Computer Science Research Projects at the 8th Symposium of Rising Scholars
4 minute read
Recently, over 160 Polygence students had the opportunity to showcase their work at Polygence’s 8th Symposium for Rising Scholars. In this article, I will highlight projects related to computer science and mechanical engineering. Computer science projects were prevalent throughout the Symposium, and many of the projects I discussed in my other articles also involved computer science. (After you’re done with this article, you can learn about how students used computer science to predict breast cancer and explore genetic variations in dolphins.) Keep reading to learn more about some incredible computer science projects!
Create a research project tailored to your interests and your schedule
Polygence pairs you with an expert mentor in your area of passion. Together, you work to create a high quality research project that is uniquely your own. We also offer options to explore multiple topics, or to showcase your final product!
Research Projects with Machine Learning/Artificial Intelligence
Xinyuan Duan discussed her novel approach to addressing natural disasters in her talk, “Optimizing Disaster Impact Detection Pipelines.” Natural disasters are occurring increasingly often, causing widespread damage to infrastructure, so computer scientists have developed models that predict how much damage has been done to buildings as a result of a natural disaster. Xinyuan aimed to reduce the complexity of current machine learning models so that the models predicted the damage more quickly. This faster prediction would in turn allow rescue personnel to know which buildings need the most attention and act more quickly.
First, Xinyuan collected data from an online source that included images of buildings in various states of damage. There was a pre- and post- disaster version of each image. Every image also came with a classification number indicating the extent of the damage. Previous approaches for predicting a building’s damage classified every pixel of the image, so Xinyuan’s approach was to separate the image into boxes, and then classify each box instead. She thought of this approach because buildings are generally shaped like a box. After cleaning the data, she developed a convolutional neural network with five layers that used this box detection approach. While the model accuracy was lower than expected, there is still potential for this approach to become very accurate and effective. Xinyuan called for future research to build on her work by updating the model to understand images with more buildings and adding more layers to the neural network.
In another Artificial Intelligence (AI) project, Nithin Sivakumar examined which machine learning method was the best at detecting fake news. Fake news is false or misleading information that is often found on social media and poses a significant threat to our society. For example, researchers from Stanford in 2020 found that 72% of tweets about the Covid-19 pandemic included false information. Fake news can spread extremely quickly - one person shares the fake news with people they know, who in turn share it with people they know, and so on. One potential solution is to use AI to identify news as fake news and prevent that first person from spreading the false information.
Nithin compared the effectiveness of three machine learning algorithms that are commonly discussed in fake news research: Naive Bayes, Random Forest, and Support Vector Machine. He did a great job at describing these algorithms, so I encourage you to check out his presentation here. He gathered three datasets that included political articles from several web sources. Then he cleaned the data, for example removing filler words and converting everything to lowercase. Next, he created three different models for each algorithm and did multiple evaluations on each model. Overall, his main finding was that each algorithm resulted in overfitting. In his words, overfitting is “when a certain model performs well when tested on the same model it was trained on, but not on others.” Nithin called for future research to develop more advanced machine learning methods that can address the problem of overfitting. He also suggested that combining algorithms, rather than examining them separately as he did in his research, may increase effectiveness.
Elizabeth Zhang was inspired to conduct her research on mental health and productivity by this finding from the American Psychiatric Association: “unresolved depression accounts for a 35% reduction in productivity, costing the economy $210.5 billion annually.” Her proposed solution was that governments address this problem by spending money on mental health programs. The first step to validating her proposal was identifying a relationship between increased spending and increased productivity. She used statistical techniques to investigate if there was a relationship between federal mental health expenditures and the respective country’s GDP per capita. GDP per capacity served as a proxy for productivity.
After gathering data of federal spending and GDP per capita from 2011 that spanned 77 countries, Elizabeth began her analysis. She developed R code to run multiple linear regression on the data. Multiple linear regression is useful because it estimates the relationship between multiple independent variables and one dependent variable. She found that for every 1% increase in spending on mental health (as a percentage of total health spending), there was a $1,639.20 increase in GDP. These results show that federal spending on mental health potentially increases the vitality of global economies and broadens standards of living. Elizabeth is currently expanding on her work by analyzing the results by geographical location and transforming the data into a better format for analysis.
Research Projects with Computer Science and Mechanical Engineering
Ethan Kou’s project combines robotics and computer science. He made a computer simulation of a robot that navigated using an Extended Kalman Filter (EKF). I had never heard of an EKF before watching his presentation, but Ethan gave this great analogy: if you are in your kitchen and the lights go out, you intuitively know how to estimate where you are in your kitchen. You know how many steps you’ve taken and what direction those steps are leading you (the “prediction”) and any physical things you can touch around you, like a table (the “measurement”). You combine the prediction and the measurement to create your estimation.
Ethan developed code in Python to simulate a robot that used these types of predictions and measurements to estimate its location. Specifically, this robot had four special wheels that allowed it to move forwards and sideways, a camera that detected images, and odometry pods, which are wheels that measure how far they’ve spun. He wrote code for the robot to calculate its position and velocity by combining multiple data sources (odometry, computer vision landmark detection, and a physics model). After running 100 simulations of the robot following a test path, he found that he was successful in combining multiple data sources. Next, he plans to create a physical robot that uses EKF, instead of a simulation. To create the physical robot, he will develop a more realistic model, for instance taking friction into account.
Rhea Vashishtha shared her ongoing work that combines mechanical engineering and computer science in her presentation, “Design and Control of 3D Printed, Actuated Prosthetic Hands.” Unfortunately, prosthetics are not accessible for many people because they are very expensive. Additionally, they are often bulky and difficult to control, which leads to people not wearing them. Designing prosthetics for children poses an additional challenge because the hand needs to be adjustable since children are constantly growing.
The goal of Rhea’s project is to create a less expensive and more sleek design of a prosthetic hand that is accessible to children in underprivileged areas. Rhea sketched out her initial design, which uses fishing wire and axles to create a pulley system that will allow the hand to open and close. Next, modeled her design in 3D using software called SolidWorks, and she will 3D-print this design. She will also program her prosthetic hand to be controllable in Python. Her final project goals include developing one full prosthetic hand, sending 15 simpler partial hand prosthetics to underprivileged areas through the non-profit organization Hands for Heroes, and creating a website detailing the project.
This article concludes my coverage of the Symposium talks! From watching so many presentations, I’ve seen the wide breadth of projects that Polygence students can do! I’ve also learned that the boundaries between subjects are very fluid. Many projects combine multiple fields, and it’s great seeing that the students can be creative and combine their interests.
If you want to do your own research under the guidance of an experienced mentor, you should apply to our flagship research program. Doing original research can give you that edge you need on your college admissions. You can also check out this article for creative ways to explore your passions.
Want to start a project of your own?
Click below to get matched with one of our expert mentors who can help take your project off the ground!