About

Boxuan Zhao is an Assistant Professor in the Department of Cell and Developmental Biology and the Neuroscience Program at the University of Illinois. He is also an affiliate of the Carl R. Woese Institute for Genomic Biology. His research focuses on the molecular biology and biochemistry of RNA, with particular emphasis on RNA modifications such as N6-methyladenosine (m6A) and 5-hydroxymethylcytosine (5hmC). Zhao's work involves developing innovative technologies to study RNA modifications and their roles in living cells, including high-throughput mapping of neuronal connectivity at single-synapse resolution via barcode sequencing. His research integrates advanced biochemical and genetic approaches to unravel the complexities of RNA regulation and its impact on gene expression and cellular function. Zhao has contributed to the understanding of RNA modification mechanisms and engineered tools for precise spatiotemporal control of molecular processes in living cells, advancing the fields of neuroscience, molecular biology, and bioorganic chemistry.

Research topics

• Artificial Intelligence
• Computer Science
• Machine Learning
• Algorithm
• Mathematics
• Theoretical computer science

Selected publications

• Reinforced physiology-informed learning for image completion from partial-frame dynamic PET imaging

  Medical Image Analysis · 2025-09-03

  article
  PublisherDOI

• Substitution of DMSO for Stabilized Perovskite Solar Cells with Extended Process Window

  Advanced Functional Materials · 2025-06-25 · 4 citations

  articleOpen access1st author

  Abstract Dimethyl sulfoxide (DMSO) is frequently employed to boost the crystal quality of solution‐processed perovskites, while it is prone to remain trapped within the films and leads to defective interface within the resultant perovskite solar cells (PSCs). Herein, a small molecule of hydroxyethyl methacrylate (HEMA) is introduced to substitute the DMSO. The hydroxyl (─OH) and carbonyl (─C═O) groups in HEMA are simultaneously associated with formamidinium (FA + ) and Pb 2+ via hydrogen bonds and coordination bonds, respectively, which facilitates the formation of strongly bonded FAI‐HEMA‐PbI 2 complexes in the precursor solution to regulate perovskite crystallization with improved crystallinity and preferred orientation. Moreover, the solidification of residual HEMA via in situ polymerization can stabilize crystal structure with suppressed defects and released lattice strain. Consequently, PSCs based on HEMA‐treated perovskite films achieve a decent power conversion efficiency (PCE) of 25.31% with superior stability, retaining 90% of their initial PCE after 1000 h storage. Importantly, the incorporated hydrophilic HEMA can largely promote the moisture resistance of the precursor solution by preventing water molecules from direct contact with perovskite components. More than 90% of the initial efficiency is maintained by using old precursor solutions aged in ambient air for 20 days, indicating an extended process window for device fabrication.

  PublisherOA PDFDOI

• Enhancing Long Video Understanding via Hierarchical Event-Based Memory

  2025-06-30

  articleSenior author

  Recently, integrating visual foundation models into large language models (LLMs) to form video understanding systems has attracted widespread attention. Most of the existing models compress diverse semantic information within the whole video and feed it into LLMs for content comprehension. While this method excels in short video understanding, it may result in a blend of multiple event information in long videos due to coarse compression, which causes information redundancy. Consequently, the semantics of key events might be obscured within the vast information that hinders the model’s understanding capabilities. To address this issue, we propose a Hierarchical Event-based Memory-enhanced LLM (HEM-LLM) for better understanding of long videos. Firstly, we design a novel adaptive sequence segmentation scheme to divide multiple events within long videos. In this way, we can perform individual memory modeling for each event to establish intra-event contextual connections, thereby reducing information redundancy. Secondly, while modeling current event, we compress and inject the information of the previous event to enhance the long-term inter-event dependencies in videos. Finally, we perform extensive experiments on various video understanding tasks and the results show that our model achieves state-of-the-art performances.

  PublisherDOI

• Towards Universal Dataset Distillation via Task-Driven Diffusion

  2025-06-10 · 2 citations

  article

  Dataset distillation (DD) condenses key information from large-scale datasets into smaller synthetic datasets, reducing storage and computational costs for training networks. However, most recent research has primarily focused on image classification tasks, with limited exploration in detection and segmentation. Two key challenges remain: (i) Task Optimization Heterogeneity, where existing methods focus on class-level information but fail to address the diverse needs of detection and segmentation, and (ii) Inflexible Image Generation, where current generation methods rely on global updates for single-class targets and lack localized optimization for specific object regions. To address these challenges, we propose UniDD, a universal dataset distillation framework built on a task-driven diffusion model for diverse DD tasks, as shown in Fig. 1. Our approach operates in two stages: Universal Task Knowledge Mining, which captures task-relevant information through task-specific proxy model training, and Universal Task-Driven Diffusion, where these proxies guide the diffusion process to generate task-specific synthetic images. Extensive experiments across ImageNet-1K, Pascal VOC, and MS COCO demonstrate that UniDD consistently outperforms state-of-the-art methods. In particular, on ImageNet-1K with IPC-10, UniDD surpasses previous diffusion-based methods by 6.1%, while also reducing deployment costs.

  PublisherDOI

• Using barium isotopes to distinguish metamorphic and magmatic fluids for the gold deposits

  Ore Geology Reviews · 2024-11-14 · 4 citations

  articleOpen access

  • Variation Ba isotopes were controlled by crystallization and dissolution of Ba-bearing minerals, particularly barite. • Ba isotopes exhibiting a significant difference between the magmatic and metamorphic Au-bearing fluids. • Barium isotopes are potential tracers for distinguishing origin of ore-forming fluids. The orogenic and intrusion-related gold deposits represent the two most significant types of gold reserves globally, collectively accounting for over half of the total and are formed associated with metamorphic and magmatic hydrous fluids, respectively. Theoretically, gold deposits should be a result of activities of the metamorphic and magmatic hydrous fluids that sourced, carried, and reserved gold. However, distinguishing differences between the metamorphic and magmatic fluids proves challenging mainly due to overlapping of trace elements, S, and Re-Os isotopes of sulfides and C, H, O, He, and Ar isotopes of the ore-forming fluids. Barium, as a fluid sensitive cation, is believed to faithfully record the sources and evolution of fluids. In this work, we presented the Ba isotopes of various ore-rocks from the Hadamen and Haoyaoerhudong gold deposits along the northern margin of the North China Craton. These two deposits have been well-defined the ore-forming fluids that originated from crystal-melt separation in alkaline granitic magma and dehydration of chlorite and mica from the black sales during metamorphism, respectively. The Ba isotopes exhibit significantly fractionation among the various ore-rocks from above two gold deposits. With increasing SiO 2 content, δ 138/134 Ba values increased from −0.28 ‰ in the potassium silicate alteration zone, crossed −0.21 ‰ ∼ −0.19 ‰ in the potassium silicate alteration zone filled with sulfide-quartz veins, to −0.13 ‰ ∼ +0.01 ‰ in the sulfide-quartz veins. This suggests that the heavier Ba isotopes were preferentially incorporated into the evolving magmatic fluids primarily due to the crystallization of K-feldspar and barite. In contrast, δ 138/134 Ba values decreased from the carbonaceous slate (+0.73 ‰ ∼ +0.95 %) to sulfide veins (−0.28 ‰ ∼ +0.07 ‰), then increased to sulfide-quartz veins (+0.01 ‰). This phenomenon results from the continuously enhanced dissolution of diagenetic barite accompanying metamorphism and the crystallization of barite during evolution of metamorphic fluids. The microstructural characteristics also support that crystallization and dissolution of barite control significant Ba isotope fractionation in two types of fluids with different features. Furthermore, the magmatic and metamorphic fluids exhibit relative positive and negative relationships between Ba content and δ 138/134 Ba values, respectively. These geochemical features are also useful in defining the origin of the ore-forming fluids. Therefore, we propose that Ba isotope composition will be a new tool for deciphering the evolution of the Au-bearing ore-forming fluids and distinguishing the origin of the ore-forming fluids.

  PublisherDOI

• Accelerated 4D Flow MRI with Low-Rank Modeling and A Deep Generative Prior

  Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition · 2024-11-26

  articleSenior author

  Motivation: Conventional 4D flow MRI provides valuable insights into blood flow but suffers from long scan times. Recent machine learning methods improve MRI reconstruction; however, they often require a large amount of training data to achieve desired performance. Goal(s): This work is aimed to introduce a novel learning-based image reconstruction method to accelerate 4D flow MRI without using training datasets. Approach: The proposed method integrates low-rank modelling with a deep generative prior by utilizing an untrained generative neural network to represent the spatial subspace of the model. Results: The effectiveness of the proposed method has been demonstrated with in-vivo aortic 4D flow experiments. Impact: This work introduced an innovative learning-based image reconstruction method for accelerating 4D flow MRI, which produces accurate velocity measurements even under high acceleration factor, all without the need for training datasets.

  PublisherDOI

• OLAP on Modern Chiplet-Based Processors

  Proceedings of the VLDB Endowment · 2024-07-01 · 4 citations

  article

  Chiplet-based CPUs, which combine multiple independent dies on a single package, allow hardware to scale to higher CPU core counts at the cost of more memory heterogeneity and performance variability. This introduces challenges when existing query engines are deployed on chiplet-based CPUs, as current designs make assumptions about uniform memory access, cache locality and consistent core performance, e.g., leading to ineffective CPU utilization. In this paper, we analyse the performance impact when query engines ignore chiplet-specific properties. We demonstrate that a naïve deployment can result in a significant degradation of query processing efficiency, exhibiting non-linear scaling even within a single CPU socket domain. Based on comprehensive experiments, we explore approaches to deploy query engines on chiplet-based CPUs with improved performance: we show that distributing processing tasks according to a chiplet-aware strategy achieves higher resource utilization and scalability, yielding an up to 7× speedup compared to hardware-oblivious approaches.

  PublisherDOI

• Bayesian Cramér-Rao Bound Estimation With Score-Based Models

  IEEE Transactions on Information Theory · 2024-08-21 · 4 citations

  articleSenior author

  The Bayesian Cramér-Rao bound (CRB) provides a lower bound on the mean square error of any Bayesian estimator under mild regularity conditions. It can be used to benchmark the performance of statistical estimators, and provides a principled metric for system design and optimization. However, the Bayesian CRB depends on the underlying prior distribution, which is often unknown for many problems of interest. This work introduces a new data-driven estimator for the Bayesian CRB using score matching, i.e., a statistical estimation technique that models the gradient of a probability distribution from a given set of training data. The performance of the proposed estimator is analyzed in both the classical parametric modeling regime and the neural network modeling regime. In both settings, we develop novel non-asymptotic bounds on the score matching error and our Bayesian CRB estimator based on the results from empirical process theory, including classical bounds and recently introduced techniques for characterizing neural networks. We illustrate the performance of the proposed estimator with two application examples: a signal denoising problem and a dynamic phase offset estimation problem with applications in communication systems.

  PublisherDOI

• Enhancing Long Video Understanding via Hierarchical Event-Based Memory

  arXiv (Cornell University) · 2024-09-10

  preprintOpen accessSenior author

  Recently, integrating visual foundation models into large language models (LLMs) to form video understanding systems has attracted widespread attention. Most of the existing models compress diverse semantic information within the whole video and feed it into LLMs for content comprehension. While this method excels in short video understanding, it may result in a blend of multiple event information in long videos due to coarse compression, which causes information redundancy. Consequently, the semantics of key events might be obscured within the vast information that hinders the model's understanding capabilities. To address this issue, we propose a Hierarchical Event-based Memory-enhanced LLM (HEM-LLM) for better understanding of long videos. Firstly, we design a novel adaptive sequence segmentation scheme to divide multiple events within long videos. In this way, we can perform individual memory modeling for each event to establish intra-event contextual connections, thereby reducing information redundancy. Secondly, while modeling current event, we compress and inject the information of the previous event to enhance the long-term inter-event dependencies in videos. Finally, we perform extensive experiments on various video understanding tasks and the results show that our model achieves state-of-the-art performances.

  PublisherOA PDFDOI

• Score Matching with Deep Neural Networks: A Non-Asymptotic Analysis

  2024-11-24

  articleSenior author

  Score matching is a statistical approach for estimating the score (the gradient of the log-density) of a probability distribution from samples. It has found a number of applications, including in generative modeling, where it serves as a key component of the state-of-the-art diffusion modeling framework. The goal of this work is to provide non-asymptotic bounds on the score matching risk in the setting where the score model is a deep neural network. Here key challenges include the fact that the score model is vector-valued and that the score matching loss depends on the Jacobian of the score model. Our approach integrates results from empirical process theory, including classical bounds and recently introduced techniques for bounding covering numbers of neural network models, with novel covering results to address these challenges. The resulting bound has logarithmic dependence on the network width, allowing the network size to grow exponentially with the number of training samples without compromising the bound.

  PublisherDOI

Recent grants

• NIH Grant F32EB024381

  NIH · $120k · 2019

• Optimized MR Fingerprinting for Rapid Volumetric Quantitative Neuroimaging

  NIH · $185k · 2019–2021

Frequent coauthors

• Lawrence L. Wald

  74 shared

• Kawin Setsompop

  59 shared

• Berkin Bilgic̦

  34 shared

• Wei Zhao

  Beihang University

  24 shared

• Huihui Ye

  19 shared

• Mark A. Griswold

  Case Western Reserve University

  17 shared

• Sabee Molloi

  University of California, Irvine

  16 shared

• Huanjun Ding

  Nanjing University

  15 shared

Labs

• Zhao LabPI

Education

• Ph. D. , Electrical and Computer Engineering

  University of Illinois at Urbana-Champaign

  2014

• MS, Automation

  Tianjin University

  2007

• BS, Automation

  Tianjin University

  2005

Awards & honors

• 2019 Life Sciences Research Foundation Shurl and Kay Curci F…
• 2018 Stanford ChEM-H Postdocs at the Interface Seed Grant, S…
• 2017 Wu Tsai Neurosciences Institute Interdisciplinary Schol…
• 2017 Scaringe Graduate Student Career Award, The RNA Society
• 2017 Elizabeth R. Norton Prize for Excellence in Research in…