arXiv subjects

Zhiyuan Jiang

Publications and source records attributed to Zhiyuan Jiang.

At least 19 recordsLinked to original sources

arXiv2026

MOMAT: Mixture of Multiple Atlases for Low-Power Jailbreak Defense of Quantized LLMs

Quantized large language models are increasingly deployed on edge devices for their low latency and energy efficiency. However, model quantization weakens alignment safeguards, leaving qLLMs (quantized large language models) highly vulnerable to jailbreak attacks. To address this challenge, we present MOMAT (Mixture of Multiple Atlases), a hardware-enhanced safety framework that combines structured knowledge retrieval with low-power defense acceleration. Each atlas represents a semantic cluster of harmful or benign sample sets and policy templates, enabling domain-localized Retrieval-Augmented Generation guarding that mitigates the curse of dimensionality and the resulting semantic sparsity problem in large, heterogeneous safety databases. MOMAT retrieves top-$k$ similarity features from all atlases for each prompt and evaluates them using a lightweight MoE (Mixture of Experts) detector, while a CiM (Compute-in-Memory)-accelerated similarity engine performs fast, low-power atlas-local retrieval. MOMAT's CiM-based retrieval accelerates a 100-query batch from 15,052.44 ms to 3,207.21 ns (a $4.69 \times 10^6\times$ speedup) and reduces energy from $8.1 \times 10^7$ $μ$J to 3.32 $μ$J, yielding an approximately $2.5 \times 10^5\times$ energy reduction over DRAM-based (Raspberry Pi) baselines. Red-team evaluations across standard benchmarks show that MOMAT matches the defense performance of state-of-the-art methods while avoiding benign overkill and providing substantial efficiency gains, demonstrating that CiM-based modular defenses can make edge-deployed qLLMs both safer and more energy-efficient. We will release the full 223.2k-sample dataset to foster future research.

cs.CR↗

arXiv2026

HyGRAIL: Cost-Aware and Evidence-Grounded Scientific Hypothesis Discovery over Knowledge Graphs

Scientific knowledge graphs organize entities and relations extracted from scientific literature, but they remain inherently incomplete. Missing typed links in such graphs can therefore represent plausible scientific hypotheses, such as unexplored associations between materials and applications. However, scientific hypothesis discovery is challenging because true discoveries are extremely sparse among typed candidate pairs: graph neural networks (GNNs) are efficient but unreliable for ambiguous cases, while large language models (LLMs) are knowledgeable but too costly to apply exhaustively and are not naturally grounded in graph structures. We propose HyGRAIL, a cost-aware and evidence-grounded framework that combines heterogeneous GNN triage with LLM-based hypothesis review. HyGRAIL first uses a GNN to score candidate hypotheses and identify a validation-calibrated ambiguous region, routing only graph-uncertain cases to LLM review. For each routed hypothesis, HyGRAIL retrieves node-level associations and multi-hop relational paths from the knowledge graph (KG), then converts this structured evidence into natural language through template-based or LLM-based naturalization. An LLM review agent finally judges each hard hypothesis using the naturalized evidence and validation-selected decision criteria. On MatKG, HyGRAIL achieves the best F1 score of 0.429, improving over the strongest prior baseline by 0.242 F1 points and over the GNN-only baseline by 0.322. Meanwhile, GNN triage reduces the LLM call rate by 54.36% on average. Ablation studies further show that retrieved graph evidence is crucial for reliable hypothesis verification and that compact, two-sided evidence is more effective than simply increasing retrieval quantity.

cs.CL↗

arXiv2026

Demystifying Agent Skills: Why They Work-Until They Don't

Skills have emerged as a practical and effective approach for enhancing LLM agents at inference time through structured packages of knowledge. However, existing evaluations largely measure whether skills improve aggregated task success, leaving a more fundamental question underexplored: \emph{\textbf{When do skills help, why do they work, and where do they fail?}} Through controlled experiments across various benchmarks, agent harnesses and LLMs, we isolate the effects of representation, outcome annotation, retrieval difficulty, and cross-framework robustness of skills. To further answer this question, we design a contrastive study that combines controlled quantitative experiments with paired trajectory analysis. We normalize 8,135 trial records from controlled experiments and retain 238 valid unique labels from 240 open-coded records. We consolidate these observations into a taxonomy of three high-level categories and twelve skill-use modes: skills work when noisy trajectories become procedural anchors that stabilize execution. Skills improve over Workflow Memory by 6.06 points in matched comparisons. Procedural anchoring accounts for 65.7\% of skill cases, versus 4.5\% for explicit knowledge injection, showing that skills stabilize action rather than inject missing facts. Retrieval is a separate bottleneck: as pools grow from 5 to 100, actual-use precision falls from 29.6\% to 3.3\%. Confusable distractors impair offline identification, yet downstream success remains stable; exact ground-truth invocation is neither sufficient nor necessary. Skills fail under brittle assumptions, incompatible contexts, or insufficient adaptation. These findings move evaluation beyond aggregate success rates and guide reliable self-evolving agents.

cs.AI↗

arXiv2026

Cohomology of complex supertori

We consider supertori which are quotients of affine superspace by translations by algebraically independent odd parameters. Specifically, we describe the ring structure of its space of global sections by generators and relations and completely determine the coherent cohomology groups of its structure sheaf by reducing it to a problem of Lie algebra cohomology. We also show that Poincaré duality on sheaf cohomology is compatible with that of the group cohomology of the translation group and give explicit tables of examples. Finally, we compute the Picard groups of these supertori.

math.AG↗

arXiv2026

TGPP: Trajectory-Guided Plug-and-Play Priors for Sparse Radio Map Reconstruction

Radio map (RM) reconstruction is essential for environment-aware wireless networks, but practical measurements are often collected along mobility trajectories rather than randomly scattered over the target region. Such trajectory-sampled observations induce spatially heterogeneous uncertainty: near-trajectory regions are directly constrained, whereas distant or occluded regions remain weakly observed, leading to degraded reconstruction accuracy in under-constrained areas. To address this problem, we propose Trajectory-Guided Plug-and-Play Priors (TGPP), a general guidance module for sparse RM reconstruction. TGPP learns an explicit guidance map as an interpretable input-space risk prior, and an implicit guide feature that is projected and fused with backbone hidden representations. TGPP can be attached to different reconstruction backbones without changing their original task formulation. We further introduce RadioFlow-LDM, a latent flow-based generative backbone, and apply TGPP to deterministic, adversarial, graph-based, and latent generative reconstruction models. Experiments on RadioMapSeer with five trajectory sampling rates show that trajectory-sampled reconstruction differs substantially from random sparse interpolation. TGPP improves most reconstruction metrics across backbones, achieving up to 43.1% NMSE reduction relative to the corresponding base backbone without trajectory-guided priors.

eess.SP↗

arXiv2026

LLM-as-Judge Framework for Evaluating Tone-Induced Hallucination in Vision-Language Models

Vision-Language Models (VLMs) are increasingly deployed in settings where reliable visual grounding carries operational consequences, yet their behavior under progressively coercive prompt phrasing remains undercharacterized. Existing hallucination benchmarks predominantly rely on neutral prompts and binary detection, leaving open how both the incidence and the intensity of fabrication respond to graded linguistic pressure across structurally distinct task types. We present Ghost-100, a procedurally constructed benchmark of 800 synthetically generated images spanning eight categories across three task families: text-illegibility, time-reading, and object-absence, each designed under a negative-ground-truth principle that guarantees the queried target is absent, illegible, or indeterminate by construction. Every image is paired with five prompts drawn from a structured 5-Level Prompt Intensity Framework, holding the image and task identity fixed while varying only directive force, so that tone is isolated as the sole independent variable. We adopt a dual-track evaluation protocol: a rule-based H-Rate measuring the proportion of responses in which a model crosses from grounded refusal into unsupported positive commitment, and a GPT-4o-mini-judged H-Score on a 1-5 scale characterizing the confidence and specificity of fabrication once it occurs. We additionally release a three-stage automated validation workflow, which retrospectively confirms 717 of 800 images as strictly compliant. Evaluating nine open-weight VLMs, we find that H-Rate and H-Score dissociate substantially across model families, reading-style and presence-detection subsets respond to prompt pressure in qualitatively different ways, and several models exhibit non-monotonic sensitivity peaking at intermediate tone levels: patterns that aggregate metrics obscure.

cs.CV↗

arXiv2026

An approach to the abundance conjecture for Kähler varieties via algebraic reduction

In this article, we establish a strategy to the abundance conjecture for Kähler varieties via induction on algebraic dimension. Our strategy is to reduce the abundance conjecture for Kähler varieties to the abundance conjecture for projective varieties using the algebraic reduction fibration. In dimension 4, we apply our inductive strategy to obtain some cases of the abundance conjecture for Kähler fourfolds that are not algebraic or have trivial $K_X$.

math.AG↗

arXiv2026

FuzzySQL: Uncovering Hidden Vulnerabilities in DBMS Special Features with LLM-Driven Fuzzing

Traditional database fuzzing techniques primarily focus on syntactic correctness and general SQL structures, leaving critical yet obscure DBMS features, such as system-level modes (e.g., GTID), programmatic constructs (e.g., PROCEDURE), advanced process commands (e.g., KILL), largely underexplored. Although rarely triggered by typical inputs, these features can lead to severe crashes or security issues when executed under edge-case conditions. In this paper, we present FuzzySQL, a novel LLM-powered adaptive fuzzing framework designed to uncover subtle vulnerabilities in DBMS special features. FuzzySQL combines grammar-guided SQL generation with logic-shifting progressive mutation, a novel technique that explores alternative control paths by negating conditions and restructuring execution logic, synthesizing structurally and semantically diverse test cases. To further ensure deeper execution coverage of the back end, FuzzySQL employs a hybrid error repair pipeline that unifies rule-based patching with LLM-driven semantic repair, enabling automatic correction of syntactic and context-sensitive failures. We evaluate FuzzySQL across multiple DBMSs, including MySQL, MariaDB, SQLite, PostgreSQL and Clickhouse, uncovering 64 vulnerabilities, 27 of which are tied to under-tested DBMS special features. As of this writing, 60 cases have been confirmed with 9 assigned CVE identifiers, 31 already fixed by vendors, and additional vulnerabilities scheduled to be patched in upcoming releases. Our results highlight the limitations of conventional fuzzers in semantic feature coverage and demonstrate the potential of LLM-based fuzzing to discover deeply hidden bugs in complex database systems.

cs.DB↗

arXiv2026

Tone Matters: The Impact of Linguistic Tone on Hallucination in VLMs

Vision-Language Models (VLMs) are increasingly used in safety-critical applications that require reliable visual grounding. However, these models often hallucinate details that are not present in the image to satisfy user prompts. While recent datasets and benchmarks have been introduced to evaluate systematic hallucinations in VLMs, many hallucination behaviors remain insufficiently characterized. In particular, prior work primarily focuses on object presence or absence, leaving it unclear how prompt phrasing and structural constraints can systematically induce hallucinations. In this paper, we investigate how different forms of prompt pressure influence hallucination behavior. We introduce Ghost-100, a procedurally generated dataset of synthetic scenes in which key visual details are deliberately removed, enabling controlled analysis of absence-based hallucinations. Using a structured 5-Level Prompt Intensity Framework, we vary prompts from neutral queries to toxic demands and rigid formatting constraints. We evaluate three representative open-weight VLMs: MiniCPM-V 2.6-8B, Qwen2-VL-7B, and Qwen3-VL-8B. Across all three models, hallucination rates do not increase monotonically with prompt intensity. All models exhibit reductions at higher intensity levels at different thresholds, though not all show sustained reduction under maximum coercion. These results suggest that current safety alignment is more effective at detecting semantic hostility than structural coercion, revealing model-specific limitations in handling compliance pressure. Our dataset is available at: https://github.com/bli1/tone-matters

cs.CV↗

arXiv2026

Wow, wo, val! A Comprehensive Embodied World Model Evaluation Turing Test

As world models gain momentum in Embodied AI, an increasing number of works explore using video foundation models as predictive world models for downstream embodied tasks like 3D prediction or interactive generation. However, before exploring these downstream tasks, video foundation models still have two critical questions unanswered: (1) whether their generative generalization is sufficient to maintain perceptual fidelity in the eyes of human observers, and (2) whether they are robust enough to serve as a universal prior for real-world embodied agents. To provide a standardized framework for answering these questions, we introduce the Embodied Turing Test benchmark: WoW-World-Eval (Wow,wo,val). Building upon 609 robot manipulation data, Wow-wo-val examines five core abilities, including perception, planning, prediction, generalization, and execution. We propose a comprehensive evaluation protocol with 22 metrics to assess the models' generation ability, which achieves a high Pearson Correlation between the overall score and human preference (>0.93) and establishes a reliable foundation for the Human Turing Test. On Wow-wo-val, models achieve only 17.27 on long-horizon planning and at best 68.02 on physical consistency, indicating limited spatiotemporal consistency and physical reasoning. For the Inverse Dynamic Model Turing Test, we first use an IDM to evaluate the video foundation models' execution accuracy in the real world. However, most models collapse to $\approx$ 0% success, while WoW maintains a 40.74% success rate. These findings point to a noticeable gap between the generated videos and the real world, highlighting the urgency and necessity of benchmarking World Model in Embodied AI.

cs.RO↗

arXiv2025

Zoom in, Click out: Unlocking and Evaluating the Potential of Zooming for GUI Grounding

Grounding is a fundamental capability for building graphical user interface (GUI) agents. Although existing approaches rely on large-scale bounding box supervision, they still face various challenges, such as cross-platform generalization, complex layout analysis, and fine-grained element localization. In this paper, we investigate zoom as a strong yet underexplored prior for GUI grounding, and propose a training-free method, ZoomClick. By characterizing four key properties of zoom (i.e., pre-zoom, depth, shrink size, minimal crop size), we unlock its full capabilities for dynamic spatial focusing and adaptive context switching. Experiments demonstrate that our method significantly boosts the performance of both general vision-language and specialized GUI grounding models, achieving state-of-the-art results on several mainstream benchmarks; for example, UI-Venus-72B attains a 73.1% success rate on ScreenSpot-Pro. Furthermore, we present GUIZoom-Bench, a benchmark for evaluating model adaptability to zoom, aiming to inspire future research on improving zoom for further training and test-time scaling in GUI grounding tasks.

cs.CV↗

arXiv2025

WoW: Towards a World omniscient World model Through Embodied Interaction

Humans develop an understanding of intuitive physics through active interaction with the world. This approach is in stark contrast to current video models, such as Sora, which rely on passive observation and therefore struggle with grasping physical causality. This observation leads to our central hypothesis: authentic physical intuition of the world model must be grounded in extensive, causally rich interactions with the real world. To test this hypothesis, we present WoW, a 14-billion-parameter generative world model trained on 2 million robot interaction trajectories. Our findings reveal that the model's understanding of physics is a probabilistic distribution of plausible outcomes, leading to stochastic instabilities and physical hallucinations. Furthermore, we demonstrate that this emergent capability can be actively constrained toward physical realism by SOPHIA, where vision-language model agents evaluate the DiT-generated output and guide its refinement by iteratively evolving the language instructions. In addition, a co-trained Inverse Dynamics Model translates these refined plans into executable robotic actions, thus closing the imagination-to-action loop. We establish WoWBench, a new benchmark focused on physical consistency and causal reasoning in video, where WoW achieves state-of-the-art performance in both human and autonomous evaluation, demonstrating strong ability in physical causality, collision dynamics, and object permanence. Our work provides systematic evidence that large-scale, real-world interaction is a cornerstone for developing physical intuition in AI. Models, data, and benchmarks will be open-sourced.

cs.RO↗

arXiv2025

Unlimited Vector Processing for Wireless Baseband Based on RISC-V Extension

Wireless baseband processing (WBP) serves as an ideal scenario for utilizing vector processing, which excels in managing data-parallel operations due to its parallel structure. However, conventional vector architectures face certain constraints such as limited vector register sizes, reliance on power-of-two vector length multipliers, and vector permutation capabilities tied to specific architectures. To address these challenges, we have introduced an instruction set extension (ISE) based on RISC-V known as unlimited vector processing (UVP). This extension enhances both the flexibility and efficiency of vector computations. UVP employs a novel programming model that supports non-power-of-two register groupings and hardware strip-mining, thus enabling smooth handling of vectors of varying lengths while reducing the software strip-mining burden. Vector instructions are categorized into symmetric and asymmetric classes, complemented by specialized load/store strategies to optimize execution. Moreover, we present a hardware implementation of UVP featuring sophisticated hazard detection mechanisms, optimized pipelines for symmetric tasks such as fixed-point multiplication and division, and a robust permutation engine for effective asymmetric operations. Comprehensive evaluations demonstrate that UVP significantly enhances performance, achieving up to 3.0$\times$ and 2.1$\times$ speedups in matrix multiplication and fast Fourier transform (FFT) tasks, respectively, when measured against lane-based vector architectures. Our synthesized RTL for a 16-lane configuration using SMIC 40nm technology spans 0.94 mm$^2$ and achieves an area efficiency of 21.2 GOPS/mm$^2$.

cs.AR↗

arXiv2025

Zoozve: A Strip-Mining-Free RISC-V Vector Extension with Arbitrary Register Grouping Compilation Support (WIP)

Vector processing is crucial for boosting processor performance and efficiency, particularly with data-parallel tasks. The RISC-V "V" Vector Extension (RVV) enhances algorithm efficiency by supporting vector registers of dynamic sizes and their grouping. Nevertheless, for very long vectors, the static number of RVV vector registers and its power-of-two grouping can lead to performance restrictions. To counteract this limitation, this work introduces Zoozve, a RISC-V vector instruction extension that eliminates the need for strip-mining. Zoozve allows for flexible vector register length and count configurations to boost data computation parallelism. With a data-adaptive register allocation approach, Zoozve permits any register groupings and accurately aligns vector lengths, cutting down register overhead and alleviating performance declines from strip-mining. Additionally, the paper details Zoozve's compiler and hardware implementations using LLVM and SystemVerilog. Initial results indicate Zoozve yields a minimum 10.10$\times$ reduction in dynamic instruction count for fast Fourier transform (FFT), with a mere 5.2\% increase in overall silicon area.

cs.PL↗

arXiv2025

A Hierarchical Dataflow-Driven Heterogeneous Architecture for Wireless Baseband Processing

Wireless baseband processing (WBP) is a key element of wireless communications, with a series of signal processing modules to improve data throughput and counter channel fading. Conventional hardware solutions, such as digital signal processors (DSPs) and more recently, graphic processing units (GPUs), provide various degrees of parallelism, yet they both fail to take into account the cyclical and consecutive character of WBP. Furthermore, the large amount of data in WBPs cannot be processed quickly in symmetric multiprocessors (SMPs) due to the unpredictability of memory latency. To address this issue, we propose a hierarchical dataflow-driven architecture to accelerate WBP. A pack-and-ship approach is presented under a non-uniform memory access (NUMA) architecture to allow the subordinate tiles to operate in a bundled access and execute manner. We also propose a multi-level dataflow model and the related scheduling scheme to manage and allocate the heterogeneous hardware resources. Experiment results demonstrate that our prototype achieves $2\times$ and $2.3\times$ speedup in terms of normalized throughput and single-tile clock cycles compared with GPU and DSP counterparts in several critical WBP benchmarks. Additionally, a link-level throughput of $288$ Mbps can be achieved with a $45$-core configuration.

cs.AR↗

arXiv2022

Prediction-based Hybrid Slicing Framework for Service Level Agreement Guarantee in Mobility Scenarios: A Deep Learning Approach

Network slicing is a critical driver for guaranteeing the diverse service level agreements (SLA) in 5G and future networks. Inter-slice radio resource allocation (IS-RRA) in the radio access network (RAN) is very important. However, user mobility brings new challenges for optimal IS-RRA. This paper first proposes a soft and hard hybrid slicing framework where a common slice is introduced to realize a trade-off between isolation and spectrum efficiency (SE). To address the challenges posed by user mobility, we propose a two-step deep learning-based algorithm: joint long short-term memory (LSTM)-based network state prediction and deep Q network (DQN)-based slicing strategy. In the proposal, LSTM networks are employed to predict traffic demand and the location of each user in a slicing window level. Moreover, channel gain is mapped by location and a radio map. Then, the predicted channel gain and traffic demand are input to the DQN to output the precise slicing adjustment. Finally, experiment results confirm the effectiveness of our proposed slicing framework: the slices' SLA can be guaranteed well, and the proposed algorithm can achieve near-optimal performance in terms of the SLA satisfaction ratio, isolation degree and SE.

eess.SY↗

arXiv2022

A Hard and Soft Hybrid Slicing Framework for Service Level Agreement Guarantee via Deep Reinforcement Learning

Network slicing is a critical driver for guaranteeing the diverse service level agreements (SLA) in 5G and future networks. Recently, deep reinforcement learning (DRL) has been widely utilized for resource allocation in network slicing. However, existing related works do not consider the performance loss associated with the initial exploration phase of DRL. This paper proposes a new performance-guaranteed slicing strategy with a soft and hard hybrid slicing setting. Mainly, a common slice setting is applied to guarantee slices' SLA when training the neural network. Moreover, the resource of the common slice tends to precisely redistribute to slices with the training of DRL until it converges. Furthermore, experiment results confirm the effectiveness of our proposed slicing framework: the slices' SLA of the training phase can be guaranteed, and the proposed algorithm can achieve the near-optimal performance in terms of the SLA satisfaction ratio, isolation degree and spectrum maximization after convergence.

cs.NI↗

arXiv2021

Fairness for Freshness: Optimal Age of Information Based OFDMA Scheduling with Minimal Knowledge

It is becoming increasingly clear that an important task for wireless networks is to minimize the age of information (AoI), i.e., the timeliness of information delivery. While mainstream approaches generally rely on the real-time observation of user AoI and channel state, there has been little attention to solve the problem in a complete (or partial) absence of such knowledge. In this article, we present a novel study to address the optimal blind radio resource scheduling problem in orthogonal frequency division multiplexing access (OFDMA) systems towards minimizing long-term average AoI, which is proven to be the composition of time-domain-fair clustered round-robin and frequency-domain-fair intra-cluster sub-carrier assignment. Heuristic solutions that are near-optimal as shown by simulation results are also proposed to effectively improve the performance upon presence of various degrees of extra knowledge, e.g., channel state and AoI.

cs.IT↗