data/chunks.json

[
 {
  "doc_id": "profile",
  "title": "Profile of Prof. Liang Zhao — Prof. Liang Zhao — Profile",
  "source": "profile",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "# Prof. Liang Zhao — Profile",
  "chunk_id": 0
 },
 {
  "doc_id": "profile",
  "title": "Profile of Prof. Liang Zhao — Brief description (at a glance)",
  "source": "profile",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## Brief description (at a glance)\nDr. Liang Zhao is an award-winning AI researcher at Emory University who builds\n**scalable and trustworthy machine learning for structured, spatial, and scientific\nproblems**. Over more than a decade his work has spanned a remarkable arc — from the\ndeployed **EMBERS** civil-unrest forecasting system, through foundational work on\n**graph neural networks and deep graph generation**, to today's frontier of\n**retrieval-augmented and agentic large language models**, **AI for science**\n(molecules, epidemiology, power grids), and **LLM efficiency**. He is among\n**Stanford's \"Top 2%\" most-cited scientists**, with **~12,200 citations, an h-index of\n52, and an i10-index of 213**, and his research is funded by **NSF (including a CAREER\nAward), NIH**, and industry partners such as **Amazon, Meta, and NVIDIA**. His guiding\nphilosophy: *\"Whatever science domain I work with, I abstract the core problem into\nmathematical language.\"*",
  "chunk_id": 1
 },
 {
  "doc_id": "profile",
  "title": "Profile of Prof. Liang Zhao — Who he is",
  "source": "profile",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## Who he is\nDr. Liang Zhao is a **Winship Distinguished Research Professor** and **Associate\nProfessor** in the **Department of Computer Science at Emory University**, with a\njoint appointment in the **Cell & Molecular Biology Division of the Winship Cancer\nInstitute**. He earned his **Ph.D. from Virginia Tech in 2017**. He has authored\n**200+ peer-reviewed papers**, with roughly **12,000 citations, an h-index of 52,\nand an i10-index of 212** (Google Scholar). His work is supported by **NSF, NIH**,\nand industry partners including **Amazon, Meta, and NVIDIA**.\n\nHomepage: https://cs.emory.edu/~lzhao41/\nGoogle Scholar: https://scholar.google.com/citations?user=qnvyqtwAAAAJ&hl=en",
  "chunk_id": 2
 },
 {
  "doc_id": "profile",
  "title": "Profile of Prof. Liang Zhao — Research mission",
  "source": "profile",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## Research mission\nHe builds **scalable and trustworthy machine learning for structured, spatial, and\nscientific domains**. A recurring theme in his own words: \"Whatever science domain\nI work with, I abstract the core problem into mathematical language.\"",
  "chunk_id": 3
 },
 {
  "doc_id": "profile",
  "title": "Profile of Prof. Liang Zhao — Research mission",
  "source": "profile",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "### Core research areas\n- **Graph neural networks & graph learning** — foundations, deep graph generation,\n  graph transformation, text-attributed graphs, distributed/efficient GNN training,\n  spectral–spatial unification, GNN explainability.\n- **Geospatial & spatio-temporal reasoning** — spatial event forecasting, trajectory\n  generation, polyhedra/3D geometry representation learning, spatial RAG and\n  spatial agents.\n- **AI for science / scientific ML** — AI-guided molecule and disinfectant design,\n  neurobiology simulation, wildfire-smoke modeling, tumor detection, MRI-based brain\n  health, healthcare knowledge graphs.\n- **Large language models** — retrieval-augmented generation (RAG), graph & citation\n  RAG, agentic systems, domain specialization, uncertainty quantification, federated\n  pruning, resource-efficient LLMs.\n- **Trustworthy & efficient AI** — explainability, robustness, optimization for\n  infrastructure and energy efficiency, LLM cost reduction.",
  "chunk_id": 4
 },
 {
  "doc_id": "profile",
  "title": "Profile of Prof. Liang Zhao — Research impact & reach",
  "source": "profile",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## Research impact & reach\n- **Metrics (Google Scholar):** ~12,200 total citations, h-index 52, i10-index 213;\n  most citations are recent (since 2021), reflecting an accelerating research program.\n- **Highly cited works** include the *Graph Neural Networks: Foundations, Frontiers,\n  and Applications* book (1400+ citations), the *Domain Specialization of LLMs* survey,\n  the *EMBERS* civil-unrest forecasting system, and *GRAG: Graph Retrieval-Augmented\n  Generation*.\n- **Frequent collaborators** span top institutions and labs: Chang-Tien Lu and Naren\n  Ramakrishnan (Virginia Tech), Lingfei Wu, Feng Chen (UT Dallas), Jian Pei (Duke),\n  Jieping Ye (Alibaba), Charu Aggarwal (IBM), Hui Xiong, and Lise Getoor (UC Santa\n  Cruz), among others.\n- **Newest directions (2026):** \"world models\" for science and decision-making —\n  e.g. *Toward World Models for Epidemiology*, *MolWorld* for molecular optimization,\n  and *LUMINA*, a grid foundation model for power-system optimization — plus\n  self-maintaining **LLM-agent skill libraries** (*SkillOps*).",
  "chunk_id": 5
 },
 {
  "doc_id": "profile",
  "title": "Profile of Prof. Liang Zhao — Selected awards & recognition",
  "source": "profile",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## Selected awards & recognition\n- **2025**: KDD Test of Time Award; CIKM Best Paper Runner-up.\n- **2024**: Stanford \"Top 2% Scientists\" ranking.\n- **2023**: IEEE Middle-Career Award.\n- **2022**: ICDM Best Paper Award; Meta Research Award.\n- **2020**: NSF CAREER Award; Amazon Research Award.",
  "chunk_id": 6
 },
 {
  "doc_id": "profile",
  "title": "Profile of Prof. Liang Zhao — Current Ph.D. students and their topics",
  "source": "profile",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## Current Ph.D. students and their topics\n- **Dazhou Yu** (2021–): 3D geometry, spatial reasoning, polyhedra representation learning (PolyhedronNet; Spatial-RAG).\n- **Yifei Zhang** (2022–): visual explanations, saliency analysis, active learning (Saliency-Bench).\n- **Bo Pan** (2022–): graph neural networks, text-attributed graphs, GNN explanations (GraphNarrator).\n- **Mengdan Zhu** (2023–): latent representation explanation, generative models (LatentExplainer).\n- **Yuntong Hu** (2023–): RAG systems, citation networks, memory-augmented LLMs (GRAG, CG-RAG, RAG-without-Forgetting).\n- **Zeeshan Memon** (2024–): information propagation, diffusion networks (propagation-tree identification).\n- **Yang Qiao** (2024–).\n- **Ziyuan Qin** (2024–, co-advised): biomedical informatics applications.\n- **Xinyuan Song** (2024–).\n- **Riyang Bao** (2025–): spatial reasoning, agentic systems (Spatial-Agent).\n- **Yuzhu Mao** (2025–).\n- **Ziyang Yu** (2025–).\n- **Yuxiang Lai** (2026–, co-advised): biomedical applications.\n- **Ken Su** (2026–).",
  "chunk_id": 7
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — Frequently Asked Questions — Prospective Students & Collaborators",
  "source": "faq",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "# Frequently Asked Questions — Prospective Students & Collaborators",
  "chunk_id": 8
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — Is Prof. Zhao taking new Ph.D. students?",
  "source": "faq",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## Is Prof. Zhao taking new Ph.D. students?\nHis group grows most years (students joined in 2024, 2025, and 2026), so he\nregularly considers strong applicants. Admission to work with him is through the\n**Emory University Computer Science Ph.D. program** — you apply to the department\nand indicate interest in his areas. Prospective students who already understand\nhis research (graph learning, spatio-temporal/geospatial ML, RAG/agentic LLMs,\nAI-for-science) and can point to relevant projects or coursework tend to stand out.",
  "chunk_id": 9
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — What does he look for in students?",
  "source": "faq",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## What does he look for in students?\nStrong fundamentals in machine learning and mathematics, the ability to \"abstract\nthe core problem into mathematical language,\" solid programming, and genuine\ncuriosity about turning a scientific or real-world problem into a learning problem.\nEvidence of initiative — a paper, an open-source project, a thoughtful re-implementation\nof recent work — speaks louder than a résumé line.",
  "chunk_id": 10
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — How should I reach out?",
  "source": "faq",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## How should I reach out?\nEmail is best. A good cold email is short and specific: name the paper or research\ndirection of his you care about, say in one or two sentences how your background\nconnects, and attach a CV. Generic mass emails rarely land. You can also leave your\ndetails with this assistant (below) and they will be passed along.",
  "chunk_id": 11
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — Can he supervise undergraduate / master's research or internships?",
  "source": "faq",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## Can he supervise undergraduate / master's research or internships?\nHe works with motivated students at multiple levels through Emory. The same advice\napplies: come in having read some of the lab's work and with a concrete interest.",
  "chunk_id": 12
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — I'd like to collaborate (academia or industry). Is that possible?",
  "source": "faq",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## I'd like to collaborate (academia or industry). Is that possible?\nYes — his work spans CS, biomedical informatics, public health, and the physical\nsciences, and is supported by industry partners (Amazon, Meta, NVIDIA). Describe the\nproblem, the data, and what a successful collaboration would look like.",
  "chunk_id": 13
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — Where can I learn more about the research?",
  "source": "faq",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## Where can I learn more about the research?\n- Homepage: https://cs.emory.edu/~lzhao41/\n- Google Scholar: https://scholar.google.com/citations?user=qnvyqtwAAAAJ&hl=en\nAsk this assistant about a specific topic (e.g. \"graph RAG\", \"Spatial-RAG\",\n\"GNN explainability\", \"spatial event forecasting\") for a cited summary.",
  "chunk_id": 14
 },
 {
  "doc_id": "faq",
  "title": "Prospective Students & Collaborators FAQ — Disclaimer",
  "source": "faq",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## Disclaimer\nThis is an AI assistant prototype that answers questions about the lab's public\nresearch. Responses are generated automatically and may be incomplete or imperfect;\nthey do not constitute official statements from Prof. Zhao, the lab, or Emory\nUniversity. Always confirm important details (deadlines, admissions, funding) through\nofficial channels.",
  "chunk_id": 15
 },
 {
  "doc_id": "tools",
  "title": "Lab Tools & Systems — Lab Tools & Systems",
  "source": "tools",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "# Lab Tools & Systems",
  "chunk_id": 16
 },
 {
  "doc_id": "tools",
  "title": "Lab Tools & Systems — ResearchAtlas — graph-based literature discovery",
  "source": "tools",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## ResearchAtlas — graph-based literature discovery\nA system that helps researchers map a field rather than just rank hits. Instead of\nreturning a ranked list like a standard search engine, ResearchAtlas is\n**coverage-oriented**: it builds a graph over the literature so a user can see the\nstructure of a research area, discover related work, and navigate connections\nbetween papers and ideas. It reflects the lab's broader thesis that **graph structure\nis a first-class signal** for understanding and retrieval.\nLink: https://www.researchatlas.dpdns.org/",
  "chunk_id": 17
 },
 {
  "doc_id": "tools",
  "title": "Lab Tools & Systems — LLMCostCut — cutting LLM API cost by up to 10×",
  "source": "tools",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## LLMCostCut — cutting LLM API cost by up to 10×\nA method/toolkit that reduces the number of expensive large-LLM API calls **by up to\n10×** through **selective invocation** (only calling the large model when it is\nactually needed) combined with **online distillation** (learning from the large\nmodel's outputs so a cheaper model can handle more cases over time). This is the\nresearch basis for the **cost-aware model routing** used inside this very assistant:\ncheap questions are answered by a small/fast model and only substantive research\nquestions escalate to a stronger model.\nLink: https://github.com/zhaoliangvaio/llmcostcut",
  "chunk_id": 18
 },
 {
  "doc_id": "tools",
  "title": "Lab Tools & Systems — How this assistant embodies the lab's research",
  "source": "tools",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## How this assistant embodies the lab's research\nThis lab assistant is itself a small demonstration of several of the lab's lines of\nwork:\n- **Retrieval-augmented generation (RAG)** over the lab's publications — see *GRAG*\n  and *Spatial-RAG*.\n- **Citation-/graph-aware retrieval**: after vector search, results are expanded one\n  hop along a paper graph (shared authors and topics), echoing *GRAG* and *CG-RAG*.\n- **Trustworthy generation**: answers are grounded in retrieved sources with inline\n  citations, and the assistant declines to answer when the corpus does not support a\n  claim — reflecting the lab's emphasis on robustness and uncertainty.\n- **Cost-aware routing**: a direct nod to *LLMCostCut*.",
  "chunk_id": 19
 },
 {
  "doc_id": "tools",
  "title": "Lab Tools & Systems — Note on \"Causal Dynamics Lab\" / Cielara",
  "source": "tools",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [],
  "authors": [],
  "text": "## Note on \"Causal Dynamics Lab\" / Cielara\nSeparately, Dr. Zhao is involved as a research scientist with the **Causal Dynamics\nLab**, an AI lab whose product **Cielara** builds a \"production world model\" to\nsimulate code changes against live environments before deployment. This is an\nindustry effort distinct from his Emory academic group; for academic research,\nadvising, and publications, refer to the Emory lab.",
  "chunk_id": 20
 },
 {
  "doc_id": "spatial-rag",
  "title": "Spatial-RAG: Spatial Retrieval Augmented Generation for Real-World Spatial Reasoning",
  "source": "publication",
  "url": "https://arxiv.org/abs/2502.18470",
  "topics": [
   "rag",
   "spatial",
   "llm"
  ],
  "authors": [
   "Dazhou Yu",
   "Riyang Bao",
   "Liang Zhao"
  ],
  "text": "Spatial-RAG: Spatial Retrieval Augmented Generation for Real-World Spatial Reasoning (ACL 2026). Authors: Dazhou Yu, Riyang Bao, Liang Zhao. Answering real-world geospatial questions--such as finding restaurants along a travel route or amenities near a landmark--requires reasoning over both geographic relationships and semantic user intent. However, existing large language models (LLMs) lack spatial computing capabilities and access to up-to-date, ubiquitous real-world geospatial data, while traditional geospatial systems fall short in interpreting natural language. To bridge this gap, we introduce Spatial-RAG, a Retrieval-Augmented Generation (RAG) framework designed for geospatial question answering. Spatial-RAG integrates structured spatial databases with LLMs via a hybrid spatial retriever that combines sparse spatial filtering and dense semantic matching. It formulates the answering process as a multi-objective optimization over spatial and semantic relevance, identifying Pareto-optimal candidates and dynamically selecting the best response based on user intent. Experiments across multiple tourism and map-based QA datasets show that Spatial-RAG significantly improves accuracy, precision, and ranking performance over strong baselines.",
  "chunk_id": 21
 },
 {
  "doc_id": "grag",
  "title": "GRAG: Graph Retrieval-Augmented Generation",
  "source": "publication",
  "url": "https://arxiv.org/abs/2405.16506",
  "topics": [
   "rag",
   "graph",
   "llm"
  ],
  "authors": [
   "Yuntong Hu",
   "Zhihan Lei",
   "Zheng Zhang",
   "Liang Zhao"
  ],
  "text": "GRAG: Graph Retrieval-Augmented Generation (NAACL Findings 2025). Authors: Yuntong Hu, Zhihan Lei, Zheng Zhang, Liang Zhao. Naive Retrieval-Augmented Generation (RAG) focuses on individual documents during retrieval and, as a result, falls short in handling networked documents which are very popular in many applications such as citation graphs, social media, and knowledge graphs. To overcome this limitation, we introduce Graph Retrieval-Augmented Generation (GRAG), which tackles the fundamental challenges in retrieving textual subgraphs and integrating the joint textual and topological information into Large Language Models (LLMs) to enhance its generation. To enable efficient textual subgraph retrieval, we propose a novel divide-and-conquer strategy that retrieves the optimal subgraph structure in linear time. To achieve graph context-aware generation, incorporate textual graphs into LLMs through two complementary views-the text view and the graph view-enabling LLMs to more effectively comprehend and utilize the graph context. Extensive experiments on graph reasoning benchmarks demonstrate that in scenarios requiring multi-hop reasoning on textual graphs, our GRAG approach significantly outperforms current state-of-the-art RAG methods. Our datasets as well as codes of GRAG are available at https://github.com/HuieL/GRAG.",
  "chunk_id": 22
 },
 {
  "doc_id": "cg-rag",
  "title": "CG-RAG: Research Question Answering by Citation Graph Retrieval-Augmented Generation",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "rag",
   "graph",
   "llm"
  ],
  "authors": [
   "Yuntong Hu",
   "Liang Zhao"
  ],
  "text": "CG-RAG: Research Question Answering by Citation Graph Retrieval-Augmented Generation (SIGIR 2025). Authors: Yuntong Hu, Liang Zhao. Answers scholarly research questions by retrieving over a citation graph: it follows citation links between papers to assemble a connected, well-grounded context for the language model, improving answer faithfulness for literature questions.",
  "chunk_id": 23
 },
 {
  "doc_id": "rag-without-forgetting",
  "title": "RAG without Forgetting: Continual Query-Infused Key Memory for Retrieval-Augmented Generation",
  "source": "publication",
  "url": "https://arxiv.org/abs/2602.05152",
  "topics": [
   "rag",
   "llm",
   "memory"
  ],
  "authors": [
   "Yuntong Hu",
   "Sha Li",
   "Liang Zhao"
  ],
  "text": "RAG without Forgetting: Continual Query-Infused Key Memory for Retrieval-Augmented Generation (ICML 2026). Authors: Yuntong Hu, Sha Li, Liang Zhao. Retrieval-augmented generation (RAG) systems commonly improve robustness via query-time adaptations such as query expansion and iterative retrieval. While effective, these approaches are inherently stateless: adaptations are recomputed for each query and discarded thereafter, precluding cumulative learning and repeatedly incurring inference-time cost. Index-side approaches like key expansion introduce persistence but rely on offline preprocessing or heuristic updates that are weakly aligned with downstream task utility, leading to semantic drift and noise accumulation. We propose Evolving Retrieval Memory (ERM), a training-free framework that transforms transient query-time gains into persistent retrieval improvements. ERM updates the retrieval index through correctness-gated feedback, selectively attributes atomic expansion signals to the document keys they benefit, and progressively evolves keys via stable, norm-bounded updates. We show that query and key expansion are theoretically equivalent under standard similarity functions and prove convergence of ERM's selective updates, amortizing optimal query expansion into a stable index with zero inference-time overhead. Experiments on BEIR and BRIGHT across 13 domains demonstrate consistent gains in retrieval and generation, particularly on reasoning-intensive tasks, at native retrieval speed.",
  "chunk_id": 24
 },
 {
  "doc_id": "spatial-agent",
  "title": "Spatial-Agent: Agentic Geo-spatial Reasoning with Scientific Core Concepts",
  "source": "publication",
  "url": "https://arxiv.org/abs/2601.16965",
  "topics": [
   "agentic",
   "spatial",
   "llm"
  ],
  "authors": [
   "Riyang Bao",
   "Cheng Yang",
   "Liang Zhao"
  ],
  "text": "Spatial-Agent: Agentic Geo-spatial Reasoning with Scientific Core Concepts (ACL 2026). Authors: Riyang Bao, Cheng Yang, Liang Zhao. Geospatial reasoning is essential for real-world applications such as urban analytics, transportation planning, and disaster response. However, existing LLM-based agents often fail at genuine geospatial computation, relying instead on web search or pattern matching while hallucinating spatial relationships. We present Spatial-Agent, an AI agent grounded in foundational theories of spatial information science. Our approach formalizes geo-analytical question answering as a concept transformation problem, where natural-language questions are parsed into executable workflows represented as GeoFlow Graphs -- directed acyclic graphs with nodes corresponding to spatial concepts and edges representing transformations. Drawing on spatial information theory, Spatial-Agent extracts spatial concepts, assigns functional roles with principled ordering constraints, and composes transformation sequences through template-based generation. Extensive experiments on MapEval-API and MapQA benchmarks demonstrate that Spatial-Agent significantly outperforms existing baselines including ReAct and Reflexion, while producing interpretable and executable geospatial workflows.",
  "chunk_id": 25
 },
 {
  "doc_id": "domain-specialization-survey",
  "title": "Domain Specialization as the Key to Make Large Language Models Disruptive: A Comprehensive Survey",
  "source": "publication",
  "url": "https://arxiv.org/abs/2305.18703",
  "topics": [
   "llm",
   "survey"
  ],
  "authors": [
   "Chen Ling",
   "Xujiang Zhao",
   "Jiaying Lu",
   "Liang Zhao"
  ],
  "text": "Domain Specialization as the Key to Make Large Language Models Disruptive: A Comprehensive Survey (ACM Computing Surveys 2025). Authors: Chen Ling, Xujiang Zhao, Jiaying Lu, Liang Zhao. Large language models (LLMs) have significantly advanced the field of natural language processing (NLP), providing a highly useful, task-agnostic foundation for a wide range of applications. However, directly applying LLMs to solve sophisticated problems in specific domains meets many hurdles, caused by the heterogeneity of domain data, the sophistication of domain knowledge, the uniqueness of domain objectives, and the diversity of the constraints (e.g., various social norms, cultural conformity, religious beliefs, and ethical standards in the domain applications). Domain specification techniques are key to make large language models disruptive in many applications. Specifically, to solve these hurdles, there has been a notable increase in research and practices conducted in recent years on the domain specialization of LLMs. This emerging field of study, with its substantial potential for impact, necessitates a comprehensive and systematic review to better summarize and guide ongoing work in this area. In this article, we present a comprehensive survey on domain specification techniques for large language models, an emerging direction critical for large language model applications. First, we propose a systematic taxonomy that categorizes the LLM domain-specialization techniques based on the accessibility to LLMs and summarizes the framework for all the subcategories as well as their relations and differences to each other. Second, we present an extensive taxonomy of critical application domains that can benefit dramatically from specialized LLMs, discussing their practical significance and open challenges. Last, we offer our insights into the current research status and future trends in this area.",
  "chunk_id": 26
 },
 {
  "doc_id": "gnn-book",
  "title": "Graph Neural Networks: Foundations, Frontiers, and Applications",
  "source": "publication",
  "url": "https://graph-neural-networks.github.io/",
  "topics": [
   "gnn",
   "graph",
   "survey"
  ],
  "authors": [
   "Lingfei Wu",
   "Peng Cui",
   "Jian Pei",
   "Liang Zhao"
  ],
  "text": "Graph Neural Networks: Foundations, Frontiers, and Applications (Springer (book) 2022). Authors: Lingfei Wu, Peng Cui, Jian Pei, Liang Zhao. An edited reference book covering the foundations of graph neural networks, current research frontiers, and applications across domains; one of the most cited works in the GNN literature (1400+ citations).",
  "chunk_id": 27
 },
 {
  "doc_id": "transferable-deep-clustering",
  "title": "Transferable Deep Clustering Model",
  "source": "publication",
  "url": "https://arxiv.org/abs/2310.04946",
  "topics": [
   "clustering",
   "representation-learning"
  ],
  "authors": [
   "Zheng Zhang",
   "Liang Zhao"
  ],
  "text": "Transferable Deep Clustering Model (CIKM (Best Paper Runner-up) 2025). Authors: Zheng Zhang, Liang Zhao. Deep learning has shown remarkable success in the field of clustering recently. However, how to transfer a trained clustering model on a source domain to a target domain by leveraging the acquired knowledge to guide the clustering process remains challenging. Existing deep clustering methods often lack generalizability to new domains because they typically learn a group of fixed cluster centroids, which may not be optimal for the new domain distributions. In this paper, we propose a novel transferable deep clustering model that can automatically adapt the cluster centroids according to the distribution of data samples. Rather than learning a fixed set of centroids, our approach introduces a novel attention-based module that can adapt the centroids by measuring their relationship with samples. In addition, we theoretically show that our model is strictly more powerful than some classical clustering algorithms such as k-means or Gaussian Mixture Model (GMM). Experimental results on both synthetic and real-world datasets demonstrate the effectiveness and efficiency of our proposed transfer learning framework, which significantly improves the performance on target domain and reduces the computational cost.",
  "chunk_id": 28
 },
 {
  "doc_id": "taga",
  "title": "TAGA: Text-Attributed Graph Self-Supervised Learning by Synergizing Graph and Text Mutual Transformations",
  "source": "publication",
  "url": "https://arxiv.org/abs/2405.16800",
  "topics": [
   "graph",
   "gnn",
   "llm",
   "representation-learning"
  ],
  "authors": [
   "Zheng Zhang",
   "Yuntong Hu",
   "Liang Zhao"
  ],
  "text": "TAGA: Text-Attributed Graph Self-Supervised Learning by Synergizing Graph and Text Mutual Transformations (CIKM 2025). Authors: Zheng Zhang, Yuntong Hu, Liang Zhao. Text-Attributed Graphs (TAGs) enhance graph structures with natural language descriptions, enabling detailed representation of data and their relationships across a broad spectrum of real-world scenarios. Despite the potential for deeper insights, existing TAG representation learning primarily relies on supervised methods, necessitating extensive labeled data and limiting applicability across diverse contexts. This paper introduces a new self-supervised learning framework, Text-And-Graph Multi-View Alignment (TAGA), which overcomes these constraints by integrating TAGs' structural and semantic dimensions. TAGA constructs two complementary views: Text-of-Graph view, which organizes node texts into structured documents based on graph topology, and the Graph-of-Text view, which converts textual nodes and connections into graph data. By aligning representations from both views, TAGA captures joint textual and structural information. In addition, a novel structure-preserving random walk algorithm is proposed for efficient training on large-sized TAGs. Our framework demonstrates strong performance in zero-shot and few-shot scenarios across eight real-world datasets.",
  "chunk_id": 29
 },
 {
  "doc_id": "graphnarrator",
  "title": "GraphNarrator: Generating Textual Explanations for Graph Neural Networks",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "gnn",
   "explainability",
   "graph"
  ],
  "authors": [
   "Bo Pan",
   "Zhen Xiong",
   "Liang Zhao"
  ],
  "text": "GraphNarrator: Generating Textual Explanations for Graph Neural Networks (ACL 2025). Authors: Bo Pan, Zhen Xiong, Liang Zhao. Produces natural-language explanations of GNN predictions, translating structural evidence behind a graph model's decision into human-readable narratives.",
  "chunk_id": 30
 },
 {
  "doc_id": "latentexplainer",
  "title": "LatentExplainer: Explaining Latent Representations of Generative Models with Multimodal LLMs",
  "source": "publication",
  "url": "https://arxiv.org/abs/2406.14862",
  "topics": [
   "explainability",
   "generative",
   "llm"
  ],
  "authors": [
   "Mengdan Zhu",
   "Liang Zhao"
  ],
  "text": "LatentExplainer: Explaining Latent Representations of Generative Models with Multimodal LLMs (CIKM 2025). Authors: Mengdan Zhu, Liang Zhao. Deep generative models like VAEs and diffusion models have advanced various generation tasks by leveraging latent variables to learn data distributions and generate high-quality samples. Despite the field of explainable AI making strides in interpreting machine learning models, understanding latent variables in generative models remains challenging. This paper introduces LatentExplainer, a framework for automatically generating semantically meaningful explanations of latent variables in deep generative models. LatentExplainer tackles three main challenges: inferring the meaning of latent variables, aligning explanations with inductive biases, and handling varying degrees of explainability. Our approach perturbs latent variables, interprets changes in generated data, and uses multimodal large language models (MLLMs) to produce human-understandable explanations. We evaluate our proposed method on several real-world and synthetic datasets, and the results demonstrate superior performance in generating high-quality explanations for latent variables. The results highlight the effectiveness of incorporating inductive biases and uncertainty quantification, significantly enhancing model interpretability.",
  "chunk_id": 31
 },
 {
  "doc_id": "polyhedronnet",
  "title": "PolyhedronNet: Representation Learning for Polyhedra with Surface-attributed Graph",
  "source": "publication",
  "url": "https://arxiv.org/abs/2502.01814",
  "topics": [
   "spatial",
   "graph",
   "representation-learning"
  ],
  "authors": [
   "Dazhou Yu",
   "Genpei Zhang",
   "Liang Zhao"
  ],
  "text": "PolyhedronNet: Representation Learning for Polyhedra with Surface-attributed Graph (ICLR 2025). Authors: Dazhou Yu, Genpei Zhang, Liang Zhao. Ubiquitous geometric objects can be precisely and efficiently represented as polyhedra. The transformation of a polyhedron into a vector, known as polyhedra representation learning, is crucial for manipulating these shapes with mathematical and statistical tools for tasks like classification, clustering, and generation. Recent years have witnessed significant strides in this domain, yet most efforts focus on the vertex sequence of a polyhedron, neglecting the complex surface modeling crucial in real-world polyhedral objects. This study proposes \\textbf{PolyhedronNet}, a general framework tailored for learning representations of 3D polyhedral objects. We propose the concept of the surface-attributed graph to seamlessly model the vertices, edges, faces, and their geometric interrelationships within a polyhedron. To effectively learn the representation of the entire surface-attributed graph, we first propose to break it down into local rigid representations to effectively learn each local region's relative positions against the remaining regions without geometric information loss. Subsequently, we propose PolyhedronGNN to hierarchically aggregate the local rigid representation via intra-face and inter-face geometric message passing modules, to obtain a global representation that minimizes information loss while maintaining rotation and translation invariance. Our experimental evaluations on four distinct datasets, encompassing both classification and retrieval tasks, substantiate PolyhedronNet's efficacy in capturing comprehensive and informative representations of 3D polyhedral objects. Code and data are available at {https://github.com/dyu62/3D_polyhedron}.",
  "chunk_id": 32
 },
 {
  "doc_id": "fedspallm",
  "title": "FedSpaLLM: Federated Pruning of Large Language Models",
  "source": "publication",
  "url": "https://arxiv.org/abs/2410.14852",
  "topics": [
   "llm",
   "federated",
   "efficiency"
  ],
  "authors": [
   "Guangji Bai",
   "Liang Zhao"
  ],
  "text": "FedSpaLLM: Federated Pruning of Large Language Models (NAACL Findings 2025). Authors: Guangji Bai, Liang Zhao. Large Language Models (LLMs) achieve state-of-the-art performance but are challenging to deploy due to their high computational and storage demands. Pruning can reduce model size, yet existing methods assume public access to calibration data, which is impractical for privacy-sensitive applications. To address the challenge of pruning LLMs in privacy-preserving settings, we propose FedSpaLLM, the first federated learning framework designed specifically for pruning LLMs. FedSpaLLM enables clients to prune their models locally based on private data while accounting for system heterogeneity and maintaining communication efficiency. Our framework introduces several key innovations: (1) a novel $\\ell_0$-norm aggregation function that ensures only non-zero weights are averaged across clients, preserving important model parameters; (2) an adaptive mask expansion technique that meets global sparsity targets while accommodating client-specific pruning decisions; and (3) a layer sampling strategy that reduces communication overhead and personalizes the pruning process based on client resources. Extensive experiments show that FedSpaLLM improves pruning performance in diverse federated settings.",
  "chunk_id": 33
 },
 {
  "doc_id": "saliency-bench",
  "title": "Saliency-Bench: A Comprehensive Benchmark for Evaluating Visual Explanations",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "explainability",
   "benchmark",
   "vision"
  ],
  "authors": [
   "Yifei Zhang",
   "James Song",
   "Liang Zhao"
  ],
  "text": "Saliency-Bench: A Comprehensive Benchmark for Evaluating Visual Explanations (KDD 2025). Authors: Yifei Zhang, James Song, Liang Zhao. A benchmark suite for systematically evaluating saliency-based visual explanation methods across datasets and metrics.",
  "chunk_id": 34
 },
 {
  "doc_id": "propagation-trees",
  "title": "Deep Identification of Propagation Trees",
  "source": "publication",
  "url": "https://arxiv.org/abs/2503.00646",
  "topics": [
   "forecasting",
   "graph",
   "diffusion"
  ],
  "authors": [
   "Zeeshan Memon",
   "Chen Ling",
   "Liang Zhao"
  ],
  "text": "Deep Identification of Propagation Trees (IJCAI 2026). Authors: Zeeshan Memon, Chen Ling, Liang Zhao. Understanding propagation structures in graph diffusion processes, such as epidemic spread or misinformation diffusion, is a fundamental yet challenging problem. While existing methods primarily focus on source localization, they cannot reconstruct the underlying propagation trees i.e., \"who infected whom\", which are substantial for tracking the propagation pathways and investigate diffusion mechanisms. In this work, we propose Deep Identification of Propagation Trees (DIPT), a probabilistic framework that infers propagation trees from observed diffused states. DIPT models local influence strengths between nodes and leverages an alternating optimization strategy to jointly learn the diffusion mechanism and reconstruct the propagation structure. Extensive experiments on five real-world datasets demonstrate the effectiveness of DIPT in accurately reconstructing propagation trees.",
  "chunk_id": 35
 },
 {
  "doc_id": "continuous-domain-generalization",
  "title": "Continuous Domain Generalization",
  "source": "publication",
  "url": "https://arxiv.org/abs/2505.13519",
  "topics": [
   "domain-generalization",
   "representation-learning"
  ],
  "authors": [
   "Zekun Cai",
   "Yiheng Yao",
   "Liang Zhao"
  ],
  "text": "Continuous Domain Generalization (NeurIPS 2025). Authors: Zekun Cai, Yiheng Yao, Liang Zhao. Real-world data distributions often shift continuously across multiple latent factors such as time, geography, and socioeconomic contexts. However, existing domain generalization approaches typically treat domains as discrete or as evolving along a single axis (e.g., time). This oversimplification fails to capture the complex, multidimensional nature of real-world variation. This paper introduces the task of Continuous Domain Generalization (CDG), which aims to generalize predictive models to unseen domains defined by arbitrary combinations of continuous variations. We present a principled framework grounded in geometric and algebraic theories, showing that optimal model parameters across domains lie on a low-dimensional manifold. To model this structure, we propose a Neural Lie Transport Operator (NeuralLio), which enables structure-preserving parameter transitions by enforcing geometric continuity and algebraic consistency. To handle noisy or incomplete domain variation descriptors, we introduce a gating mechanism to suppress irrelevant dimensions and a local chart-based strategy for robust generalization. Extensive experiments on synthetic and real-world datasets, including remote sensing, scientific documents, and traffic forecasting, demonstrate that our method significantly outperforms existing baselines in both generalization accuracy and robustness.",
  "chunk_id": 36
 },
 {
  "doc_id": "healthcare-kg-review",
  "title": "A Review on Knowledge Graphs for Healthcare",
  "source": "publication",
  "url": "https://arxiv.org/abs/2306.04802",
  "topics": [
   "graph",
   "scientific-ml",
   "survey",
   "healthcare"
  ],
  "authors": [
   "Hejie Cui",
   "Carl Yang",
   "Liang Zhao"
  ],
  "text": "A Review on Knowledge Graphs for Healthcare (Journal of Biomedical Informatics 2025). Authors: Hejie Cui, Carl Yang, Liang Zhao. This comprehensive review aims to provide an overview of the current state of Healthcare Knowledge Graphs (HKGs), including their construction, utilization models, and applications across various healthcare and biomedical research domains. We thoroughly analyzed existing literature on HKGs, covering their construction methodologies, utilization techniques, and applications in basic science research, pharmaceutical research and development, clinical decision support, and public health. The review encompasses both model-free and model-based utilization approaches and the integration of HKGs with large language models (LLMs). We searched Google Scholar for relevant papers on HKGs and classified them into the following topics: HKG construction, HKG utilization, and their downstream applications in various domains. We also discussed their special challenges and the promise for future work. The review highlights the potential of HKGs to significantly impact biomedical research and clinical practice by integrating vast amounts of biomedical knowledge from multiple domains. The synergy between HKGs and LLMs offers promising opportunities for constructing more comprehensive knowledge graphs and improving the accuracy of healthcare applications. HKGs have emerged as a powerful tool for structuring medical knowledge, with broad applications across biomedical research, clinical decision-making, and public health. This survey serves as a roadmap for future research and development in the field of HKGs, highlighting the potential of combining knowledge graphs with advanced machine learning models for healthcare transformation.",
  "chunk_id": 37
 },
 {
  "doc_id": "embers",
  "title": "Beating the News with EMBERS: Forecasting Civil Unrest using Open Source Indicators",
  "source": "publication",
  "url": "https://arxiv.org/abs/1402.7035",
  "topics": [
   "forecasting",
   "spatial",
   "scientific-ml"
  ],
  "authors": [
   "Naren Ramakrishnan",
   "Liang Zhao"
  ],
  "text": "Beating the News with EMBERS: Forecasting Civil Unrest using Open Source Indicators (KDD 2014). Authors: Naren Ramakrishnan, Liang Zhao. We describe the design, implementation, and evaluation of EMBERS, an automated, 24x7 continuous system for forecasting civil unrest across 10 countries of Latin America using open source indicators such as tweets, news sources, blogs, economic indicators, and other data sources. Unlike retrospective studies, EMBERS has been making forecasts into the future since Nov 2012 which have been (and continue to be) evaluated by an independent T&E team (MITRE). Of note, EMBERS has successfully forecast the uptick and downtick of incidents during the June 2013 protests in Brazil. We outline the system architecture of EMBERS, individual models that leverage specific data sources, and a fusion and suppression engine that supports trading off specific evaluation criteria. EMBERS also provides an audit trail interface that enables the investigation of why specific predictions were made along with the data utilized for forecasting. Through numerous evaluations, we demonstrate the superiority of EMBERS over baserate methods and its capability to forecast significant societal happenings.",
  "chunk_id": 38
 },
 {
  "doc_id": "influence-maximization",
  "title": "Deep Graph Representation Learning and Optimization for Influence Maximization",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "graph",
   "gnn",
   "optimization",
   "diffusion"
  ],
  "authors": [
   "Chen Ling",
   "Junji Jiang",
   "Junxiang Wang",
   "Liang Zhao"
  ],
  "text": "Deep Graph Representation Learning and Optimization for Influence Maximization (ICML 2023). Authors: Chen Ling, Junji Jiang, Junxiang Wang, Liang Zhao. Learns graph representations to solve influence maximization — choosing a seed set of nodes to maximize spread — with a differentiable, learning-based approach.",
  "chunk_id": 39
 },
 {
  "doc_id": "generative-graph-survey",
  "title": "A Systematic Survey on Deep Generative Models for Graph Generation",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "graph",
   "generative",
   "survey"
  ],
  "authors": [
   "Xiaojie Guo",
   "Liang Zhao"
  ],
  "text": "A Systematic Survey on Deep Generative Models for Graph Generation (IEEE TPAMI 2022). Authors: Xiaojie Guo, Liang Zhao. A systematic survey of deep generative models for generating graphs, organizing methods by problem setting, model family, and application.",
  "chunk_id": 40
 },
 {
  "doc_id": "event-prediction-survey",
  "title": "Event Prediction in the Big Data Era: A Systematic Survey",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "forecasting",
   "survey",
   "spatial"
  ],
  "authors": [
   "Liang Zhao"
  ],
  "text": "Event Prediction in the Big Data Era: A Systematic Survey (ACM Computing Surveys 2021). Authors: Liang Zhao. Surveys methods for predicting future events from large-scale data, spanning problem formulations, data, models, and evaluation across domains.",
  "chunk_id": 41
 },
 {
  "doc_id": "resource-efficient-llm-survey",
  "title": "Beyond Efficiency: A Systematic Survey of Resource-Efficient Large Language Models",
  "source": "publication",
  "url": "https://arxiv.org/abs/2401.00625",
  "topics": [
   "llm",
   "efficiency",
   "survey"
  ],
  "authors": [
   "Guangji Bai",
   "Zheng Chai",
   "Chen Ling",
   "Liang Zhao"
  ],
  "text": "Beyond Efficiency: A Systematic Survey of Resource-Efficient Large Language Models (arXiv 2024). Authors: Guangji Bai, Zheng Chai, Chen Ling, Liang Zhao. The burgeoning field of Large Language Models (LLMs), exemplified by sophisticated models like OpenAI's ChatGPT, represents a significant advancement in artificial intelligence. These models, however, bring forth substantial challenges in the high consumption of computational, memory, energy, and financial resources, especially in environments with limited resource capabilities. This survey aims to systematically address these challenges by reviewing a broad spectrum of techniques designed to enhance the resource efficiency of LLMs. We categorize methods based on their optimization focus: computational, memory, energy, financial, and network resources and their applicability across various stages of an LLM's lifecycle, including architecture design, pretraining, finetuning, and system design. Additionally, the survey introduces a nuanced categorization of resource efficiency techniques by their specific resource types, which uncovers the intricate relationships and mappings between various resources and corresponding optimization techniques. A standardized set of evaluation metrics and datasets is also presented to facilitate consistent and fair comparisons across different models and techniques. By offering a comprehensive overview of the current sota and identifying open research avenues, this survey serves as a foundational reference for researchers and practitioners, aiding them in developing more sustainable and efficient LLMs in a rapidly evolving landscape.",
  "chunk_id": 42
 },
 {
  "doc_id": "uq-icl",
  "title": "Uncertainty Quantification for In-Context Learning of Large Language Models",
  "source": "publication",
  "url": "https://arxiv.org/abs/2402.10189",
  "topics": [
   "llm",
   "uncertainty",
   "trustworthy"
  ],
  "authors": [
   "Chen Ling",
   "Xujiang Zhao",
   "Wei Cheng",
   "Liang Zhao"
  ],
  "text": "Uncertainty Quantification for In-Context Learning of Large Language Models (NAACL 2024). Authors: Chen Ling, Xujiang Zhao, Wei Cheng, Liang Zhao. In-context learning has emerged as a groundbreaking ability of Large Language Models (LLMs) and revolutionized various fields by providing a few task-relevant demonstrations in the prompt. However, trustworthy issues with LLM's response, such as hallucination, have also been actively discussed. Existing works have been devoted to quantifying the uncertainty in LLM's response, but they often overlook the complex nature of LLMs and the uniqueness of in-context learning. In this work, we delve into the predictive uncertainty of LLMs associated with in-context learning, highlighting that such uncertainties may stem from both the provided demonstrations (aleatoric uncertainty) and ambiguities tied to the model's configurations (epistemic uncertainty). We propose a novel formulation and corresponding estimation method to quantify both types of uncertainties. The proposed method offers an unsupervised way to understand the prediction of in-context learning in a plug-and-play fashion. Extensive experiments are conducted to demonstrate the effectiveness of the decomposition. The code and data are available at: https://github.com/lingchen0331/UQ_ICL.",
  "chunk_id": 43
 },
 {
  "doc_id": "staleness-distributed-gnn",
  "title": "Staleness-Alleviated Distributed GNN Training",
  "source": "publication",
  "url": "https://arxiv.org/abs/2308.13466",
  "topics": [
   "gnn",
   "graph",
   "efficiency",
   "systems"
  ],
  "authors": [
   "Guangji Bai",
   "Liang Zhao"
  ],
  "text": "Staleness-Alleviated Distributed GNN Training (SDM 2025). Authors: Guangji Bai, Liang Zhao. Despite the recent success of Graph Neural Networks (GNNs), it remains challenging to train GNNs on large-scale graphs due to neighbor explosions. As a remedy, distributed computing becomes a promising solution by leveraging abundant computing resources (e.g., GPU). However, the node dependency of graph data increases the difficulty of achieving high concurrency in distributed GNN training, which suffers from the massive communication overhead. To address it, Historical value approximation is deemed a promising class of distributed training techniques. It utilizes an offline memory to cache historical information (e.g., node embedding) as an affordable approximation of the exact value and achieves high concurrency. However, such benefits come at the cost of involving dated training information, leading to staleness, imprecision, and convergence issues. To overcome these challenges, this paper proposes SAT (Staleness-Alleviated Training), a novel and scalable distributed GNN training framework that reduces the embedding staleness adaptively. The key idea of SAT is to model the GNN's embedding evolution as a temporal graph and build a model upon it to predict future embedding, which effectively alleviates the staleness of the cached historical embedding. We propose an online algorithm to train the embedding predictor and the distributed GNN alternatively and further provide a convergence analysis. Empirically, we demonstrate that SAT can effectively reduce embedding staleness and thus achieve better performance and convergence speed on multiple large-scale graph datasets.",
  "chunk_id": 44
 },
 {
  "doc_id": "spectral-temporal-gnn",
  "title": "Towards Expressive Spectral-Temporal Graph Neural Networks for Time Series Forecasting",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "gnn",
   "forecasting",
   "graph",
   "time-series"
  ],
  "authors": [
   "Ming Jin",
   "Liang Zhao"
  ],
  "text": "Towards Expressive Spectral-Temporal Graph Neural Networks for Time Series Forecasting (IEEE TPAMI 2025). Authors: Ming Jin, Liang Zhao. Develops more expressive spectral-temporal GNNs for multivariate time-series forecasting, jointly modeling relationships across variables and over time.",
  "chunk_id": 45
 },
 {
  "doc_id": "bridging-spatial-spectral",
  "title": "Bridging the Gap between Spatial and Spectral Domains: A Unified Framework for Graph Neural Networks",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "gnn",
   "graph",
   "survey"
  ],
  "authors": [
   "Zhiqian Chen",
   "Fanglan Chen",
   "Lei Zhang",
   "Liang Zhao"
  ],
  "text": "Bridging the Gap between Spatial and Spectral Domains: A Unified Framework for Graph Neural Networks (ACM Computing Surveys 2023). Authors: Zhiqian Chen, Fanglan Chen, Lei Zhang, Liang Zhao. Proposes a unified framework that connects spatial and spectral formulations of graph neural networks, clarifying how seemingly different GNNs relate.",
  "chunk_id": 46
 },
 {
  "doc_id": "admm-deep-learning",
  "title": "ADMM for Efficient Deep Learning with Global Convergence",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "optimization",
   "efficiency"
  ],
  "authors": [
   "Junxiang Wang",
   "Fuxun Yu",
   "Xiang Chen",
   "Liang Zhao"
  ],
  "text": "ADMM for Efficient Deep Learning with Global Convergence (KDD 2019). Authors: Junxiang Wang, Fuxun Yu, Xiang Chen, Liang Zhao. Trains deep networks with an Alternating Direction Method of Multipliers formulation that comes with global convergence guarantees, as an alternative to gradient descent.",
  "chunk_id": 47
 },
 {
  "doc_id": "implicit-gnn-transformation",
  "title": "Implicit Graph Neural Networks for Deep Graph Transformation",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "gnn",
   "graph",
   "generative"
  ],
  "authors": [
   "Xiaojie Guo",
   "Yuanqi Du",
   "Liang Zhao"
  ],
  "text": "Implicit Graph Neural Networks for Deep Graph Transformation (KAIS 2025). Authors: Xiaojie Guo, Yuanqi Du, Liang Zhao. Uses implicit GNNs to perform deep graph transformation — mapping an input graph to a target graph — for structured prediction over graphs.",
  "chunk_id": 48
 },
 {
  "doc_id": "sst-mamba-transformer",
  "title": "SST: Multi-Scale Hybrid Mamba-Transformer Experts for Long-Short Range Time Series Forecasting",
  "source": "publication",
  "url": "https://arxiv.org/abs/2404.14757",
  "topics": [
   "forecasting",
   "time-series"
  ],
  "authors": [
   "Xiongxiao Xu",
   "Liang Zhao"
  ],
  "text": "SST: Multi-Scale Hybrid Mamba-Transformer Experts for Long-Short Range Time Series Forecasting (CIKM 2025). Authors: Xiongxiao Xu, Liang Zhao. Time series forecasting has made significant advances, including with Transformer-based models. The attention mechanism in Transformer effectively captures temporal dependencies by attending to all past inputs simultaneously. However, its quadratic complexity with respect to sequence length limits the scalability for long-range modeling. Recent state space models (SSMs) such as Mamba offer a promising alternative by achieving linear complexity without attention. Yet, Mamba compresses historical information into a fixed-size latent state, potentially causing information loss and limiting representational effectiveness. This raises a key research question: Can we design a hybrid Mamba-Transformer architecture that is both effective and efficient for time series forecasting? To address it, we adapt a hybrid Mamba-Transformer architecture Mambaformer, originally proposed for language modeling, to the time series domain. Preliminary experiments reveal that naively stacking Mamba and Transformer layers in Mambaformer is suboptimal for time series forecasting, due to an information interference problem. To mitigate this issue, we introduce a new time series decomposition strategy that separates time series into long-range patterns and short-range variations. Then we show that Mamba excels at capturing long-term structures, while Transformer is more effective at modeling short-term dynamics. Building on this insight, we propose State Space Transformer (SST), a multi-scale hybrid model with expert modules: a Mamba expert for long-range patterns and a Transformer expert for short-term variations. SST also employs a multi-scale patching mechanism to adaptively adjust time series resolution: low resolution for long-term patterns and high resolution for short-term variations. Experiments show that SST obtains SOTA performance with linear scalability. The code is at https://github.com/XiongxiaoXu/SST.",
  "chunk_id": 49
 },
 {
  "doc_id": "fedat",
  "title": "FedAT: A High-Performance and Communication-Efficient Federated Learning System with Asynchronous Tiers",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "federated",
   "efficiency",
   "systems"
  ],
  "authors": [
   "Zheng Chai",
   "Yujing Chen",
   "Ali Anwar",
   "Liang Zhao"
  ],
  "text": "FedAT: A High-Performance and Communication-Efficient Federated Learning System with Asynchronous Tiers (SC / HiPC 2021). Authors: Zheng Chai, Yujing Chen, Ali Anwar, Liang Zhao. FedAT groups federated-learning clients into asynchronous tiers by speed, combining synchronous intra-tier and asynchronous cross-tier updates to improve convergence speed and accuracy while cutting communication cost.",
  "chunk_id": 50
 },
 {
  "doc_id": "misinformation-twitter",
  "title": "Misinformation Propagation in the Age of Twitter",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "diffusion",
   "forecasting",
   "scientific-ml"
  ],
  "authors": [
   "Fang Jin",
   "Wei Wang",
   "Liang Zhao"
  ],
  "text": "Misinformation Propagation in the Age of Twitter (IEEE Computer 2014). Authors: Fang Jin, Wei Wang, Liang Zhao. Characterizes how rumors and misinformation spread on Twitter during real-world events, analyzing propagation dynamics versus true information.",
  "chunk_id": 51
 },
 {
  "doc_id": "multitask-spatiotemporal-forecasting",
  "title": "Multi-task Learning for Spatio-temporal Event Forecasting",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "forecasting",
   "spatial"
  ],
  "authors": [
   "Liang Zhao",
   "Qian Sun",
   "Jieping Ye"
  ],
  "text": "Multi-task Learning for Spatio-temporal Event Forecasting (KDD 2015). Authors: Liang Zhao, Qian Sun, Jieping Ye. Forecasts spatio-temporal societal events by sharing statistical strength across related locations and event types through a multi-task learning formulation.",
  "chunk_id": 52
 },
 {
  "doc_id": "covid-spatiotemporal",
  "title": "Taking the Pulse of COVID-19: A Spatiotemporal Perspective",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "spatial",
   "forecasting",
   "scientific-ml",
   "healthcare"
  ],
  "authors": [
   "Chaowei Yang",
   "Dexuan Sha",
   "Qian Liu",
   "Liang Zhao"
  ],
  "text": "Taking the Pulse of COVID-19: A Spatiotemporal Perspective (International Journal of Digital Earth 2020). Authors: Chaowei Yang, Dexuan Sha, Qian Liu, Liang Zhao. A spatiotemporal analysis of the COVID-19 pandemic, examining how the outbreak and its drivers evolved across geography and time.",
  "chunk_id": 53
 },
 {
  "doc_id": "heterogeneous-temporal-gnn",
  "title": "Heterogeneous Temporal Graph Neural Network",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "gnn",
   "graph",
   "time-series"
  ],
  "authors": [
   "Yujie Fan",
   "Mingxuan Ju",
   "Chuxu Zhang",
   "Liang Zhao",
   "Yanfang Ye"
  ],
  "text": "Heterogeneous Temporal Graph Neural Network (SIAM SDM 2022). Authors: Yujie Fan, Mingxuan Ju, Chuxu Zhang, Liang Zhao, Yanfang Ye. A graph neural network for heterogeneous, time-evolving graphs that jointly models multiple node and edge types as they change over time.",
  "chunk_id": 54
 },
 {
  "doc_id": "explanation-guided-learning-survey",
  "title": "Going Beyond XAI: A Systematic Survey for Explanation-Guided Learning",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "explainability",
   "survey",
   "trustworthy"
  ],
  "authors": [
   "Yuyang Gao",
   "Siyi Gu",
   "Junji Jiang",
   "Liang Zhao"
  ],
  "text": "Going Beyond XAI: A Systematic Survey for Explanation-Guided Learning (ACM Computing Surveys 2024). Authors: Yuyang Gao, Siyi Gu, Junji Jiang, Liang Zhao. Surveys explanation-guided learning, where model explanations are fed back into training to improve models, moving beyond explaining models only after the fact.",
  "chunk_id": 55
 },
 {
  "doc_id": "deep-graph-translation",
  "title": "Deep Graph Translation",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "graph",
   "generative",
   "gnn"
  ],
  "authors": [
   "Xiaojie Guo",
   "Lingfei Wu",
   "Liang Zhao"
  ],
  "text": "Deep Graph Translation (IEEE TNNLS 2022). Authors: Xiaojie Guo, Lingfei Wu, Liang Zhao. Introduces deep graph translation: learning to transform an input graph into a target output graph, a generative structured-prediction task over graphs.",
  "chunk_id": 56
 },
 {
  "doc_id": "interpretable-deep-graph-generation",
  "title": "Interpretable Deep Graph Generation with Node-Edge Co-Disentanglement",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "graph",
   "generative",
   "explainability"
  ],
  "authors": [
   "Xiaojie Guo",
   "Liang Zhao"
  ],
  "text": "Interpretable Deep Graph Generation with Node-Edge Co-Disentanglement (KDD 2020). Authors: Xiaojie Guo, Liang Zhao. A deep generative model for graphs that disentangles node and edge factors, making the graph generation process more interpretable and controllable.",
  "chunk_id": 57
 },
 {
  "doc_id": "source-localization-vae",
  "title": "Source Localization of Graph Diffusion via Variational Autoencoders",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "diffusion",
   "graph",
   "generative"
  ],
  "authors": [
   "Chen Ling",
   "Junji Jiang",
   "Liang Zhao"
  ],
  "text": "Source Localization of Graph Diffusion via Variational Autoencoders (KDD 2022). Authors: Chen Ling, Junji Jiang, Liang Zhao. Uses a variational autoencoder (SL-VAE) to infer the origin of a diffusion process on a network from observed spread, an inverse problem on graphs.",
  "chunk_id": 58
 },
 {
  "doc_id": "temporal-domain-generalization",
  "title": "Temporal Domain Generalization with Drift-Aware Dynamic Neural Networks",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "domain-generalization",
   "representation-learning",
   "time-series"
  ],
  "authors": [
   "Guangji Bai",
   "Chen Ling",
   "Liang Zhao"
  ],
  "text": "Temporal Domain Generalization with Drift-Aware Dynamic Neural Networks (ICLR 2023). Authors: Guangji Bai, Chen Ling, Liang Zhao. Trains models that stay accurate as data distributions drift forward in time by making network parameters evolve with the temporal trend.",
  "chunk_id": 59
 },
 {
  "doc_id": "key-player-underground-forums",
  "title": "Key Player Identification in Underground Forums over Attributed Heterogeneous Information Network Embedding",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "graph",
   "security",
   "representation-learning"
  ],
  "authors": [
   "Yanfang Ye",
   "Yiming Zhang",
   "Yujie Fan",
   "Liang Zhao"
  ],
  "text": "Key Player Identification in Underground Forums over Attributed Heterogeneous Information Network Embedding (CIKM 2019). Authors: Yanfang Ye, Yiming Zhang, Yujie Fan, Liang Zhao. Identifies the most influential actors in cybercriminal underground forums by embedding an attributed heterogeneous information network of users, posts, and topics.",
  "chunk_id": 60
 },
 {
  "doc_id": "world-models-epidemiology",
  "title": "Toward World Models for Epidemiology",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "world-model",
   "scientific-ml",
   "epidemiology",
   "forecasting"
  ],
  "authors": [
   "Zeeshan Memon",
   "Yi Su",
   "Christian K. Thomas",
   "Walid Saad",
   "Liang Zhao",
   "Naren Ramakrishnan"
  ],
  "text": "Toward World Models for Epidemiology (arXiv 2026). Authors: Zeeshan Memon, Yi Su, Christian K. Thomas, Walid Saad, Liang Zhao, Naren Ramakrishnan. Proposes building learnable 'world models' of epidemic dynamics that can simulate disease spread and interventions, bridging mechanistic epidemiology and modern AI.",
  "chunk_id": 61
 },
 {
  "doc_id": "molworld",
  "title": "MolWorld: Molecule World Models for Actionable Molecular Optimization",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "world-model",
   "scientific-ml",
   "generative"
  ],
  "authors": [
   "Yang Qiao",
   "Bo Pan",
   "Liang Zhang",
   "Liang Zhao"
  ],
  "text": "MolWorld: Molecule World Models for Actionable Molecular Optimization (arXiv 2026). Authors: Yang Qiao, Bo Pan, Liang Zhang, Liang Zhao. Learns a world model of molecular space so that an agent can take actionable steps to optimize molecules toward desired properties, supporting AI-driven molecular design.",
  "chunk_id": 62
 },
 {
  "doc_id": "lumina-grid",
  "title": "LUMINA: A Grid Foundation Model for Benchmarking AC Optimal Power Flow Surrogate Learning",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "power-grid",
   "optimization",
   "scientific-ml",
   "benchmark"
  ],
  "authors": [
   "Hua Jin",
   "Kibaek Song",
   "Zeeshan Memon",
   "Liang Zhao",
   "Kibaek Kim"
  ],
  "text": "LUMINA: A Grid Foundation Model for Benchmarking AC Optimal Power Flow Surrogate Learning (arXiv 2026). Authors: Hua Jin, Kibaek Song, Zeeshan Memon, Liang Zhao, Kibaek Kim. A foundation model and benchmark for learning fast surrogates of AC optimal power flow, helping operate electrical grids more efficiently.",
  "chunk_id": 63
 },
 {
  "doc_id": "skillops",
  "title": "SkillOps: Managing LLM Agent Skill Libraries as Self-Maintaining Software Ecosystems",
  "source": "publication",
  "url": "https://cs.emory.edu/~lzhao41/",
  "topics": [
   "agentic",
   "llm"
  ],
  "authors": [
   "Haoyu Pu",
   "Xinyuan Song",
   "Liang Zhao"
  ],
  "text": "SkillOps: Managing LLM Agent Skill Libraries as Self-Maintaining Software Ecosystems (arXiv 2026). Authors: Haoyu Pu, Xinyuan Song, Liang Zhao. Treats the library of skills an LLM agent accumulates as a self-maintaining software ecosystem, with mechanisms to add, refactor, and retire skills over time.",
  "chunk_id": 64
 }
]