arXiv subjects

Zihao Li

Publications and source records attributed to Zihao Li.

At least 145 records · Page 8Linked to original sources

arXiv2025

Can Graph Neural Networks Learn Language with Extremely Weak Text Supervision?

While great success has been achieved in building vision models with Contrastive Language-Image Pre-training (CLIP) over internet-scale image-text pairs, building transferable Graph Neural Networks (GNNs) with CLIP pipeline is challenging because of the scarcity of labeled data and text supervision, different levels of downstream tasks, and the conceptual gaps between domains. In this work, to address these issues, we propose a multi-modal prompt learning paradigm to effectively adapt pre-trained GNN to downstream tasks and data, given only a few semantically labeled samples, each with extremely weak text supervision. Our new paradigm embeds the graphs directly in the same space as the Large Language Models (LLMs) by learning both graph prompts and text prompts simultaneously. We demonstrate the superior performance of our paradigm in few-shot, multi-task-level, and cross-domain settings. Moreover, we build the first CLIP-style zero-shot classification prototype that can generalize GNNs to unseen classes with extremely weak text supervision. The code is available at https://github.com/Violet24K/Morpher.

cs.LG↗

arXiv2025

Approximability Landscape of Welfare Maximization within Fair Allocations

Fair allocation of indivisible goods studies allocating $m$ goods among $n$ agents in a fair manner. While fairness is a fundamental requirement in many real-world applications, it often conflicts with (economic) efficiency. This raises a natural and important question: How can we identify the most welfare-efficient allocation among all fair allocations? This paper answers from the perspective of computational complexity. Specifically, we study the problem of maximizing utilitarian social welfare under two widely studied fairness criteria: envy-freeness up to any item (EFX) and envy-freeness up to one item (EF1). We examine both normalized and unnormalized valuations, where normalized valuations require that each agent's total utility for all items is identical. The key contributions of this paper can be summarized as follows: (i) we sketch the complete complexity landscape of welfare maximization subject to fair allocation constraints; and (ii) we provide interesting bounds on the price of fairness for both EFX and EF1. Specifically: (1) For $n=2$ agents, we develop polynomial-time approximation schemes (PTAS) and provide NP-hardness results for EFX and EF1 constraints; (2) For $n>2$ agents, under EFX constraints, we design algorithms that achieve approximation ratios of $O(n)$ and $O(\sqrt{n})$ for unnormalized and normalized valuations, respectively. These results are complemented by asymptotically tight inapproximability results. We also obtain similar results for EF1 constraints; (3) When the number of agents is a fixed constant, we show that the optimal solution can be computed in polynomial time by slightly relaxing the fairness constraints, whereas exact fairness leads to strong inapproximability; (4) Furthermore, our results imply the price of EFX is $Θ(\sqrt{n})$ for normalized valuations, which is unknown in the literature.

cs.GT↗

arXiv2025

FMNet: Frequency-Assisted Mamba-Like Linear Attention Network for Camouflaged Object Detection

Camouflaged Object Detection (COD) is challenging due to the strong similarity between camouflaged objects and their surroundings, which complicates identification. Existing methods mainly rely on spatial local features, failing to capture global information, while Transformers increase computational costs. To address this, the Frequency-Assisted Mamba-Like Linear Attention Network (FMNet) is proposed, which leverages frequency-domain learning to efficiently capture global features and mitigate ambiguity between objects and the background. FMNet introduces the Multi-Scale Frequency-Assisted Mamba-Like Linear Attention (MFM) module, integrating frequency and spatial features through a multi-scale structure to handle scale variations while reducing computational complexity. Additionally, the Pyramidal Frequency Attention Extraction (PFAE) module and the Frequency Reverse Decoder (FRD) enhance semantics and reconstruct features. Experimental results demonstrate that FMNet outperforms existing methods on multiple COD datasets, showcasing its advantages in both performance and efficiency. Code available at https://github.com/Chranos/FMNet.

cs.CV↗

arXiv2025

Model Unlearning via Sparse Autoencoder Subspace Guided Projections

Large language models (LLMs) store vast amounts of information, making them powerful yet raising privacy and safety concerns when selective knowledge removal is required. Existing unlearning strategies, ranging from gradient-based fine-tuning and model editing to sparse autoencoder (SAE) steering, either lack interpretability or fail to provide a robust defense against adversarial prompts. We propose SAE-Guided Subspace Projection Unlearning (SSPU), a novel framework that leverages SAE features to drive targeted updates in the model's parameter space, enabling precise, interpretable, and robust unlearning. SSPU's three-stage pipeline performs data-driven layer and feature selection, subspace construction via QR decomposition, and constrained optimization that controls activations into an "irrelevant" subspace while preserving retained knowledge. Overall, we use SAE features to construct a subspace that supervises unlearning, refining the loss and adding a regularization term to guide interpretable parameter updates. In experiments on the WMDP-Cyber forget set and three utility benchmarks (MMLU, TruthfulQA, GSM8K), SSPU reduces harmful knowledge accuracy by 3.22% compared to the strongest baseline. It also improves adversarial robustness, lowering malicious accuracy under jailbreak prompts compared to baselines. Our findings expose the limitations of prior unlearning methods and demonstrate how interpretable subspace-guided optimization can achieve robust, controllable model behavior.

cs.CL↗

arXiv2025

Simulating the Unseen: Crash Prediction Must Learn from What Did Not Happen

Traffic safety science has long been hindered by a fundamental data paradox: the crashes we most wish to prevent are precisely those events we rarely observe. Existing crash-frequency models and surrogate safety metrics rely heavily on sparse, noisy, and under-reported records, while even sophisticated, high-fidelity simulations undersample the long-tailed situations that trigger catastrophic outcomes such as fatalities. We argue that the path to achieving Vision Zero, i.e., the complete elimination of traffic fatalities and severe injuries, requires a paradigm shift from traditional crash-only learning to a new form of counterfactual safety learning: reasoning not only about what happened, but also about the vast set of plausible yet perilous scenarios that could have happened under slightly different circumstances. To operationalize this shift, our proposed agenda bridges macro to micro. Guided by crash-rate priors, generative scene engines, diverse driver models, and causal learning, near-miss events are synthesized and explained. A crash-focused digital twin testbed links micro scenes to macro patterns, while a multi-objective validator ensures that simulations maintain statistical realism. This pipeline transforms sparse crash data into rich signals for crash prediction, enabling the stress-testing of vehicles, roads, and policies before deployment. By learning from crashes that almost happened, we can shift traffic safety from reactive forensics to proactive prevention, advancing Vision Zero.

cs.LG↗

arXiv2025

Chain-of-Model Learning for Language Model

In this paper, we propose a novel learning paradigm, termed Chain-of-Model (CoM), which incorporates the causal relationship into the hidden states of each layer as a chain style, thereby introducing great scaling efficiency in model training and inference flexibility in deployment. We introduce the concept of Chain-of-Representation (CoR), which formulates the hidden states at each layer as a combination of multiple sub-representations (i.e., chains) at the hidden dimension level. In each layer, each chain from the output representations can only view all of its preceding chains in the input representations. Consequently, the model built upon CoM framework can progressively scale up the model size by increasing the chains based on the previous models (i.e., chains), and offer multiple sub-models at varying sizes for elastic inference by using different chain numbers. Based on this principle, we devise Chain-of-Language-Model (CoLM), which incorporates the idea of CoM into each layer of Transformer architecture. Based on CoLM, we further introduce CoLM-Air by introducing a KV sharing mechanism, that computes all keys and values within the first chain and then shares across all chains. This design demonstrates additional extensibility, such as enabling seamless LM switching, prefilling acceleration and so on. Experimental results demonstrate our CoLM family can achieve comparable performance to the standard Transformer, while simultaneously enabling greater flexiblity, such as progressive scaling to improve training efficiency and offer multiple varying model sizes for elastic inference, paving a a new way toward building language models. Our code will be released in the future at: https://github.com/microsoft/CoLM.

cs.CL↗

arXiv2025

Signal Timing Optimization for Mixed Connected Automated Traffic Based on A Markov Delay Approximation

Connected Automated Vehicles (CAVs) offer unparalleled opportunities to revolutionize existing transportation systems. In the near future, CAVs and human-driven vehicles (HDVs) are expected to coexist, forming a mixed traffic system. Although several prototype traffic signal systems leveraging CAVs have been developed, a simple yet realistic approximation of mixed traffic delay and optimal signal timing at intersections remains elusive. This paper presents an analytical approximation for delay and optimal cycle length at an isolated intersection of mixed traffic using a stochastic framework that combines Markov chain analysis, a car following model, and queuing theory. Given the intricate nature of mixed traffic delay, the proposed framework systematically incorporates the impacts of multiple factors, such as the distinct arrival and departure behaviors and headway characteristics of CAVs and HDVs, through mathematical derivations to ensure both realism and analytical tractability. Subsequently, closed-form expressions for intersection delay and optimal cycle length are derived. Numerical experiments are then conducted to validate the model and provide insights into the dynamics of mixed traffic delays at signalized intersections.

eess.SY↗

arXiv2025

Modeling Cascading Driver Interventions in Partially Automated Traffic: A Semi-Markov Chain Approach

This paper presents an analytical modeling framework for partially automated traffic, incorporating cascading driver intervention behaviors. In this framework, drivers of partially automated vehicles have the flexibility to switch driving modes (either AV or HDV) under lockout constraints. The cascading impact is captured by making the switching probability leader-dependent, highlighting the influence of the leading vehicle on mode choice and the potential propagation of mode changes throughout traffic. Due to the complexity of this system, traditional Markov-based methods are insufficient. To address this, the paper introduces an innovative semi-Markov chain framework with lockout constraints, ideally suited for modeling the system dynamics. This framework reformulates the system as a nonlinear model whose solution can be efficiently approximated using numerical methods from control theory, such as the Runge-Kutta algorithm. Moreover, the system is proven to be a piecewise affine bilinear system, with the existence of solutions and both local and global stability established via Brouwer's Fixed Point Theorem and the 1D Uncertainty Polytopes Theorem. Numerical experiments corroborate these theoretical findings, confirming the presence of cascading impacts and elucidating the influence of modeling parameters on traffic throughput, thereby deepening our understanding of the system's properties.

math.DS↗

arXiv2025

Recent Advances in Disaster Emergency Response Planning: Integrating Optimization, Machine Learning, and Simulation

The increasing frequency and severity of natural disasters underscore the critical importance of effective disaster emergency response planning to minimize human and economic losses. This survey provides a comprehensive review of recent advancements (2019--2024) in five essential areas of disaster emergency response planning: evacuation, facility location, casualty transport, search and rescue, and relief distribution. Research in these areas is systematically categorized based on methodologies, including optimization models, machine learning, and simulation, with a focus on their individual strengths and synergies. A notable contribution of this work is its examination of the interplay between machine learning, simulation, and optimization frameworks, highlighting how these approaches can address the dynamic, uncertain, and complex nature of disaster scenarios. By identifying key research trends and challenges, this study offers valuable insights to improve the effectiveness and resilience of emergency response strategies in future disaster planning efforts.

math.OC↗

arXiv2025

AI2-Active Safety: AI-enabled Interaction-aware Active Safety Analysis with Vehicle Dynamics

This paper introduces an AI-enabled, interaction-aware active safety analysis framework that accounts for groupwise vehicle interactions. Specifically, the framework employs a bicycle model-augmented with road gradient considerations-to accurately capture vehicle dynamics. In parallel, a hypergraph-based AI model is developed to predict probabilistic trajectories of ambient traffic. By integrating these two components, the framework derives vehicle intra-spacing over a 3D road surface as the solution of a stochastic ordinary differential equation, yielding high-fidelity surrogate safety measures such as time-to-collision (TTC). To demonstrate its effectiveness, the framework is analyzed using stochastic numerical methods comprising 4th-order Runge-Kutta integration and AI inference, generating probability-weighted high-fidelity TTC (HF-TTC) distributions that reflect complex multi-agent maneuvers and behavioral uncertainties. Evaluated with HF-TTC against traditional constant-velocity TTC and non-interaction-aware approaches on highway datasets, the proposed framework offers a systematic methodology for active safety analysis with enhanced potential for improving safety perception in complex traffic environments.

cs.RO↗

arXiv2025

Virtual Roads, Smarter Safety: A Digital Twin Framework for Mixed Autonomous Traffic Safety Analysis

This paper presents a digital-twin platform for active safety analysis in mixed traffic environments. The platform is built using a multi-modal data-enabled traffic environment constructed from drone-based aerial LiDAR, OpenStreetMap, and vehicle sensor data (e.g., GPS and inclinometer readings). High-resolution 3D road geometries are generated through AI-powered semantic segmentation and georeferencing of aerial LiDAR data. To simulate real-world driving scenarios, the platform integrates the CAR Learning to Act (CARLA) simulator, Simulation of Urban MObility (SUMO) traffic model, and NVIDIA PhysX vehicle dynamics engine. CARLA provides detailed micro-level sensor and perception data, while SUMO manages macro-level traffic flow. NVIDIA PhysX enables accurate modeling of vehicle behaviors under diverse conditions, accounting for mass distribution, tire friction, and center of mass. This integrated system supports high-fidelity simulations that capture the complex interactions between autonomous and conventional vehicles. Experimental results demonstrate the platform's ability to reproduce realistic vehicle dynamics and traffic scenarios, enhancing the analysis of active safety measures. Overall, the proposed framework advances traffic safety research by enabling in-depth, physics-informed evaluation of vehicle behavior in dynamic and heterogeneous traffic environments.

cs.RO↗

arXiv2025

ClimateBench-M: A Multi-Modal Climate Data Benchmark with a Simple Generative Method

Climate science studies the structure and dynamics of Earth's climate system and seeks to understand how climate changes over time, where the data is usually stored in the format of time series, recording the climate features, geolocation, time attributes, etc. Recently, much research attention has been paid to the climate benchmarks. In addition to the most common task of weather forecasting, several pioneering benchmark works are proposed for extending the modality, such as domain-specific applications like tropical cyclone intensity prediction and flash flood damage estimation, or climate statement and confidence level in the format of natural language. To further motivate the artificial general intelligence development for climate science, in this paper, we first contribute a multi-modal climate benchmark, i.e., ClimateBench-M, which aligns (1) the time series climate data from ERA5, (2) extreme weather events data from NOAA, and (3) satellite image data from NASA HLS based on a unified spatial-temporal granularity. Second, under each data modality, we also propose a simple but strong generative method that could produce competitive performance in weather forecasting, thunderstorm alerts, and crop segmentation tasks in the proposed ClimateBench-M. The data and code of ClimateBench-M are publicly available at https://github.com/iDEA-iSAIL-Lab-UIUC/ClimateBench-M.

cs.LG↗

arXiv2025

Large Language Models for Blockchain Security: A Systematic Literature Review

Large Language Models (LLMs) have emerged as powerful tools across various domains within cyber security. Notably, recent studies are increasingly exploring LLMs applied to the context of blockchain security (BS). However, there remains a gap in a comprehensive understanding regarding the full scope of applications, impacts, and potential constraints of LLMs on blockchain security. To fill this gap, we undertake a literature review focusing on the studies that apply LLMs in blockchain security (LLM4BS). Our study aims to comprehensively analyze and understand existing research, and elucidate how LLMs contribute to enhancing the security of blockchain systems. Through a thorough examination of existing literature, we delve into the integration of LLMs into various aspects of blockchain security. We explore the mechanisms through which LLMs can bolster blockchain security, including their applications in smart contract auditing, transaction anomaly detection, vulnerability repair, program analysis of smart contracts, and serving as participants in the cryptocurrency community. Furthermore, we assess the challenges and limitations associated with leveraging LLMs for enhancing blockchain security, considering factors such as scalability, privacy concerns, and ethical concerns. Our thorough review sheds light on the opportunities and potential risks of tasks on LLM4BS, providing valuable insights for researchers, practitioners, and policymakers alike.

cs.CR↗

arXiv2025

Medium-band Astrophysics with the Grism of NIRCam In Frontier fields (MAGNIF): Spectroscopic Census of H$α$ Luminosity Functions and Cosmic Star Formation at $z\sim 4.5$ and 6.3

We measure H$α$ luminosity functions (LFs) at redshifts $z \sim 4.5$ and 6.3 using the JWST MAGNIF (Medium-band Astrophysics with the Grism of NIRCam In Frontier fields) survey. MAGNIF obtained NIRCam grism spectra with the F360M and F480M filters in four Frontier Fields. We identify 248 H$α$ emitters based on the grism spectra and photometric redshifts from combined HST and JWST imaging data. The numbers of the H$α$ emitters show a large field-to-field variation, highlighting the necessity of multiple fields to mitigate cosmic variance. We calculate both observed and dust-corrected H$α$ LFs in the two redshift bins. Thanks to the gravitational lensing, the measured H$α$ LFs span three orders of magnitude in luminosity, and the faint-end luminosity reaches $L_{\mathrm{H}α} \sim 10^{40.3} \mathrm{erg} \mathrm{s}^{-1}$ at $z \sim 4.5$ and $10^{41.5} \mathrm{erg} \mathrm{s}^{-1}$ at $z \sim 6.3$, corresponding to star-formation rates (SFRs) of $\sim$ 0.1 and 1.7 $\mathrm{M}_\odot \mathrm{yr}^{-1}$. We conclude no or weak redshift evolution of the faint-end slope of H$α$ LF across $z\simeq0.4-6.3$, and the comparison with the faint-end slopes of UV LF indicates stochastic star formation history among low-mass H$α$ emitters. The derived cosmic SFR densities are $0.058^{+0.008}_{-0.006}\ \ M_\odot\ \mathrm{yr}^{-1}\ \mathrm{Mpc}^{-3}$ at $z \sim 4.5$ and $0.025^{+0.009}_{-0.007}\ \ M_\odot\ \mathrm{yr}^{-1}\ \mathrm{Mpc}^{-3}$ at $z \sim 6.3$. These are approximately 2.2 times higher than previous estimates based on dust-corrected UV LFs, but consistent with recent measurements from infrared surveys. We discuss uncertainties in the H$α$ LF measurements, including those propagate from the lens models, cosmic variance, and AGN contribution.

astro-ph.GA↗

arXiv2025

DBR: Divergence-Based Regularization for Debiasing Natural Language Understanding Models

Pre-trained language models (PLMs) have achieved impressive results on various natural language processing tasks. However, recent research has revealed that these models often rely on superficial features and shortcuts instead of developing a genuine understanding of language, especially for natural language understanding (NLU) tasks. Consequently, the models struggle to generalize to out-of-domain data. In this work, we propose Divergence Based Regularization (DBR) to mitigate this shortcut learning behavior. Our method measures the divergence between the output distributions for original examples and examples where shortcut tokens have been masked. This process prevents the model's predictions from being overly influenced by shortcut features or biases. We evaluate our model on three NLU tasks and find that it improves out-of-domain performance with little loss of in-domain accuracy. Our results demonstrate that reducing the reliance on shortcuts and superficial features can enhance the generalization ability of large pre-trained language models.

cs.CL↗

arXiv2025

One Model for All: Large Language Models are Domain-Agnostic Recommendation Systems

Sequential recommendation systems aim to predict users' next likely interaction based on their history. However, these systems face data sparsity and cold-start problems. Utilizing data from other domains, known as multi-domain methods, is useful for alleviating these problems. However, traditional multi-domain methods rely on meaningless ID-based item representation, which makes it difficult to align items with similar meanings from different domains, yielding sup-optimal knowledge transfer. This paper introduces LLM-Rec, a framework that utilizes pre-trained large language models (LLMs) for domain-agnostic recommendation. Specifically, we mix user's behaviors from multiple domains and concatenate item titles into a sentence, then use LLMs for generating user and item representations. By mixing behaviors across different domains, we can exploit the knowledge encoded in LLMs to bridge the semantic across over multi-domain behaviors, thus obtaining semantically rich representations and improving performance in all domains. Furthermore, we explore the underlying reasons why LLMs are effective and investigate whether LLMs can understand the semantic correlations as the recommendation model, and if advanced techniques like scaling laws in NLP also work in recommendations. We conduct extensive experiments with LLMs ranging from 40M to 6.7B to answer the above questions and to verify the effectiveness of LLM-Rec in multi-domain recommendation.

cs.IR↗

arXiv2025

What Do LLMs Need to Understand Graphs: A Survey of Parametric Representation of Graphs

Graphs, as a relational data structure, have been widely used for various application scenarios, like molecule design and recommender systems. Recently, large language models (LLMs) are reorganizing in the AI community for their expected reasoning and inference abilities. Making LLMs understand graph-based relational data has great potential, including but not limited to (1) distillate external knowledge base for eliminating hallucination and breaking the context window limit for LLMs' inference during the retrieval augmentation generation process; (2) taking graph data as the input and directly solve the graph-based research tasks like protein design and drug discovery. However, inputting the entire graph data to LLMs is not practical due to its complex topological structure, data size, and the lack of effective and efficient semantic graph representations. A natural question arises: Is there a kind of graph representation that can be described by natural language for LLM's understanding and is also easy to require to serve as the raw input for LLMs? Based on statistical computation, graph laws pre-define a set of parameters (e.g., degree, time, diameter) and identifie their relationships and values by observing the topological distribution of plenty of real-world graph data. We believe this kind of parametric representation of graphs, graph laws, can be a solution for making LLMs understand graph data as the input. In this survey, we first review the previous study of graph laws from multiple perspectives, i.e., macroscope and microscope of graphs, low-order and high-order graphs, static and dynamic graphs, different observation spaces, and newly proposed graph parameters. After we review various real-world applications benefiting from the guidance of graph laws, we conclude the paper with current challenges and future research directions.

cs.AI↗

arXiv2025

MaLei at the PLABA Track of TREC 2024: RoBERTa for Term Replacement -- LLaMA3.1 and GPT-4o for Complete Abstract Adaptation

This report is the system description of the MaLei team (Manchester and Leiden) for the shared task Plain Language Adaptation of Biomedical Abstracts (PLABA) 2024 (we had an earlier name BeeManc following last year), affiliated with TREC2024 (33rd Text REtrieval Conference https://ir.nist.gov/evalbase/conf/trec-2024). This report contains two sections corresponding to the two sub-tasks in PLABA-2024. In task one (term replacement), we applied fine-tuned ReBERTa-Base models to identify and classify the difficult terms, jargon, and acronyms in the biomedical abstracts and reported the F1 score (Task 1A and 1B). In task two (complete abstract adaptation), we leveraged Llamma3.1-70B-Instruct and GPT-4o with the one-shot prompts to complete the abstract adaptation and reported the scores in BLEU, SARI, BERTScore, LENS, and SALSA. From the official Evaluation from PLABA-2024 on Task 1A and 1B, our much smaller fine-tuned RoBERTa-Base model ranked 3rd and 2nd respectively on the two sub-tasks, and the 1st on averaged F1 scores across the two tasks from 9 evaluated systems. Our LLaMA-3.1-70B-instructed model achieved the highest Completeness score for Task 2. We share our source codes, fine-tuned models, and related resources at https://github.com/HECTA-UoM/PLABA2024

cs.CL↗