling a platform for future private data set sharing. FPGAs can provide several orders-of-magnitude improvements and the FPGAs are becoming ubiquitous in data centers and can be leveraged to accelerate a diverse set of both internal and 3rd party workloads.

 “Can I get that in Writing?” Lessons from a Contracting Field Experiment in Urban Malawi

PI: Mahesh Karra, Assistant Professor of Global Development Policy, CAS
Track: HIC

In some low- and middle- income countries, transactions between the buyers and the sellers are often conducted informally and are based on credibility. Formalizing contracts can be important and beneficial. The team wants to measure the potential advantages of formalizing contracts. By conducting a field experiment in Lilongwe, Malawi, the team wants to test the extent to which formal contracts improve performance and accountability and crowd out informal contracts.

Predicting Individualized Rehabilitation Outcomes based on Brain and Behavioral Biomarkers

PI: Swathi Kiran, Associate Dean, SAR
Co-PI: Margrit Betke, Professor of Computer Sciences, CAS
Track: Digital Health Initiative & Artificial Intelligence Research

Over 2 million people are currently living with aphasia due to a stroke, and the patients do not know whether they will ever recover to improve their communication/cognitive skills. Aphasia therapy is prescribed largely on clinical presentation but the information about brain or behavior that may impact recovery is not taken into account. This project aims to develop machine-learning algorithms that take as input a complex set of brain and behavioral markers to classify and predict responders and non-responders to therapy. This study takes an important first step to predict individualized rehabilitation outcomes. All the data is already collected and the proposed project is ready to be implemented.

Adversarial Learning of MRI data for improved Assessment of Alzheimer’s Disease

PI: Vijaya B. Kolachalama, Assistant Professor of Medicine, MED
Co-PI: Rhoda Au, Professor of Anatomy & Neurobiology, MED
Track: Digital Health Initiative & Artificial Intelligence Research

Alzheimer’s disease (AD) is a disease that will require a high cost of care, and the total amount keeps rising. Identifying risk factors that are amenable to intervention at the preclinical stage represents the most effective method for combatting AD. The team wants to apply an adversarial learning framework to enhance contrast to noise ratio (CNR).

Revolutionizing Flow, Heat and Dispersion Predictions over Complex Urban Environments

PI: Dan Li, Assistant Professor of Earth and Environment, CAS
Co-PI: Kia Teymourian, Assistant Professor of Computer Science, MET
Track: Urban Climate Initiative

As more and more of the population lives in the city, it is more and more challenging to understand and make predictions of heat transfer and dispersion of air pollutants over urban environments. The team will use and develop the machine learning models to execute the project. It is less computationally expensive and much faster, and may better capture the small-scale variabilities. The new model could help to identify key locations for sensor deployment. The investigators aim to submit full-blown proposals to the Army Research Office at the Department of Defense and/or the Environmental Sustainability Program at the National Science Foundation.

Multi-faceted Data Analysis of Blazars, Highly Time-Variable Cosmic Objects

PI: Alan Marscher, Professor OF Astronomy, CAS
Co-PIs: Svetlana Jorstad, Senior Research Scientist of Institute for Astrophysical Research; Manasvita Joshi, Research Assistant of Institute for Astrophysical Research
Track: HIC

The team proposed to develop procedures to cross-compare astronomical time series data, obtained at different wavelengths, of highly variable cosmic objects called “blazars”. The analysis will require new techniques that can correlate, in an objective manner, time variations of brightness at both visible and gamma-ray wavelengths, and linear polarization at optical wavelengths. In this case, the team seeks to develop machine-learning algorithms that can identify both random components and underlying patterns.

Hiding the State: Outsourcing Security to Avoid Transparency and Accountability

PI: Kaija Schilde, Assistant Professor of International Relations, CAS
Track: HIC

Researchers, journalists, and citizens have used the Freedom of Information (FOIA) Requests since 1965 as a proxy for transparency and democratic oversight. The researcher hypothesized that increases in transparency requests in a given area will produce increased government outsourcing in that area to shield themselves from democratic accountability. This project will help the society to understand these processes and could lead to pressures to better measure, report and regulate the processes of government outsourcing and contractor accountability.

Using Reinforcement Learning to Rationally Reprogram Multicellular Decision Making

PI: Allyson Sgro, Assistant Professor of Biomedical Engineering, ENG
Co-PI: Pankaj Mehta, Associate Professor of Physics, CAS
Track: HIC & Artificial Intelligence Research

While standard modeling techniques can describe and predict biological behaviors, they fall short of offering strategies for manipulation and reprogramming. The team hypothesizes that single cells have algorithms for interpreting their environment and that they can reprogram these algorithms with Reinforcement Learning (RL) techniques. It would be an experimentally-informed hybrid-automata-continuum model of multicellular behavior. The eventual goal is to make an open-loop controller directly capable of learning to control cellular behaviors in real-time during experiments.

Confounder Matrix Online Tool

PI: Ludovic Trinquart, Assistant Professor of Biostatistics, SPH
Track: HIC & Digital Health Initiative

Systematic reviews (SRs) are appraisals and syntheses of original studies to answer some clinical questions. It is important because they help to inform clinicians and researchers on patient-important issues. However, the validity of the SRs of observational studies is the potential bias due to confounding, and a framework is lacking to address confounding bias. The investigator and the team developed a novel approach to examine confounding in SRs of observational studies and want to develop an online open-access tool for confounder bias analysis in SR of observational studies.

Integrated Foundations for Privacy-Protected Public Opinion Research

PI: Mayank Varia, Research Associate Professor of Computer Science, CAS
Co-PIs: Dino Christenson, Associate Professor of Political Science, CAS, and Mark Crovella, Professor of Computer Sciences, CAS
Track: HIC & Red Hat Collaboratory

Understanding public opinion dynamics is important but difficult because the polls are widely used with significant drawbacks. The team proposes to design, develop and deploy a novel architecture that passively observes the web browsing activity of a user panel, and interprets that activity through the aid of novel machine learning algorithms. The platform would allow social scientists to ask detailed questions and receive responses rapidly, enable policymakers to understand and react to the will of the public and better understand important social problems.