Dana Kim

My research focuses on developing and implementing advanced deep-learning models to significantly enhance the interpretative accuracy and efficiency of medical imaging, thereby improving disease diagnosis processes. This initiative addresses the urgent need for sophisticated diagnostic tools capable of managing complex diseases effectively and seeks to overcome the limitations associated with human-dependent analysis. The research utilizes a diverse array of medical images such as MRIs, CT scans, and X-rays as primary data for the training and refinement of algorithms. The research methodology is organized into distinct phases: Data Collection and Labeling, […]

Daniel Lee

Graphene-based “twistronics” – samples with multiple layers of graphene stacked possibly with twists – have shown to host a plethora of interesting physics, such as superconductivity and dissipationless transport. At the heart of all this physics lies the moire pattern – large periodic structures when two or more layers of graphene are stacked on top of each other with a twist. The geometry of the moire structure for a particular graphene twistronic determines its electrical properties. Recently, a “magic-angle” of 1.9° has been proposed for helical trilayer graphene (HTG), a […]

Tony Li

Edge computing explorations in POET(Private Optimal Energy Training) and large language model fine-tuning studies revolved around Gorilla LLM with RLHF(Reinforcement Learning Human Feedback)

Joshua Ho

Particle physicists seek to understand particles and the physical laws governing their interactions by building particle colliders. These colliders provide a large amount of data, and due to the probabilistic nature of particle physics, researches are turning towards ML to model complex physical processes. With the rise of foundational models like ChatGPT, particle physicists have been inspired to create a large scale, general purpose model that has been trained on a vast amount of data to serve as a strong starting point for various specific tasks, that can be fine-tuned […]

Clara Hung

Lensless imagers are low-cost, compact cameras with applications in medical imaging, photography, and more. Many designs for lensless imagers have been proposed, but the optimal design is not known as it is object-dependent. A method to capture images from different systems under similar conditions is needed to fairly compare system performance. Furthermore, as lensless imaging is moving towards data-intensive research, large-scale lensless measurement datasets are necessary for neural network evaluation. Yet, of the few existing datasets in the field, none fully address these demands. We propose a portable data acquisition […]

Stephen Chen

Since proteins carry out the majority of work in living cells, how a cell operates is heavily determined by the relative abundances of different proteins at any given time. These relative abundances are largely determined by gene expression levels: the extent to which a gene is “activated.” Predicting gene expression levels from noncoding DNA sequence is a major unsolved problem in computational biology. Recently, machine learning (ML) has emerged as one of the most effective tools for this task. While significant work has been done with human genomes, much work […]

Jorge Diaz Chao

We are building a novel platform that introduces a new dimension to asynchronous learning—like YouTube did upon launch—by empowering users with no coding experience to democratize their knowledge on tasks involving coordinated motor skills through Augmented Reality (AR) scenarios. Our work involves designing a user interface capable of inferring motor and linguistic cues from demonstrations and explanations, and writing an algorithm that synthesizes probabilistic code, namely Scenic, that models motion primitives to build complex behaviors. To do so, we employ tools such as symbolic learning and Large Language Models (LLMs) […]

Benjamin Eisley

In the past few years, neural networks have gone from obscure to ubiquitous. This technology is shockingly versatile, but conceptually ill-understood: there is a large gap between practice and theory, and much has yet to even be conjectured. For example, scientists are baffled by the overfitting paradox. Overfitting is usually a problem when programmers model a complex system such as the brain. Programmers must base their model on finitely many examples of that system’s behavior. Traditionally, programs that perfectly replicate these examples forget the underlying system. Surprisingly, large neural networks […]

Nir Elber

One goal of arithmetic geometry is to enumerate the points on geometric surfaces with rational coordinates. Over the past century, it has been profitable to study the geometry of the surface directly. For example, a “cohomology theory” is a way to assign a sequence of geometric invariants to the surface; it turns out that one can use cohomology in order to count points. Given a surface, there tend to be many reasonable cohomology theories. This project is interested in the symmetries of a cohomology theory. Given a cohomology theory, it […]

Keon Abedi

This project will investigate how the frequency of a mutation changes over time when in the presence of correlated environmental noise. This is an open question in the biophysics of evolutionary dynamics, which seeks a quantitative framework for Darwin’s theory of natural selection. The core idea of evolution is that a mutation’s frequency tends to increase over generations when it offers an advantage to its hosts. For example, in an environment with antibiotics, mutations conferring antibiotic resistance are strongly favored and can quickly become prevalent across a bacterial population; we […]