The National Center for Women & Information Technology (NCWIT) Selects Recipients of the 2021 NCWIT Collegiate Award

Author(s): 

NCWIT Collegiate Award Logo

NCWIT is pleased to announce winners and honorable mentions of the 2021 NCWIT Collegiate Award, honoring the outstanding computing accomplishments of undergraduate and graduate students who self-identify as women, genderqueer, or non-binary. Conferred annually, the award recognizes technical contributions to projects that demonstrate a high level of innovation and potential impact.

View a complete list of the 2021 recipients below.

The entire NCWIT AiC program platform is supported generously by Apple. AiC also receives support for specific national program elements; the NCWIT Collegiate Award is sponsored by Qualcomm and Amazon with additional support from Palo Alto Networks.

 

Winners

  • Laura Lewis, California Institute of Technology, Implementing Remote-State Preparation on a Noisy Intermediate-Size Quantum Device
    This project addresses the problem of blind and verifiable delegated quantum computation, in which a user performs a computation on a quantum computer and can be assured that the answer is correct (verifiable) without revealing the computation being performed (blind). The researcher programmed and created quantum circuits for the evaluation of certain functions, known as Noisy Trapdoor Claw-Free (NTCF) functions, which are crucial to the process that allows a user to check an answer that they obtain from the quantum computer. (View this project online.)
  • Isha Puri, Harvard University, A Scalable and Freely Accessible Machine Learning Based Web Application for the Early Detection of Dyslexia
    By implementing a novel combination of different machine learning algorithms, this research was able to produce the first-ever eye-tracking methodology that uses the standard computer webcam to screen for dyslexia. The application was able to predict if a child has a higher risk of dyslexia with an accuracy of 90.18 percent. Because it is completely free and doesn't require any specialized equipment, this application has the potential to make highly accurate dyslexia screening accessible to millions of families around the world. (This project video is currently unavailable to view online.)
  • Rachel Guo, Harvard University, Enhancing Poaching Predictions for Under-Resourced Wildlife Conservation Parks Using Remote Sensing Imagery
    The Protection Assistant for Wildlife Security (PAWS) is a machine learning approach to helping rangers in wildlife refuge areas around the world identify the areas with highest risk of illegal poaching. PAWS uses publicly available remote sensing satellite imagery to extract geospatial features that impact poaching probability. The pipeline inputs these extracted features, historical patrol data, and prediction labels to train on, and it outputs predictions of probability of poaching risk for each cell, resulting in a risk map that rangers can use to patrol areas with highest poaching risk. (View this project online.)
  • Marissa Sumathipala, Harvard University, Network Medicine Platform for Predicting Noncoding RNA Drug Targets
    This project leverages network science and machine learning for in silico prediction of the effectiveness of specific non-coding RNAs for use as therapeutics, bypassing more expensive experimental methods. The researcher created a computational platform called Theraplexus by constructing a heterogeneous network model from massive genomic and transcriptomic datasets that model the molecular mechanisms that go awry in disease. Novel algorithms powered by network science analytics and artificial intelligence are used to predict and prioritize drug targets for any given disease. (View this project online.)
  • Farita Tasnim, Massachusetts Institute of Technology (MIT), Real-Time Decoding of Facial Strains via Conformable Piezoelectric Interfaces
    The aim of this project is to realize conformable sensors and systems that can translate patterns of facial soft tissue biokinematics into interpretable electrical signals to enable new forms of nonverbal communication. The researchers created cFaCES, or the conformable (i.e. stretchable and flexible) Facial Code Extrapolation Sensor, which makes precise measurements of soft tissue biokinematics to recognize distinct facial motions, and thus facilitate nonverbal communication for patients who lack the ability to speak or interact with traditional electronic communication interfaces. (View this project online.)
  • Angelique Taylor, University of California - San Diego, Robot Social Navigation in the Emergency Department
    Robots can be used in hospital emergency departments (EDs) to help with tasks such as stocking and delivering supplies, but to perform these tasks, they must understand the context of complex hospital environments and the people working around them, and they must make intelligent social decisions to avoid causing disruptions to patient care. This project developed a new system called the Safety-Critical Deep Q-Network (SafeDQN), which uses a reinforcement learning agent to enable a robot to navigate in the ED while identifying and avoiding areas in which high-acuity patients are being treated. (View this project online.)

 

Honorable Mentions

  • Alexandra Chin, Wellesley College, Olfactory Communications System
    This project developed a prototype of an end-to-end device that can identify an odorant and replicate that odor across any distance, allowing users to effectively communicate a wide range of smells. The system consists of three sub-blocks: electronic nose (or “e-nose”), communication system, and synthesis. The e-nose detects unique features of a smell using volatile organic compound sensors and creates a signature for that smell. An identification algorithm compares the smell signature to a dynamic plot of past signatures to determine the odorant, then the application sends the information via bluetooth to a synthesis machine. (View this project online.)
  • Kirthi Kumar, University of California - Berkeley, Developing Novel Computational Models for Prescription and Illicit Drug Addiction Dynamics in the COVID-19-Amplified Opioid Epidemic
    Computational modeling for opioid addiction via an infectious disease model is a novel method of approaching the opioid crisis. The typical framework for mathematical models, known as Susceptible-Infected-Recovered (SIR) models, involve the classic compartmental model with ordinary differential equations (ODEs) that describe phases or dynamics of infectious diseases. These models have been widely used for viruses such as malaria and Zika, and most recently, COVID-19. This project explores two mathematical models: the first describing the dynamics of prescription-based addiction, and the second describing the dynamics of addiction influenced by social behavior. (View this project online.)
  • Alexandra Szewc, Johns Hopkins University, Cytoscope: CD4 Count Estimator
    To address the need for more affordable detection, diagnostic, and treatment monitoring options for HIV, this project developed a tool called CytoScope: a low-cost, Raspberry Pi-powered microscope capable of imaging blood smears and calculating CD4 Count estimates for low-income HIV/AIDS patients. The CytoScope is armed with a CellProfiler pipeline that takes a microscope image of a stained blood smear as input, and provides the red and white blood cell counts within the smear as output, allowing for the estimation of a CD4 count. (View this project online.)
  • Victoria Xin, Stanford University, Predicting Antihypertensive Medication Efficacy through Machine Learning Survival Analyses in a Revascularized Patient Population
    Patients who have undergone coronary revascularization procedures often experience hypertension, but the most effective antihypertensive medication for this population of patients is currently unknown. This project aimed to address this gap in medical knowledge by leveraging Random Survival Forests (RSF), a machine learning method, and Kaplan-Meier survival analyses, a statistical technique. Specifically, stroke risk was examined in revascularized patients from the Antihypertensive and Lipid Lowering Treatment to Prevent Heart Attack Trial based on the medications they received: amlodipine, chlorthalidone, or lisinopril. (View this project online.)
  • Brianna Garland, Stevens Institute of Technology, Design of Grail Automated Healthcare Records
    The Grail Automated Healthcare Records project is designed to give patients autonomy over their personal health data, protect confidentiality, and reduce redundancy in the collection of medical information. The researcher used XDL (a data description language) to create an inclusive and frictionless interface in which medical data is received, including new capabilities to ensure user dignity. Once the information is entered into the form, it can then be shared with any doctor. Individuals can also update their information at any time. (View this project online.)
  • Erin Howard, Western Washington University, Leveraging Statistical Analysis to Develop Labels for Astronomical Time Series Data
    Developing effective machine learning training labels for classifying new time series data can be a daunting task, particularly for a classification that has many representations. This project shows how statistical analysis (Lomb-Scargle, Autocorrelation, Box Least Squares, and z-score) can be used to pre-classify a portion of the data and lessen the burden of manual classification. Once these statistical patterns are found in a smaller sample data set that has already been manually classified, a statistical model can be developed and applied to a more expansive portion of the data set in order to provide classification labels. (View this project online.)
  • Evani Radiya-Dixit, Stanford University, Learning Efficient Natural Language Processing Models
    In computational research, strong deep learning performance has been achieved with increasingly large models. The goal of this project is to reduce the high memory and compute needs of NLP models by developing a training procedure that sparsifies models by reducing the number of parameters used. The researcher developed two key techniques to share parameters between the pre-trained and fine-tuned models: learning fewer parameters during fine-tuning, and learning sparsity during fine-tuning. Together, these techniques save significant memory and computational cost in large language models without notable performance degradation. (View this project online.)
  • Maya Venkatraman, Columbia University, Adding YouTube HLS Output to OBS (Open Broadcaster Software)
    This project, adding the protocol known as HLS (HTTP Live Streaming) as a method of streaming video to YouTube from OBS (Open Broadcaster Software), began as a means to solve an important problem in the field of video compression: the tradeoff between video quality and buffering. The project involved adding a new option in OBS for streaming HLS to YouTube. The researcher implemented this output in C/C++ and wrote HLS-specific methods, ensuring that the appropriate HTTP requests got sent and that video and audio packets got muxed (i.e. combined) together as needed for transmission. (View this project online.)
  • Hannaneh Barahouei Pasandi, Virginia Commonwealth University, A Learning-Based Framework for Self-Driving Design of Networking Protocols
    Machine learning (ML)-based solutions for communication protocol design can reduce manual efforts to tune individual protocol parameters. However, these solutions are hard to interpret due to the black-box nature of the ML techniques. While other proposed ML-based methods mainly focus on tuning individual protocol parameters, the main contribution of this project is to decouple a protocol into a set of parametric modules, with each module representing a main protocol functionality and used as Deep Reinforcement Learning (DRL) input in order to better understand the generated protocols’ design and analyze them in a systematic fashion. (View this project online.)
  • Isabel Cachola, Johns Hopkins University, TLDR: Extreme Summarization of Scientific Documents
    This project aimed to use machine learning to automatically generate TLDR (“Too Long; Didn’t Read”) statements, or single-sentence, extreme summaries of scientific documents that are commonly used on social media and review sites. Because there was no existing publicly available dataset available for training, the researchers curated a dataset of 5.4K TLDRs. They also introduced a new training technique that improved the generated statements in both automatic metrics and in human evaluation by exploiting the similarity between titles and TLDRs. (View this project online.)
  • Israa Qasem Jaradat, University of Texas - Arlington, Automatic Cherry-picking Detection and Correction in Textual Data
    Cherry-picking refers to the practice of building up an argument by omitting evidence against the case and exaggerating evidence that supports the case. To address this problem, the researcher developed Cherry, the first-ever end-to-end automatic tool that tackles cherry-picking detection and correction in textual data. Cherry detects cherry-picking in a given news article by analyzing the context of the event covered by the article using stories from alternative sources. Stories are then segmented, and the importance of a segment with respect to the event is the basis for determining whether it is cherry-picked or not. (View this project online.)
  • Lexi Rindone, Johns Hopkins University, High-Resolution Quantitative 3D Imaging of Blood Vessels and Bone Cells in the Skull
    Emerging technologies focused on regenerating bone through a combination of biomaterials, biologics, and stem cells could provide groundbreaking treatments for patients suffering from craniofacial injuries. However, translating these technologies to clinical practice has been challenging due to a lack of fundamental understanding about how key cellular players interact to grow new bone. This project focuses on identifying the necessary environment to induce vascularized craniofacial bone regeneration by using a novel imaging platform that couples advanced molecular biology techniques with 3D quantitative image analysis to characterize cellular relationships within the skull. (View this project online.)

 

 

Aspirations Community: 
National Award

Add comment

Log in or register to post comments