UC San Diego

Name: Christopher Theissen
Title: Measuring the properties of twin stars to the smallest known planet hosts
Description: Over the past decade we have discovered thousands of planets. We have also learned that the smallest stars host the majority of terrestrial planets. However, we still don’t know how unique or common it is for the smallest stars to host planets with only a small number of known systems, most famously TRAPPIST-1 with seven known terrestrial planets. One intriguing property of the lowest-mass planet hosts is their apparent low-surface-gravity signature, possibly caused by tidal interactions with their orbiting planets. This project will investigate the similarities between known planet hosts and “twin” stars selected with very similar photometric or spectroscopic properties to better understand the the uniqueness of the smallest planet-hosting stars. Students will analyze high-resolution near-infrared spectra taken with the NIRSPEC instrument on the Keck 10-m telescope and the IGRINS instrument on the Gemini-South 8-m telescope. There will also be opportunities to assist in additional observing runs during the program.
Preferred qualifications: Python is extremely useful in getting them a jump start. Knowing how to read in tables, or access resources such as VizieR is a bonus. None of this is a dealbreaker, but allows them to get a little farther over the summer.

Name: Kam Arnold
Title: Analysis of Cosmic Microwave Background measurement by the Simons Observatory Small Aperture Telescope
Description: The Simons Observatory Small Aperture Telescope achieved first light in October of 2023, and we have started validation and commissioning of the instrument. By Summer of 2024, we will have our commissioning and initial science operations data, and will be analyzing it to understand the calibration of the instrument and assess statistical and systematic uncertainties in the data
Preferred qualifications: Python programming, understanding of Fourier analysis



Name: Flavio Ponzina
Title: Smartwatches meet Hyperdimensional Computing
Description: Smartwatches are now equipped with multiple sensors capable of measuring activity recognition as well as health information, including heart rate, heart rate variability, and oxygen saturation among others. The noisy data acquisition environment typical of daily life (e.g., variations in light, temperature, humidity, and device position on the wrist) requires wearable devices to perform complex feature extraction, pre-processing, and data analysis to produce accurate results. This complexity leads to faster battery discharge which may impact user experience.
Hyperdimensional computing (HDC) has emerged in the last few years as a lightweight and robust machine learning approach, demonstrating effectiveness in multiple tasks. The project is related to the research activities of the SEE Lab of UC San Diego, led by Professor Tajana Rosing. The goal is to evaluate and analyze the potential application of HDC models in smartwatches to process sensor data for efficient measurement of health-related metrics, ultimately reducing the compute and memory intensity of current approaches.
Preferred qualifications: Python programming, basics of machine learning

Name: Rose Yu
Title: Generative AI
Description: Generative AI is at the forefront of AI research. It is the technology that is powering many successful applications such as Dalle and ChatGPT. The research project will dive into the fundamentals of generative AI. The goal is to develop new algorithms and models that investigate the robustness, generalizality and trustworthiness of generative AI, especially for scientific problems such as climate change, drug discovery and etc.
Preferred qualifications: Familiarity with python and machine learning. Fast learning skills.

Name: Alex Cloninger
Title: Machine Learning on Graphs
Description: Many important problems can be represented and studied using graphs — social networks, interacting particles, brain networks, hierarchical image clustering and many more. However, most machine learning techniques assume the input data comes from a finite dimensional vector space. This motivates the need to study the extension of machine learning to graphs; this can be used for everything from prediction of labels of the nodes in the network, to prediction of missing edges, to classifying different types of graphs.
We will examine a couple particular problems within this domain, focusing on developing algorithms that are both computationally feasible and with theoretical guarantees. These problems will be a choice from: (1) encoding the notion of a decision boundary into graph edges, (2) building classifiers for different types of graphs when the graph is built from k-nearest neighbors on a point cloud, and (3) predicting links in a bipartite graph with side information on each node.
Preferred qualifications: Taken a course in linear algebra, some familiarity with Python/Matlab/R programming.