Physical Sciences – Page 3 – UROP Spring Symposium 2021

Physical Sciences

Assessing the Accuracy of sPHENIX Design in Measuring Jet Charge

The purpose of this research project is to determine the efficacy of designs for the new sPHENIX detector at the Relativistic Heavy Ion Collider in measuring the charge of a quark jet with precision. The paper “Jet Charge: A Flavor Prism for Spin Asymmetries at the Electron-Ion Collider” by Kang, Liu, et al. provides the basis for this research by suggesting that determining jet charge could yield valuable clues as to the separation of various quark flavors in jet production and associated spin asymmetries. It is therefore imperative for the sake of later investigations into nucleon spin and flavor structure using sPHENIX to ensure that the detector is capable of accurately measuring jet charge. The study will utilize Monte Carlo simulations to produce known values for the charges of jets and the flavor of a parton initiating them. Code that mimics current sPHENIX detector design will then reconstruct these jet charge values in the jet-producing, simulated proton-proton collisions. Reconstructed jet charge values will then be compared against the known, generated values in order to assess the accuracy of the reconstruction. Assessing sPHENIX’s jet charge measurements will allow for improvements to its jet charge reconstruction algorithm and eventually more accurate probes of nucleon flavor and spin structure as described in Kang et al.

Acute respiratory distress syndrome (ARDS) diagnosis by image processing

Despite the vast amount of data available from hospitals on ARDS, much of it is left unused because working with all the available data can be slow and often inconsistent. This project focuses on chest x-ray scans, aiming to turn chest x-rays into privileged information (information that is not always available but can be useful) in order to improve training and increase the accuracy of ARDS classification when available. The project was done largely in Python with the exception of the segmentation algorithm (written in MATLAB), and classification training and accuracy tests for different features were done using a large amount of data from Michigan Medicine. My project’s findings have yielded a 0.72 AUC average score when doing binary classification on chest X-Rays using features extracted from image processing. Privileged information for machine learning is a very important step in ensuring that we harness all the data available, and my project’s findings will help inform future models and serve to show an extra pathway to improve classification accuracy.

Modeling Spinal Cord Decompression Sickness

Decompression sickness (DCS) occurs when dissolved gasses in the body exit the solution they’re contained in; upon depressurization, the visualizable bubbles form inside the body and compress surrounding tissue causing symptoms ranging from soreness to central nervous system damage. The current treatment for decompression sickness is to have the patient breathe 100% oxygen and be recompressed in a hyperbaric chamber. In the event that initial treatment does not completely resolve symptoms, treatment is repeated. This study investigates the impact of current treatment of DCS in the spinal cord by building a finite-element multiphysics model using COMSOL to simulate gas diffusion in the spinal cord and the mechanical response to a growing gas bubble. The model is developed through pressurizing spinal cord tissue from cows to simulate DCS and observing gas bubbles in the tissue through a series of MRIs taken during decompression and recompression. In the multiphysics model, we can analyze the simulated tissue for damages following the decompression and recompression. Currently, we have developed software to extract a 3D model from a series of MRIs of spinal cords. Our next step is to integrate this 3D model with a multiphysics model developed by another researcher in our group in order to run the aforementioned simulations.

Improving Student Mastery through Question Interleaving

Intelligent tutoring systems often recommend questions based on the estimation of students’ current ability estimates, the randomness, or the expectations from the instructors. To help students improve their masteries in a concept, we are going to propose a question recommendation model by utilizing question interleaving. Previous researches in math classes have shown that interleaving questions enhance student learning. Hence, assuming interleaving will benefit students’ learning, we examine how to effectively reordering the concepts of the questions within a session. Analyzing mainly on two online skill-builder datasets, we demonstrate that there exists a statistically significant difference in what we call sub-abilities from which interleaving can be optimized. We come up with a simple method that only depends on knowing a question concept id and a binary outcome for correctness. We are writing a Python library for simulation to compare how ideal results match with the proposed recommendation algorithms. The result of our findings aims to generate a personalized question recommendation for the students without gathering extensive data from them while maintaining the flexibility to include other models. The ultimate goal of our research is to provide students from different backgrounds with an efficient way to improve learning results.

Improving Student Mastery through Question Interleaving

Our research focuses on improving student masteries through question interleaving. Although previous researches by Rohrer and Taylor have shown that interleaving questions enhances student learning, there has been little research on question recommendation models that utilizes this concept to maximize student learning. We assume that question interleaving helps student learning. We aim to identify metrics that allow us to optimize the ordering of question interleaving and build a question recommendation model based on our findings. We are utilizing two datasets from online tutoring platforms that contain questions answered by students to analyze if our proposed metrics are statistically significant to base decisions on. We will also run simulations with Python on potential recommendation algorithms to see how well the result matches with a simple student ability model. The result of our findings aim to provide a question recommendation method that only relies upon the concept a question tests, but can be used in conjunction with further information. We hope our findings will align with our goal of having a question recommendation model that maximizes student learning. This allows for personalization in intelligent tutoring systems even before significant amounts of student/question data has been generated. The ultimate goal/implication of our research is to help improve learning results for all students simply by changing the order of their practice questions, without increasing their workload.

Information Geometry: Unified Framework for Information, Machine Learning, and Statistical Inference

This study aims at computing around KL-divergences of probability density distributions around different metrics. We try to compute 2nd/3rd derivatives for different metrics, affine connections and levi-civita Connections in particular, under these distributions in order to have an idea about the flatness of the different spaces we are working on. Given the nature of this independent study, most of these computations and results have already been proven before. We computed KL divergences, Jacobian and Fisher matrices under different metrics. We then derive some integral forms for Fisher Information matrices, affine connections, dual connections and derivative forms for coordinate transformation for these parameters. We found out that for metrics in exponentials, the affine connections turn out to be 0 everywhere, while in expectation coordinates, the dual connections are 0. This means that in either case, we are working with a flat space. Nothing spectacular has been found out around the normal metric(namely, (mean, standard deviation)).

lsa logoum logo