About

I am an interdisciplinary data scientist and geographer interested in urbanization, changing landscape, and impacts on human health. I developed remote sensing approaches combined with machine learning in order to characterize forms of cities not only at large scales but also at a spatial explicit resolution close to people's living environment (e.g., type of housing at walking distance around home) to understand health disparities.

Research topics

• Computer Science
• Artificial Intelligence
• Machine Learning
• Software engineering
• Statistical physics
• Quantum mechanics
• Physics
• Theoretical computer science
• Materials science
• Thermodynamics
• Programming language
• Chemistry
• Algorithm
• Computer engineering
• Parallel computing
• Mathematical optimization

Selected publications

• LUMR: leveraging user decision processes to model unknowns in recommendation system

  International Journal of Data Science and Analytics · 2026-05-18

  articleSenior author
  PublisherDOI

• Causal Entropy–Driven Generative Adversarial Learning for Anomalous Event Detection in Social Networks

  Frontiers in Computing and Intelligent Systems · 2025-11-27

  articleOpen accessSenior author

  In social networks, anomalous events not only disrupt normal network evolution but also pose potential threats to network security. Therefore, achieving efficient anomaly detection in dynamically and semantically complex environments is of significant importance. However, existing methods often struggle to effectively capture the dynamic evolutionary relationships between events and their underlying causal dependencies. To address this, this paper proposes a causal entropy–driven generative adversarial learning for anomalous event detection in social networks. First, a log parser converts unstructured social network logs into structured events, which contain timestamps, IP addresses, and event content. Subsequently, we standardize causal entropy to measure the strength of causal relationships between events and construct causal sequences. Building upon this foundation, we design a generative adversarial framework. The generator combines a gated recurrent unit, a self-attention mechanism, and a long short-term memory network to capture deep event semantics and generate realistic log samples. The discriminator achieves precise differentiation between normal and abnormal events through mean cross-entropy loss on fully connected layers. Experimental results demonstrate that the proposed method achieves approximately 6.5% higher detection accuracy than existing approaches, validating its effectiveness and robustness in identifying anomaly events.

  PublisherOA PDFDOI

• Tighter bounds on non-clairvoyant parallel machine scheduling with prediction to minimize makespan

  Information Processing Letters · 2025-08-29

  article1st author
  PublisherDOI

• A motion direction detecting model for colored images based on the Hassenstein–Reichardt model

  Machine Vision and Applications · 2025-01-01 · 1 citations

  article
  PublisherDOI

• FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving

  arXiv (Cornell University) · 2025-01-02 · 3 citations

  preprintOpen access

  Transformers, driven by attention mechanisms, form the foundation of large language models (LLMs). As these models scale up, efficient GPU attention kernels become essential for high-throughput and low-latency inference. Diverse LLM applications demand flexible and high-performance attention solutions. We present FlashInfer: a customizable and efficient attention engine for LLM serving. FlashInfer tackles KV-cache storage heterogeneity using block-sparse format and composable formats to optimize memory access and reduce redundancy. It also offers a customizable attention template, enabling adaptation to various settings through Just-In-Time (JIT) compilation. Additionally, FlashInfer's load-balanced scheduling algorithm adjusts to dynamism of user requests while maintaining compatibility with CUDAGraph which requires static configuration. FlashInfer have been integrated into leading LLM serving frameworks like SGLang, vLLM and MLC-Engine. Comprehensive kernel-level and end-to-end evaluations demonstrate FlashInfer's ability to significantly boost kernel performance across diverse inference scenarios: compared to state-of-the-art LLM serving solutions, FlashInfer achieve 29-69% inter-token-latency reduction compared to compiler backends for LLM serving benchmark, 28-30% latency reduction for long-context inference, and 13-17% speedup for LLM serving with parallel generation.

  PublisherOA PDFDOI

• Bioenergy Utilization and Greenhouse Gas Emission Reduction: A Global Impact Assessment

  SSRN Electronic Journal · 2025-01-01 · 1 citations

  preprintOpen accessSenior author
  PublisherDOI

• Higher-order protection of quantum gates: Hamiltonian engineering coordinated with dynamical decoupling

  Physical review. A/Physical review, A · 2025-02-20 · 2 citations

  article

  Dynamical decoupling represents an active approach towards the protection of quantum memories and quantum gates. Because dynamical decoupling operations can interfere with a system's own time evolution, the protection of quantum gates is more challenging than that of quantum states. In this work, we put forward a simple but general approach towards the realization of higher-order protection of quantum gates and further execute the first cloud-based demonstration of dynamical-decoupling-protected quantum gates at the first order and the second order. The central idea of our approach is to engineer (hence regain the control of) the gate Hamiltonian in coordination with higher-order dynamical decoupling sequences originally proposed for the protection of quantum memories. The physical demonstration on an IBM quantum processor indicates the effectiveness and potential of our approach on noisy intermediate scale quantum computers.

  PublisherDOI

• Observation of the non-Hermitian skin effect and Fermi skin on a digital quantum computer

  Nature Communications · 2025-02-04 · 33 citations

  articleOpen access

  Lately, the non-Hermitian skin effect (NHSE) has been demonstrated in various classical metamaterials and even ultracold atomic arrays. Yet, its interplay with many-body dynamics have never been experimentally investigated. Here, we report the observation of the NHSE and its many-fermion analog on a universal quantum processor. To implement NHSE accumulation on a quantum computer, the time-evolution circuit not only needs to be non-reciprocal and non-unitary, but must also contain sufficiently many lattice qubits. We demonstrate this by systematically post-selecting ancilla qubits, as demonstrated through two paradigmatic non-reciprocal models on noisy quantum processors, with clear signatures of asymmetric spatial propagation and many-body “Fermi skin” accumulation. To minimize errors from inevitable device noise, time evolution is performed using trainable, variationally optimized quantum circuits. Our demonstration represents an important step in the quantum simulation of non-Hermitian lattices on present-day quantum hardware, and can be readily generalized to more sophisticated many-body models. Non-Hermitian skin effect, a phenomenon where eigenstates accumulate at the boundaries of a non-Hermitian system, has been observed in various platforms but primarily at the single particle level. Here the authors demonstrate the interplay of this effect with many-body physics on a superconducting quantum processor.

  PublisherOA PDFDOI

• A Bio-Inspired Learning Dendritic Motion Detection Framework with Direction-Selective Horizontal Cells

  Biomimetics · 2025-05-02

  articleOpen access1st author

  Motion direction detection is an essential task for both computer vision and neuroscience. Inspired by the biological theory of the human visual system, we proposed a learnable horizontal-cell-based dendritic neuron model (HCdM) that captures motion direction with high efficiency while remaining highly robust. Unlike present deep learning models, which rely on extension of computation and extraction of global features, the HCdM mimics the localized processing of dendritic neurons, enabling efficient motion feature integration. Through synaptic learning that prunes unnecessary parts, our model maintains high accuracy in noised images, particularly against salt-and-pepper noise. Experimental results show that the HCdM reached over 99.5% test accuracy, maintained robust performance under 10% salt-and-pepper noise, and achieved cross-dataset generalization exceeding 80% in certain conditions. Comparisons with state-of-the-art (SOTA) models like vision transformers (ViTs) and convolutional neural networks (CNNs) demonstrate the HCdM's robustness and efficiency. Additionally, in contrast to previous artificial visual systems (AVSs), our findings suggest that lateral geniculate nucleus (LGN) structures, though present in biological vision, may not be essential for motion direction detection. This insight provides a new direction for bio-inspired computational models. Future research will focus on hybridizing the HCdM with SOTA models that perform well on complex visual scenes to enhance its adaptability.

  PublisherOA PDFDOI

• A strategy of chiral cation coordination to achieve a large luminescence dissymmetry factor in 1D hybrid manganese halides

  Chemical Science · 2025-01-01 · 12 citations

  articleOpen access

  A pair of Mn-based bromide enantiomers R / S -DACAMnBr 3 were synthesized via a strategy of chiral cation coordination, which exhibited a large g lum up to 0.292, positive magneto-chiroptical effect, and great optoelectronic application potential.

  PublisherOA PDFDOI

Frequent coauthors

• Tim Byrnes

  34 shared

• Yong Yu

  Institute of Economics

  17 shared

• Luís Ceze

  16 shared

• Carlos Guestrin

  15 shared

• Thierry Moreau

  University of Washington

  15 shared

• Ziheng Jiang

  14 shared

• Arvind Krishnamurthy

  Stanford University

  11 shared

• Jared Roesch

  10 shared

Labs

• Health and Urbanization LabPI

Education

• Master of Science, Engineering Physics (Theoretical physics specialization)

  KTH Royal Institute of Technology Centre for Autonomous Systems

  2016

Awards & honors

• Leading Women in Machine Learning for Earth Observation (202…