The National Center for Women & Information Technology (NCWIT) Selects Recipients of the 2020 NCWIT Collegiate Award

NCWIT is pleased to announce winners and honorable mentions of the 2020 NCWIT Collegiate Award, celebrating undergraduate and graduate students who self-identify as women, genderqueer, or non-binary from academic institutions nationwide. Conferred annually, the NCWIT Collegiate Award recognizes technical contributions to projects that demonstrate a high level of innovation and potential impact. 

View a complete list of the 2020 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

• Tiffanie Edwards, Southern Connecticut State University, Deep Learning for Score-based Serial Fusion for User Verification
  This research study proposes to determine how well a deep learning model performs with serial fusion at the score level in biometric user verification. This multimodal classification analyzes finger, facial, and palm images drawn from a publicly available dataset. User verification scores are generated from each biometric trait sequentially and also fused sequentially. Biometric fusion can be used to enhance incomplete data, improve the ability to handle large databases, or improve user protection. (View the project online.) 

• Fatemehsadat Mireshghallah, University of California - San Diego, Shredder: Learning Noise Distributions to Protect Inference Privacy
  The Shredder project uses a self-supervised learning approach to tackle the problem of privacy during inference for classification tasks, especially for edge devices that execute part of the task on the device and part of it on the cloud. To prevent attackers from extracting private information from raw data that is used to perform a task, Shredder learns the noise distribution patterns that are tolerated by the network and do not harm the accuracy of the task completion, but do decrease the availability of private information in the input. (View the project online.)

• Kelley Paskov, Stanford University, Inherited Deletions Contribute to Autism Risk
  This project uses genomic data from children with autism and their families to detect DNA deletions and to examine their relationship with autism. While small deletions are difficult to detect via genome sequencing, this project uses a hidden Markov model to detect deletions that other methods would have missed. Formulating biological knowledge as mathematical constraints on the model was a key innovation that made it computationally feasible to scale the model to the required scope. (View the project online.)

• Naba Rizvi, University of Toledo, Hera
  According to the American Heart Association, women are seven times more likely than men to have their heart disease misdiagnosed. To address this gender inequality in access to healthcare, this project created an application that uses deep learning to help women advocate for themselves to get fairer and faster heart disease diagnosis by helping them understand what an abnormal electrocardiogram (ECG) result looks like. The user uploads a photo of their ECG results, and the application provides a preliminary classification. (View the project online.)

• Eshika Saxena, Harvard University, HemaCam: A Computer Vision-Enhanced Mobile Phone Imaging System for Automated Screening of Hematological Diseases with Convolutional Neural Networks
  Early diagnosis is key to treating many diseases that affect the shape, size, and/or appearance of red blood cells. This research focused on making hematological disease screening fast, affordable, and automated by designing a low-cost, smartphone-based microscope to capture blood cell images; developing computer vision algorithms to enhance, segment, and characterize the blood cell images; and creating and training deep learning models to provide instant disease diagnosis. (View the project online.)

• Ariana Isabel Sokolov, University of Southern California, Trill Project
  Trill Project is an anonymous social network where individuals can freely express themselves in order to feel a sense of belonging. Inspired by a desire to address mental health challenges faced by LGBTQ+ teens, Trill Project has also become a refuge for users suffering from trauma, abuse, and other mental illnesses. The platform uses a combination of machine learning, human moderation, and other features to deter trolls and ensure that the community remains safe, positive, and supportive. (View the project online.)

Honorable Mentions

• Michelle Bao, Stanford University, Enhancing Real-Time Driver Identification Checks Using Liveness Detection for the Uber Driver App
  This project addresses the problem of malicious individuals impersonating legitimate Uber drivers by bypassing the selfie-based driver authentication methods that allow access to the driver app. By combining the capabilities of two native Apple frameworks that are built in with all iOS apps, the researcher was able to develop a “liveness detection” feature that requires drivers to perform a series of randomized requests to move their face in order to verify their identity. (View the project online.)

• Kelly Finke, Swarthmore College, Ancestral Haplotype Reconstruction for Identifying Recessive Risk Factors
  This project created a tool called thread (Timely Haplotype REconstruction of Ancestors based on Descendants), which expands the generational scope of genetic data available for studies of endogamous populations and thus supports medical research on genetic diseases. Given existing sequence data and a pedigree, thread is able to reconstruct estimates of ancestral chromosomal sequences, called haplotypes, for up to six generations prior to the living generation, enabling researchers to gain insight into long-term inheritance patterns of human diseases. (View the project online.)

• Grace Guan, Princeton University, Predicting Sick Patient Volume in a Pediatric Outpatient Setting Using Time Series Analysis
  This project aimed to improve the methods used to predict patient volume in a pediatric clinic in order to help providers better allocate staff and resources to meet patient needs. After investigating several time series models, the researchers found that recurrent neural network (RNN) models were able to capture the seasonality of the data and perform substantially better than state-of-the-art models, including constant predictions. (View the project online.)

• Israa Jaradat, University of Texas - Arlington, Excavator: Truth Diggers Never Get Lost
  The goal of this study was to develop an automatic research tool, named Excavator, which is capable of thoroughly investigating the veracity of a news claim and providing the user with a summary of the most important findings, in a way that mimics the manual fact-checking process currently used by professional media outlets. Excavator’s comprehensive approach includes such functions as veracity analysis, evidence extraction, source detection, summary generation and presentation, and pattern discovery. (View the project online.) 

• Mackenzie Jorgensen, Villanova University, Automated Machine Learning for Multi-Class Classification of Hate Speech on Twitter
  Due to a lack of computing resources, small online media platforms often rely on manual systems to combat hate speech on their sites. This project aimed to create a semi-automated hate speech detection system which is easy to use for non-machine learning experts. Auto-ML, which completes the machine learning algorithm selection and hyperparameter optimization processes for the researcher, provides a solution that is more accessible for small media outlets. (View the project online.) 

• Edden Kashi, Hofstra University, Investigating Techniques of User Tracking Through Mobile- and Web-Applications
  This project explored a popular technology for online user tracking, the web- and Bluetooth- beacons, in order to understand how much information about users can be gathered without their knowledge. It also investigated whether and how this tracking can be limited. A secondary focus of this project was to investigate the information that is being sent out by mobile devices while they are not being used. (View the project online.)

• Ruby Martinez Gomez, University of Colorado - Boulder, D.O.T.T.S. (Debris Orbital Tumbler and Thermal Sensor)
  The research team for this project worked with the national RockSat-X program to design a set of experiments dealing with problems related to “space junk.” The primary experiment tested the feasibility of using a static electric field to meaningfully impact the trajectory of simulated space debris fragments, which currently pose a danger to both astronauts and equipment. The secondary experiment gathered data on the viability of 3D-printed materials for use in space. (View the project online.)

• Ruth Okoilu Akintunde, North Carolina State University, Let's Watch Math Game Become Easier By ReOrdering Curriculum Sequence
  This project conducted a novel predictive analysis of a curriculum-integrated math game, ST Math, to suggest a partial ordering for the game’s curriculum sequence. In order to determine the best sequence of objectives within the game for maximum student success, the researchers replicated a prior study to find predictive relationships between 15 objectives played in different sequences by 3,328 students from five districts. (View the project online.)

• Sofia Ongele, Fordham University, ReDawn
  ReDawn is an app that addresses the lack of resources and information available to survivors of sexual violence. This project aimed to develop a platform where users can access confidential advice in a conversational matter. The app’s primary functionality is a chatbot named Dawn, and it also includes a tab that connects people to local resources, a tab to log past incidents of abuse, and a tab dedicated to hotlines and 911. (View the project online.)

• Isha Puri, Harvard University, A Scalable and Freely Accessible Machine Learning Based Application for the Early Detection of Dyslexia
  The goal of this project was to develop a free, web-based application that uses a standard, inbuilt computer webcam to screen children for dyslexia. Several medical studies have shown that dyslexic students exhibit significantly longer and more frequent eye fixations while reading than non-dyslexic readers. By implementing a novel combination of different machine learning algorithms, this project was able to produce a highly accurate eye-tracking methodology for the standard computer webcam. (View the project online.)

• Faith Williams, Columbia University, Quantum Entanglement in Medical Diagnosis
  This project studied the application of physical optics and quantum entanglement in medical diagnosis to better understand diseases and improve disease detection. The research aimed to discover whether one could distinguish human brain tissue from subjects with and without Alzheimer's disease using entangled photon pairs. The project used quantum topography to analyze the scattering of photons that came into contact with healthy and diseased brain tissue samples, as well as the change in entanglement between the photon pairs. (View the project online.)

• Hairuo Xu, Auburn University, Toward the Verifiability Problem in Distributed Machine Learning: Attack Models and Detection Algorithms
  Distributed machine learning systems have drawn a lot of attention recently; however, some security issues have been largely overlooked. This project investigates this problem from both the negative (i.e., attack) and positive (detection) standpoints. The project developed multiple threat and detection models, including the Random Noise Attack, the Gradient Ascent Attack, and the Gaussian and Cosine Similarity Detection models, to allow a system’s users to improve their verification effectiveness under various system settings. (View the project online.)