Luddy Faculty Fellows

Xiaojing Liao

Privacy-Accountable Large Language Model-Integrated System
The proliferation of large language model-integrated systems (LLMSs) e.g., healthcare chatbots or personal virtual assistants) has ushered in a new era of technological advancements and transformations across numerous aspects of our daily lives. These LLMSs, equipped with sophisticated algorithms and vast datasets, are instrumental in powering innovative solutions across diverse domains, from healthcare and finance to entertainment and communications. These systems have a remarkable capacity to understand and generate human-like text for tasks ranging from language translation to content generation. However, this rapid integration of large language model (LLMSs) into our lives has not occurred without raising important concerns, most notably in the realm of privacy. In this project, Liao argues that a privacy-accountable LLMS is the solution to implement proactive privacy-enhancing measures, continuously enforce privacy accountability, as well as fostering responsible data usage and privacy protection.

Qin Zhang

Quantum Data Management: Recording the Essence of Nature
Quantum information and computing seek to leverage the principles of quantum mechanics to fundamentally transform computation across many domains. Previous research has primarily focused on designing quantum algorithms that demonstrate a provable advantage over classical counterparts when applied to classical data. In contrast, conducting analytics directly on quantum data has largely remained an unexplored territory. The nature, along with scientific experiments, generates massive quantities of quantum data every day. In many cases, there is a natural desire for us to record the quantum data that has been collected or generated to facilitate future analysis.

In this project, Zhang aims to formulate a model for representing and storing quantum data that allows us to handle basic data analytics efficiently and sustainably. In pursuit of this goal, we must tackle distinctive challenges in the quantum world, such as post-measurement state disturbance, the no-cloning principle, and the expensive quantum state tomography. This project attempts to take the first step in this new and exciting research frontier, with the goal of paving the way for a comprehensive quantum data management system for the next generation.

Zoran Tiganj

Deep Reinforcement Learning with Human-Inspired Memory
Tiganj’s research aims to develop a novel neural network architecture that integrates insights from a computational neuroscience memory model with state-of-the-art Artificial Neural Networks (ANNs). This approach will result in a specific representation of knowledge in the form of cognitive maps, which are thought to play an essential role in human learning and reasoning. The evaluation will be done on a set of tasks that resemble tasks commonly used to study mammalian learning, such as interval timing, numerosity judgments, and spatial navigation.

Maria Bondesson

Leveraging Single Cell Sequencing for Deciphering Mechanisms Behind Chemical Exposure-Induced Birth Defects
About 8 million infants are born each year with birth defects. Although the underlying cause for each baby is often unknown, birth defects generally result from genetic alterations and/or environmental exposures. Major congenital anomalies usually occur during the embryonic period (weeks 3–8), a critical developmental phase during which fetal organogenesis takes place. Chemical and drug exposure during this susceptible embryonic period has been linked to fetal teratogenicity. Bondesson’s project aims to apply the recently developed method of single-cell mRNA sequencing to chemically exposed zebrafish embryos to elucidate the molecular mechanisms of teratogenicity. Data-producing experimentation, coupled with bioinformatic analysis, will, at a high resolution, identify the genes and cells of different cell lineages that are altered by chemical exposure and lead to birth defects.

Kahyun Choi 

Computational Poetry Analysis: AI for Literature
Choi’s research focuses on the intersection of technology, humans, and music. This project will develop AI-based NLP systems that achieve a high-level understanding of a poem, such as its emotion and theme. Her project will focus on three questions: 1) Can advanced deep learning-based language processing models, such as BERT, decipher emotions and themes of poems? 2) Can learned model on song lyrics be re-purposed for poem content analysis, given their common characteristics? 3) Do auxiliary data, such as authors’ notes on their poems and crowdsourced word emotion labels, provide additional information to the poem analysis systems?

Paul Macklin

Preparing Cancer Patient Digital Twins to Forecast Progression and Treatment
Macklin’s work focuses on cancer patient digital twins—predictive models that are tailored to individual patients to predict cancer progression and therapeutic response. This proposal will enhance the scope and expedite progress to prototype, analyze and refine CPDTs for melanoma pulmonary (lung) metastases. This project will develop the methodologies to fuse dynamical cancer models with AI and data assimilation, providing a critical step in personalizing predictive cancer models to assist clinical decision making and improve care.

Justin Wood

Reverse Engineering the Origins of Visual Intelligence
Wood creates high-precision methods for studying newborn animals, specifically, the origins of their core cognitive abilities. He then builds autonomous models that can create artificial brains that learn in the same way as newborns. As a cognitive scientist with a background in psychology, neuroscience, and artificial intelligence, Wood’s goal is to reverse engineer the learning mechanisms in newborn brains. In this project, he will use a new experimental system to build task-performing computational models of newborn visual systems.