arXiv subjects

Zi Huang

Publications and source records attributed to Zi Huang.

At least 55 records · Page 3Linked to original sources

arXiv2025

Does Homophily Help in Robust Test-time Node Classification?

Homophily, the tendency of nodes from the same class to connect, is a fundamental property of real-world graphs, underpinning structural and semantic patterns in domains such as citation networks and social networks. Existing methods exploit homophily through designing homophily-aware GNN architectures or graph structure learning strategies, yet they primarily focus on GNN learning with training graphs. However, in real-world scenarios, test graphs often suffer from data quality issues and distribution shifts, such as domain shifts across users from different regions in social networks and temporal evolution shifts in citation network graphs collected over varying time periods. These factors significantly compromise the pre-trained model's robustness, resulting in degraded test-time performance. With empirical observations and theoretical analysis, we reveal that transforming the test graph structure by increasing homophily in homophilic graphs or decreasing it in heterophilic graphs can significantly improve the robustness and performance of pre-trained GNNs on node classifications, without requiring model training or update. Motivated by these insights, a novel test-time graph structural transformation method grounded in homophily, named GrapHoST, is proposed. Specifically, a homophily predictor is developed to discriminate test edges, facilitating adaptive test-time graph structural transformation by the confidence of predicted homophily scores. Extensive experiments on nine benchmark datasets under a range of test-time data quality issues demonstrate that GrapHoST consistently achieves state-of-the-art performance, with improvements of up to 10.92%. Our code has been released at https://github.com/YanJiangJerry/GrapHoST.

cs.LG↗

arXiv2025

Cluster-Aware Prompt Ensemble Learning for Few-Shot Vision-Language Model Adaptation

Vision-language models (VLMs) such as CLIP achieve zero-shot transfer across various tasks by pre-training on numerous image-text pairs. These models often benefit from using an ensemble of context prompts to represent a class. Despite being effective, conventional prompt ensembling that averages textual features of context prompts often yields suboptimal results. This is because feature averaging shifts the class centroids away from the true class distribution. To address this issue, we propose the Cluster-Aware Prompt Ensemble Learning (CAPEL) framework, which preserves the cluster nature of context prompts. CAPEL classifies images into one of several class clusters, each represented by a distinct prompt. Instead of ensembling prompts in the feature space, we perform ensembling in the classification logits space, aligning better with the visual feature distribution. To further optimize prompt fine-tuning while maintaining cluster-specific discriminative power, we introduce a cluster-preserving regularization term. This ensures that prompts remain distinct and specialized for different clusters, preventing collapse into a uniform direction. Additionally, we integrate an adaptive prompt weighting technique to dynamically adjust the attention weights for flawed or ambiguous prompts, ensuring robust performance across diverse datasets and tasks.

cs.CV↗

arXiv2025

ContextNav: Towards Agentic Multimodal In-Context Learning

Recent advances demonstrate that multimodal large language models (MLLMs) exhibit strong multimodal in-context learning (ICL) capabilities, enabling them to adapt to novel vision-language tasks from a few contextual examples. However, existing ICL approaches face challenges in reconciling scalability with robustness across diverse tasks and noisy contextual examples: manually selecting examples produces clean contexts but is labor-intensive and task-specific, while similarity-based retrieval improves scalability but could introduce irrelevant or structurally inconsistent samples that degrade ICL performance. To address these limitations, we propose ContextNav, the first agentic framework that integrates the scalability of automated retrieval with the quality and adaptiveness of human-like curation, enabling noise-robust and dynamically optimized contextualization for multimodal ICL. ContextNav unifies context management and noise-robust contextualization within a closed-loop workflow driven by graph-based orchestration. Specifically, it builds a resource-aware multimodal embedding pipeline, maintains a retrievable vector database, and applies agentic retrieval and structural alignment to construct noise-resilient contexts. An Operational Grammar Graph (OGG) further supports adaptive workflow planning and optimization, enabling the agent to refine its operational strategies based on downstream ICL feedback. Experimental results demonstrate that ContextNav achieves state-of-the-art performance across various datasets, underscoring the promise of agentic workflows for advancing scalable and robust contextualization in multimodal ICL.

cs.AI↗

arXiv2025

Augment to Segment: Tackling Pixel-Level Imbalance in Wheat Disease and Pest Segmentation

Accurate segmentation of foliar diseases and insect damage in wheat is crucial for effective crop management and disease control. However, the insect damage typically occupies only a tiny fraction of annotated pixels. This extreme pixel-level imbalance poses a significant challenge to the segmentation performance, which can result in overfitting to common classes and insufficient learning of rare classes, thereby impairing overall performance. In this paper, we propose a Random Projected Copy-and-Paste (RPCP) augmentation technique to address the pixel imbalance problem. Specifically, we extract rare insect-damage patches from annotated training images and apply random geometric transformations to simulate variations. The transformed patches are then pasted in appropriate regions while avoiding overlaps with lesions or existing damaged regions. In addition, we apply a random projection filter to the pasted regions, refining local features and ensuring a natural blend with the new background. Experiments show that our method substantially improves segmentation performance on the insect damage class, while maintaining or even slightly enhancing accuracy on other categories. Our results highlight the effectiveness of targeted augmentation in mitigating extreme pixel imbalance, offering a straightforward yet effective solution for agricultural segmentation problems.

cs.CV↗

arXiv2025

Text Meets Topology: Rethinking Out-of-distribution Detection in Text-Rich Networks

Out-of-distribution (OOD) detection remains challenging in text-rich networks, where textual features intertwine with topological structures. Existing methods primarily address label shifts or rudimentary domain-based splits, overlooking the intricate textual-structural diversity. For example, in social networks, where users represent nodes with textual features (name, bio) while edges indicate friendship status, OOD may stem from the distinct language patterns between bot and normal users. To address this gap, we introduce the TextTopoOOD framework for evaluating detection across diverse OOD scenarios: (1) attribute-level shifts via text augmentations and embedding perturbations; (2) structural shifts through edge rewiring and semantic connections; (3) thematically-guided label shifts; and (4) domain-based divisions. Furthermore, we propose TNT-OOD to model the complex interplay between Text aNd Topology using: 1) a novel cross-attention module to fuse local structure into node-level text representations, and 2) a HyperNetwork to generate node-specific transformation parameters. This aligns topological and semantic features of ID nodes, enhancing ID/OOD distinction across structural and textual shifts. Experiments on 11 datasets across four OOD scenarios demonstrate the nuanced challenge of TextTopoOOD for evaluating OOD detection in text-rich networks.

cs.CL↗

arXiv2025

ALSA: Anchors in Logit Space for Out-of-Distribution Accuracy Estimation

Estimating model accuracy on unseen, unlabeled datasets is crucial for real-world machine learning applications, especially under distribution shifts that can degrade performance. Existing methods often rely on predicted class probabilities (softmax scores) or data similarity metrics. While softmax-based approaches benefit from representing predictions on the standard simplex, compressing logits into probabilities leads to information loss. Meanwhile, similarity-based methods can be computationally expensive and domain-specific, limiting their broader applicability. In this paper, we introduce ALSA (Anchors in Logit Space for Accuracy estimation), a novel framework that preserves richer information by operating directly in the logit space. Building on theoretical insights and empirical observations, we demonstrate that the aggregation and distribution of logits exhibit a strong correlation with the predictive performance of the model. To exploit this property, ALSA employs an anchor-based modeling strategy: multiple learnable anchors are initialized in logit space, each assigned an influence function that captures subtle variations in the logits. This allows ALSA to provide robust and accurate performance estimates across a wide range of distribution shifts. Extensive experiments on vision, language, and graph benchmarks demonstrate ALSA's superiority over both softmax- and similarity-based baselines. Notably, ALSA's robustness under significant distribution shifts highlights its potential as a practical tool for reliable model evaluation.

cs.LG↗

arXiv2025

Few-Shot Radar Signal Recognition through Self-Supervised Learning and Radio Frequency Domain Adaptation

Radar signal recognition (RSR) plays a pivotal role in electronic warfare (EW), as accurately classifying radar signals is critical for informing decision-making. Recent advances in deep learning have shown significant potential in improving RSR in domains with ample annotated data. However, these methods fall short in EW scenarios where annotated radio frequency (RF) data are scarce or impractical to obtain. To address these challenges, we introduce a self-supervised learning (SSL) method which utilises masked signal modelling and RF domain adaption to perform few-shot RSR and enhance performance in environments with limited RF samples and annotations. We propose a two-step approach, first pre-training masked autoencoders (MAE) on baseband in-phase and quadrature (I/Q) signals from diverse RF domains, and then transferring the learned representations to the radar domain, where annotated data are scarce. Empirical results show that our lightweight self-supervised ResNet1D model with domain adaptation achieves up to a 17.5% improvement in 1-shot classification accuracy when pre-trained on in-domain signals (i.e., radar signals) and up to a 16.31% improvement when pre-trained on out-of-domain signals (i.e., comm signals), compared to its baseline without using SSL. We also present reference results for several MAE designs and pre-training strategies, establishing a new benchmark for few-shot radar signal classification.

cs.LG↗

arXiv2025

SVIP: Semantically Contextualized Visual Patches for Zero-Shot Learning

Zero-shot learning (ZSL) aims to recognize unseen classes without labeled training examples by leveraging class-level semantic descriptors such as attributes. A fundamental challenge in ZSL is semantic misalignment, where semantic-unrelated information involved in visual features introduce ambiguity to visual-semantic interaction. Unlike existing methods that suppress semantic-unrelated information post hoc either in the feature space or the model space, we propose addressing this issue at the input stage, preventing semantic-unrelated patches from propagating through the network. To this end, we introduce Semantically contextualized VIsual Patches (SVIP) for ZSL, a transformer-based framework designed to enhance visual-semantic alignment. Specifically, we propose a self-supervised patch selection mechanism that preemptively learns to identify semantic-unrelated patches in the input space. This is trained with the supervision from aggregated attention scores across all transformer layers, which estimate each patch's semantic score. As removing semantic-unrelated patches from the input sequence may disrupt object structure, we replace them with learnable patch embeddings. With initialization from word embeddings, we can ensure they remain semantically meaningful throughout feature extraction. Extensive experiments on ZSL benchmarks demonstrate that SVIP achieves state-of-the-art performance results while providing more interpretable and semantically rich feature representations. Code is available at https://github.com/uqzhichen/SVIP.

cs.CV↗

arXiv2025

Box-QAymo: Box-Referring VQA Dataset for Autonomous Driving

Interpretable communication is essential for safe and trustworthy autonomous driving, yet current vision-language models (VLMs) often operate under idealized assumptions and struggle to capture user intent in real-world scenarios. Existing driving-oriented VQA datasets are limited to full-scene descriptions or waypoint prediction, preventing the assessment of whether VLMs can respond to localized user-driven queries. We introduce Box-QAymo, a box-referring dataset and benchmark designed to both evaluate and finetune VLMs on spatial and temporal reasoning over user-specified objects. Users express intent by drawing bounding boxes, offering a fast and intuitive interface for focused queries in complex scenes. Specifically, we propose a hierarchical evaluation protocol that begins with binary sanity-check questions to assess basic model capacities, and progresses to (1) attribute prediction for box-referred objects, (2) motion understanding of target instances, and (3) spatiotemporal motion reasoning over inter-object dynamics across frames. To support this, we crowd-sourced fine-grained object classes and visual attributes that reflect the complexity drivers encounter, and extract object trajectories to construct temporally grounded QA pairs. Rigorous quality control through negative sampling, temporal consistency checks, and difficulty-aware balancing guarantee dataset robustness and diversity. Our comprehensive evaluation reveals significant limitations in current VLMs when queried about perception questions, highlighting the gap in achieving real-world performance. This work provides a foundation for developing more robust and interpretable autonomous driving systems that can communicate effectively with users under real-world conditions. Project page and dataset are available at https://djamahl99.github.io/qaymo-pages/.

cs.CV↗

arXiv2025

UQLegalAI@COLIEE2025: Advancing Legal Case Retrieval with Large Language Models and Graph Neural Networks

Legal case retrieval plays a pivotal role in the legal domain by facilitating the efficient identification of relevant cases, supporting legal professionals and researchers to propose legal arguments and make informed decision-making. To improve retrieval accuracy, the Competition on Legal Information Extraction and Entailment (COLIEE) is held annually, offering updated benchmark datasets for evaluation. This paper presents a detailed description of CaseLink, the method employed by UQLegalAI, the second highest team in Task 1 of COLIEE 2025. The CaseLink model utilises inductive graph learning and Global Case Graphs to capture the intrinsic case connectivity to improve the accuracy of legal case retrieval. Specifically, a large language model specialized in text embedding is employed to transform legal texts into embeddings, which serve as the feature representations of the nodes in the constructed case graph. A new contrastive objective, incorporating a regularization on the degree of case nodes, is proposed to leverage the information within the case reference relationship for model optimization. The main codebase used in our method is based on an open-sourced repo of CaseLink: https://github.com/yanran-tang/CaseLink.

cs.IR↗

arXiv2025

CodeMerge: Codebook-Guided Model Merging for Robust Test-Time Adaptation in Autonomous Driving

Maintaining robust 3D perception under dynamic and unpredictable test-time conditions remains a critical challenge for autonomous driving systems. Existing test-time adaptation (TTA) methods often fail in high-variance tasks like 3D object detection due to unstable optimization and sharp minima. While recent model merging strategies based on linear mode connectivity (LMC) offer improved stability by interpolating between fine-tuned checkpoints, they are computationally expensive, requiring repeated checkpoint access and multiple forward passes. In this paper, we introduce CodeMerge, a lightweight and scalable model merging framework that bypasses these limitations by operating in a compact latent space. Instead of loading full models, CodeMerge represents each checkpoint with a low-dimensional fingerprint derived from the source model's penultimate features and constructs a key-value codebook. We compute merging coefficients using ridge leverage scores on these fingerprints, enabling efficient model composition without compromising adaptation quality. Our method achieves strong performance across challenging benchmarks, improving end-to-end 3D detection 14.9% NDS on nuScenes-C and LiDAR-based detection by over 7.6% mAP on nuScenes-to-KITTI, while benefiting downstream tasks such as online mapping, motion prediction and planning even without training. Code and pretrained models are released in the supplementary material.

cs.CV↗

arXiv2025

GOLD: Graph Out-of-Distribution Detection via Implicit Adversarial Latent Generation

Despite graph neural networks' (GNNs) great success in modelling graph-structured data, out-of-distribution (OOD) test instances still pose a great challenge for current GNNs. One of the most effective techniques to detect OOD nodes is to expose the detector model with an additional OOD node-set, yet the extra OOD instances are often difficult to obtain in practice. Recent methods for image data address this problem using OOD data synthesis, typically relying on pre-trained generative models like Stable Diffusion. However, these approaches require vast amounts of additional data, as well as one-for-all pre-trained generative models, which are not available for graph data. Therefore, we propose the GOLD framework for graph OOD detection, an implicit adversarial learning pipeline with synthetic OOD exposure without pre-trained models. The implicit adversarial training process employs a novel alternating optimisation framework by training: (1) a latent generative model to regularly imitate the in-distribution (ID) embeddings from an evolving GNN, and (2) a GNN encoder and an OOD detector to accurately classify ID data while increasing the energy divergence between the ID embeddings and the generative model's synthetic embeddings. This novel approach implicitly transforms the synthetic embeddings into pseudo-OOD instances relative to the ID data, effectively simulating exposure to OOD scenarios without auxiliary data. Extensive OOD detection experiments are conducted on five benchmark graph datasets, verifying the superior performance of GOLD without using real OOD data compared with the state-of-the-art OOD exposure and non-exposure baselines.

cs.LG↗

arXiv2025

Continual Text-to-Video Retrieval with Frame Fusion and Task-Aware Routing

Text-to-Video Retrieval (TVR) aims to retrieve relevant videos based on textual queries. However, as video content evolves continuously, adapting TVR systems to new data remains a critical yet under-explored challenge. In this paper, we introduce the first benchmark for Continual Text-to-Video Retrieval (CTVR) to address the limitations of existing approaches. Current Pre-Trained Model (PTM)-based TVR methods struggle with maintaining model plasticity when adapting to new tasks, while existing Continual Learning (CL) methods suffer from catastrophic forgetting, leading to semantic misalignment between historical queries and stored video features. To address these two challenges, we propose FrameFusionMoE, a novel CTVR framework that comprises two key components: (1) the Frame Fusion Adapter (FFA), which captures temporal video dynamics while preserving model plasticity, and (2) the Task-Aware Mixture-of-Experts (TAME), which ensures consistent semantic alignment between queries across tasks and the stored video features. Thus, FrameFusionMoE enables effective adaptation to new video content while preserving historical text-video relevance to mitigate catastrophic forgetting. We comprehensively evaluate FrameFusionMoE on two benchmark datasets under various task settings. Results demonstrate that FrameFusionMoE outperforms existing CL and TVR methods, achieving superior retrieval performance with minimal degradation on earlier tasks when handling continuous video streams. Our code is available at: https://github.com/JasonCodeMaker/CTVR.

cs.CV↗

arXiv2025

Text Speaks Louder than Vision: ASCII Art Reveals Textual Biases in Vision-Language Models

Vision-language models (VLMs) have advanced rapidly in processing multimodal information, but their ability to reconcile conflicting signals across modalities remains underexplored. This work investigates how VLMs process ASCII art, a unique medium where textual elements collectively form visual patterns, potentially creating semantic-visual conflicts. We introduce a novel evaluation framework that systematically challenges five state-of-the-art models (including GPT-4o, Claude, and Gemini) using adversarial ASCII art, where character-level semantics deliberately contradict global visual patterns. Our experiments reveal a strong text-priority bias: VLMs consistently prioritize textual information over visual patterns, with visual recognition ability declining dramatically as semantic complexity increases. Various mitigation attempts through visual parameter tuning and prompt engineering yielded only modest improvements, suggesting that this limitation requires architectural-level solutions. These findings uncover fundamental flaws in how current VLMs integrate multimodal information, providing important guidance for future model development while highlighting significant implications for content moderation systems vulnerable to adversarial examples.

cs.CV↗

arXiv2025

MIRAGE: Multimodal Immersive Reasoning and Guided Exploration for Red-Team Jailbreak Attacks

While safety mechanisms have significantly progressed in filtering harmful text inputs, MLLMs remain vulnerable to multimodal jailbreaks that exploit their cross-modal reasoning capabilities. We present MIRAGE, a novel multimodal jailbreak framework that exploits narrative-driven context and role immersion to circumvent safety mechanisms in Multimodal Large Language Models (MLLMs). By systematically decomposing the toxic query into environment, role, and action triplets, MIRAGE constructs a multi-turn visual storytelling sequence of images and text using Stable Diffusion, guiding the target model through an engaging detective narrative. This process progressively lowers the model's defences and subtly guides its reasoning through structured contextual cues, ultimately eliciting harmful responses. In extensive experiments on the selected datasets with six mainstream MLLMs, MIRAGE achieves state-of-the-art performance, improving attack success rates by up to 17.5% over the best baselines. Moreover, we demonstrate that role immersion and structured semantic reconstruction can activate inherent model biases, facilitating the model's spontaneous violation of ethical safeguards. These results highlight critical weaknesses in current multimodal safety mechanisms and underscore the urgent need for more robust defences against cross-modal threats.

cs.CL↗

arXiv2025

SCORE: Soft Label Compression-Centric Dataset Condensation via Coding Rate Optimization

Dataset Condensation (DC) aims to obtain a condensed dataset that allows models trained on the condensed dataset to achieve performance comparable to those trained on the full dataset. Recent DC approaches increasingly focus on encoding knowledge into realistic images with soft labeling, for their scalability to ImageNet-scale datasets and strong capability of cross-domain generalization. However, this strong performance comes at a substantial storage cost which could significantly exceed the storage cost of the original dataset. We argue that the three key properties to alleviate this performance-storage dilemma are informativeness, discriminativeness, and compressibility of the condensed data. Towards this end, this paper proposes a \textbf{S}oft label compression-centric dataset condensation framework using \textbf{CO}ding \textbf{R}at\textbf{E} (SCORE). SCORE formulates dataset condensation as a min-max optimization problem, which aims to balance the three key properties from an information-theoretic perspective. In particular, we theoretically demonstrate that our coding rate-inspired objective function is submodular, and its optimization naturally enforces low-rank structure in the soft label set corresponding to each condensed data. Extensive experiments on large-scale datasets, including ImageNet-1K and Tiny-ImageNet, demonstrate that SCORE outperforms existing methods in most cases. Even with 30$\times$ compression of soft labels, performance decreases by only 5.5\% and 2.7\% for ImageNet-1K with IPC 10 and 50, respectively. Code will be released upon paper acceptance.

cs.CV↗

arXiv2025

Is Less More? Exploring Token Condensation as Training-free Test-time Adaptation

Contrastive Language-Image Pretraining (CLIP) excels at learning generalizable image representations but often falls short in zero-shot inference on certain downstream datasets. Test-time adaptation (TTA) mitigates this issue by adjusting components like normalization layers or context prompts, yet it typically requires large batch sizes and extensive augmentations, leading to high computational costs. This raises a key question: Can VLMs' performance drop in specific test cases be mitigated through efficient, training-free approaches? To explore the solution, we investigate token condensation (TC) techniques, originally designed to enhance vision transformer efficiency by refining token usage during inference. We observe that informative tokens improve visual-text alignment in VLMs like CLIP on unseen datasets. However, existing TC methods often fail to maintain in-distribution performance when reducing tokens, prompting us to ask: How can we transform TC into an effective ``free-lunch'' adaptation strategy for VLMs? To address this, we propose Token Condensation as Adaptation (TCA), a training-free adaptation method that takes a step beyond standard TC. Rather than passively discarding tokens, TCA condenses token representation by introducing reservoir-based domain anchor tokens for information-preserving token reduction and logits correction. TCA achieves up to a 21.4% performance improvement over the strongest baseline on cross-dataset benchmark and the CIFAR-100-Corrupted dataset while reducing GFLOPs by 12.2% to 48.9%, with minimal hyperparameter dependency on both CLIP and SigLIP series.

cs.CV↗

arXiv2025

Tit-for-Tat: Safeguarding Large Vision-Language Models Against Jailbreak Attacks via Adversarial Defense

Deploying large vision-language models (LVLMs) introduces a unique vulnerability: susceptibility to malicious attacks via visual inputs. However, existing defense methods suffer from two key limitations: (1) They solely focus on textual defenses, fail to directly address threats in the visual domain where attacks originate, and (2) the additional processing steps often incur significant computational overhead or compromise model performance on benign tasks. Building on these insights, we propose ESIII (Embedding Security Instructions Into Images), a novel methodology for transforming the visual space from a source of vulnerability into an active defense mechanism. Initially, we embed security instructions into defensive images through gradient-based optimization, obtaining security instructions in the visual dimension. Subsequently, we integrate security instructions from visual and textual dimensions with the input query. The collaboration between security instructions from different dimensions ensures comprehensive security protection. Extensive experiments demonstrate that our approach effectively fortifies the robustness of LVLMs against such attacks while preserving their performance on standard benign tasks and incurring an imperceptible increase in time costs.

cs.CR↗