arXiv subjects

Kaiyuan Zhang

Publications and source records attributed to Kaiyuan Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Security Considerations for Artificial Intelligence Agents

This article, a lightly adapted version of Perplexity's response to NIST/CAISI Request for Information 2025-0035, details our observations and recommendations concerning the security of frontier AI agents. These insights are informed by Perplexity's experience operating general-purpose agentic systems used by millions of users and thousands of enterprises in both controlled and open-world environments. Agent architectures change core assumptions around code-data separation, authority boundaries, and execution predictability, creating new confidentiality, integrity, and availability failure modes. We map principal attack surfaces across tools, connectors, hosting boundaries, and multi-agent coordination, with particular emphasis on indirect prompt injection, confused-deputy behavior, and cascading failures in long-running workflows. We then assess current defenses as a layered stack: input-level and model-level mitigations, sandboxed execution, and deterministic policy enforcement for high-consequence actions. Finally, we identify standards and research gaps, including adaptive security benchmarks, policy models for delegation and privilege control, and guidance for secure multi-agent system design aligned with NIST risk management principles.

cs.LG↗

arXiv2026

AutoVerifier: An Agentic Automated Verification Framework Using Large Language Models

Scientific and Technical Intelligence (S&TI) analysis requires verifying complex technical claims across rapidly growing literature, where existing approaches fail to bridge the verification gap between surface-level accuracy and deeper methodological validity. We present AutoVerifier, an LLM-based agentic framework that automates end-to-end verification of technical claims without requiring domain expertise. AutoVerifier decomposes every technical assertion into structured claim triples of the form (Subject, Predicate, Object), constructing knowledge graphs that enable structured reasoning across six progressively enriching layers: corpus construction and ingestion, entity and claim extraction, intra-document verification, cross-source verification, external signal corroboration, and final hypothesis matrix generation. We demonstrate AutoVerifier on a contested quantum computing claim, where the framework, operated by analysts with no quantum expertise, automatically identified overclaims and metric inconsistencies within the target paper, traced cross-source contradictions, uncovered undisclosed commercial conflicts of interest, and produced a final assessment. These results show that structured LLM verification can reliably evaluate the validity and maturity of emerging technologies, turning raw technical documents into traceable, evidence-backed intelligence assessments.

cs.AI↗

arXiv2026

One neutron triaxial halo candidates in aluminum isotopes from reaction observables

Microscopic description of one neutron ($1n$) halo candidates $^{40,42}$Al, with particular triaxial shape, is presented by combining the triaxial relativistic Hartree-Bogoliubov theory in continuum (TRHBc) with the Glauber reaction model for the first time. In this scheme, the reaction cross sections of aluminum isotopes on a carbon target at 240 and 900 MeV/A are calculated, which exhibit a pronounced increase for $^{40,42}$Al + $^{12}$C deviating from the systematic trend of their neighbours. Furthermore, the predicted longitudinal momentum distributions of the residues after $1n$ removal reactions for $^{40,42}$Al + $^{12}$C are narrower than those for $^{36,38}$Al + $^{12}$C, which suggest halo structure with spatially extended density distribution. Based on the large occupation probabilities of $p$-wave valence neutrons, we identify $^{40,42}$Al as the first triaxially deformed $1n$ $p$-wave halo candidates. This work cast a new light on the search for the heavier halo nuclei for future experiments in the mass region of $A\approx40$, through theoretical predictions from triaxial structure to reaction observables.

nucl-th↗

arXiv2026

Predicting reaction observables for the two-neutron halo candidates $^{31}$F and $^{39}$Na

Microscopic description of two-neutron ($2n$) halo candidates $^{31}$F and $^{39}$Na has been realized from nuclear structure to reaction observables for the first time. The reliability of the Glauber reaction model has been confirmed by exactly reproducing the momentum distributions of the benchmark $2n$ halo nucleus $^{11}$Li, with the identical structural inputs from the former work. Combined with the structure from the deformed relativistic Hartree-Bogoliubov theory in continuum (DRHBc), the Glauber model is applied to predict the reaction observables, including the reaction cross sections (RCSs) for the fluorine and sodium isotopes bombarding a carbon target at 240~MeV/A and the longitudinal momentum distributions of the fragments after $2n$ knockout reactions. It turns out that the calculated RCSs agree well with the available experimental data and a pronounced increase occurs to $^{29, 31}$F + $^{12}$C and $^{37, 39}$Na + $^{12}$C, which deviate from the original trend of their neighbours. Furthermore, the narrower longitudinal momentum distributions of the fragments after $2n$ knockout reactions demonstrate that $^{31}$F and $^{39}$Na have the dilute $2n$ halo structure. Such a new combination is promising to suggest new $2n$ halo candidates for future measurements.

nucl-th↗

arXiv2026

Window-based Membership Inference Attacks Against Fine-tuned Large Language Models

Most membership inference attacks (MIAs) against Large Language Models (LLMs) rely on global signals, like average loss, to identify training data. This approach, however, dilutes the subtle, localized signals of memorization, reducing attack effectiveness. We challenge this global-averaging paradigm, positing that membership signals are more pronounced within localized contexts. We introduce WBC (Window-Based Comparison), which exploits this insight through a sliding window approach with sign-based aggregation. Our method slides windows of varying sizes across text sequences, with each window casting a binary vote on membership based on loss comparisons between target and reference models. By ensembling votes across geometrically spaced window sizes, we capture memorization patterns from token-level artifacts to phrase-level structures. Extensive experiments across eleven datasets demonstrate that WBC substantially outperforms established baselines, achieving higher AUC scores and 2-3 times improvements in detection rates at low false positive thresholds. Our findings reveal that aggregating localized evidence is fundamentally more effective than global averaging, exposing critical privacy vulnerabilities in fine-tuned LLMs.

cs.CL↗

arXiv2026

Membership Inference Attacks Against Fine-tuned Diffusion Language Models

Diffusion Language Models (DLMs) represent a promising alternative to autoregressive language models, using bidirectional masked token prediction. Yet their susceptibility to privacy leakage via Membership Inference Attacks (MIA) remains critically underexplored. This paper presents the first systematic investigation of MIA vulnerabilities in DLMs. Unlike the autoregressive models' single fixed prediction pattern, DLMs' multiple maskable configurations exponentially increase attack opportunities. This ability to probe many independent masks dramatically improves detection chances. To exploit this, we introduce SAMA (Subset-Aggregated Membership Attack), which addresses the sparse signal challenge through robust aggregation. SAMA samples masked subsets across progressive densities and applies sign-based statistics that remain effective despite heavy-tailed noise. Through inverse-weighted aggregation prioritizing sparse masks' cleaner signals, SAMA transforms sparse memorization detection into a robust voting mechanism. Experiments on nine datasets show SAMA achieves 30% relative AUC improvement over the best baseline, with up to 8 times improvement at low false positive rates. These findings reveal significant, previously unknown vulnerabilities in DLMs, necessitating the development of tailored privacy defenses.

cs.LG↗

arXiv2026

STACodec: Semantic Token Assignment for Balancing Acoustic Fidelity and Semantic Information in Audio Codecs

Neural audio codecs are widely used for audio compression and can be integrated into token-based language models. Traditional codecs preserve acoustic details well but lack semantic information. Recent hybrid codecs attempt to incorporate semantic information through distillation, but this often degrades reconstruction performance, making it difficult to achieve both. To address this limitation, we introduce STACodec, a unified codec that integrates semantic information from self-supervised learning (SSL) models into the first layer of residual vector quantization (RVQ-1) via semantic token assignment (STA). To further eliminate reliance on SSL-based semantic tokenizers and improve efficiency during inference, we propose a semantic pre-distillation (SPD) module, which predicts semantic tokens directly for assignment to the first RVQ layer during inference. Experimental results show that STACodec outperforms existing hybrid codecs in both audio reconstruction and downstream semantic tasks, demonstrating a better balance between acoustic fidelity and semantic capability.

eess.AS↗

arXiv2026

Mind the Shift: Using Delta SSL Embeddings to Enhance Child ASR

Self-supervised learning (SSL) models have achieved impressive results across many speech tasks, yet child automatic speech recognition (ASR) remains challenging due to limited data and pretraining domain mismatch. Fine-tuning SSL models on child speech induces shifts in the representation space. We hypothesize that delta SSL embeddings, defined as the differences between embeddings from a finetuned model and those from its pretrained counterpart, encode task-specific information that complements finetuned features from another SSL model. We evaluate multiple fusion strategies on the MyST childrens corpus using different models. Results show that delta embedding fusion with WavLM yields up to a 10 percent relative WER reduction for HuBERT and a 4.4 percent reduction for W2V2, compared to finetuned embedding fusion. Notably, fusing WavLM with delta W2V2 embeddings achieves a WER of 9.64, setting a new state of the art among SSL models on the MyST corpus. These findings demonstrate the effectiveness of delta embeddings and highlight feature fusion as a promising direction for advancing child ASR.

cs.CL↗

arXiv2026

FutureX-Pro: Extending Future Prediction to High-Value Vertical Domains

Building upon FutureX, which established a live benchmark for general-purpose future prediction, this report introduces FutureX-Pro, including FutureX-Finance, FutureX-Retail, FutureX-PublicHealth, FutureX-NaturalDisaster, and FutureX-Search. These together form a specialized framework extending agentic future prediction to high-value vertical domains. While generalist agents demonstrate proficiency in open-domain search, their reliability in capital-intensive and safety-critical sectors remains under-explored. FutureX-Pro targets four economically and socially pivotal verticals: Finance, Retail, Public Health, and Natural Disaster. We benchmark agentic Large Language Models (LLMs) on entry-level yet foundational prediction tasks -- ranging from forecasting market indicators and supply chain demands to tracking epidemic trends and natural disasters. By adapting the contamination-free, live-evaluation pipeline of FutureX, we assess whether current State-of-the-Art (SOTA) agentic LLMs possess the domain grounding necessary for industrial deployment. Our findings reveal the performance gap between generalist reasoning and the precision required for high-value vertical applications.

cs.AI↗

arXiv2025

MME-CC: A Challenging Multi-Modal Evaluation Benchmark of Cognitive Capacity

As reasoning models scale rapidly, the essential role of multimodality in human cognition has come into sharp relief, driving a growing need to probe vision-centric cognitive behaviors. Yet, existing multimodal benchmarks either overemphasize textual reasoning or fall short of systematically capturing vision-centric cognitive behaviors, leaving the cognitive capacity of MLLMs insufficiently assessed. To address this limitation, we introduce MME-CC (Multi-Modal Evaluation benchmark of Cognitive Capacity), a vision-grounded benchmark that organizes 11 representative reasoning tasks into three fundamental categories of visual information: spatial, geometric, and knowledge-based reasoning, and provides fine-grained analyses of MLLMs' cognitive capacity across these dimensions. Based on MME-CC, we conduct extensive experiments over 16 representative MLLMs. Our study reveals that closed-source models currently lead overall (e.g., 42.66 for Gemini-2.5-Pro vs. 30.45 for GLM-4.5V), while spatial and geometric reasoning remain broadly weak (less than or equal to 30%). We further identify common error patterns, including orientation mistakes, fragile cross-view identity persistence, and poor adherence to counterfactual instructions, and observe that Chain-of-Thought typically follows a three-stage process (extract -> reason -> verify) with heavy reliance on visual extraction. We hope this work catalyzes a shift toward treating the cognitive capacity of MLLMs as central to both evaluation and model design.

cs.CL↗

arXiv2025

Comparing Unsupervised and Supervised Semantic Speech Tokens: A Case Study of Child ASR

Discrete speech tokens have gained attention for their storage efficiency and integration with Large Language Models (LLMs). They are commonly categorized into acoustic and semantic tokens, with the latter being more advantageous for Automatic Speech Recognition (ASR). Traditionally, unsupervised K-means clustering has been used to extract semantic speech tokens from Speech Foundation Models (SFMs). Recently, supervised methods, such as finite scalar quantization (FSQ) trained with ASR loss, have emerged for speech generation. Both approaches leverage pre-trained SFMs, benefiting low-resource tasks such as child ASR. This paper systematically compares supervised and unsupervised semantic speech tokens for child ASR. Results show that supervised methods not only outperform unsupervised ones but even unexpectedly surpass continuous representations, and they perform well even in ultra-low bitrate settings. These findings highlight the advantages of supervised semantic tokens and offer insights for improving discrete speech tokenization.

eess.AS↗

arXiv2025

Nanosecond-Scale Proton Emission from Triaxially Deformed Lu-148 Predicted with High Accuracy Qp Value via Novel Bayesian Evaluation

The half-life of the odd-odd deformed proton emitter $^{148}$Lu is predicted to be $196_{-129}^{+420}$ ns via the Wentzel-Kramers-Brillouin (WKB) approximation, in which the potential is extracted from the triaxial relativistic Hartree-Bogoliubov theory in continuum (TRHBc) and the proton decay energy $Q_{\rm p}$ is computed as 2.015(89) MeV by the Bayesian Neural Network - Beihang (BNN-BH) model for the first time. As a decisive factor, the uncertainty of $S_{\rm p}$ has been improved from 411 keV (Bayesian Machine Learning, BML) to 89 keV (BNN-BH) by taking the ensemble uncertainty into account and confining the error estimation to the neighboring nuclei. In consequence, the magnitude of the half-life's uncertainty can be reduced from 4 orders to 1 order, compared to that ($5.5_{-5.3}^{+636}$ ns) with $S_{\rm p}$ from the BML model. We also found that the range of half-life predicted by the TRHBc + WKB approach is consistent with those from the deformed relativistic Hartree-Bogoliubov theory in continuum (DRHBc) + WKB approach, and with those from an empirical formalism with the $S_{\rm p}$ obtained with the BNN-BH model. Furthermore, the means from the above 3 ways agree well with the experimental data for $^{149}$Lu, which gives us confidence to recommend a measurement of the half-life of proton emitter $^{148}$Lu.

nucl-th↗

arXiv2025

Random walks in Weyl chambers

We study a $d$-dimensional random walk with zero mean and finite variance in the Weyl chambers of type C and D. Under optimal moment assumptions we construct positive harmonic functions for random walks killed on exiting Weyl chambers. We also find the tail asymptotics for the exit time of the random walk from Weyl chambers.

math.PR↗

arXiv2025

TAI3: Testing Agent Integrity in Interpreting User Intent

LLM agents are increasingly deployed to automate real-world tasks by invoking APIs through natural language instructions. While powerful, they often suffer from misinterpretation of user intent, leading to the agent's actions that diverge from the user's intended goal, especially as external toolkits evolve. Traditional software testing assumes structured inputs and thus falls short in handling the ambiguity of natural language. We introduce TAI3, an API-centric stress testing framework that systematically uncovers intent integrity violations in LLM agents. Unlike prior work focused on fixed benchmarks or adversarial inputs, TAI3 generates realistic tasks based on toolkits' documentation and applies targeted mutations to expose subtle agent errors while preserving user intent. To guide testing, we propose semantic partitioning, which organizes natural language tasks into meaningful categories based on toolkit API parameters and their equivalence classes. Within each partition, seed tasks are mutated and ranked by a lightweight predictor that estimates the likelihood of triggering agent errors. To enhance efficiency, TAI3 maintains a datatype-aware strategy memory that retrieves and adapts effective mutation patterns from past cases. Experiments on 80 toolkit APIs demonstrate that TAI3 effectively uncovers intent integrity violations, significantly outperforming baselines in both error-exposing rate and query efficiency. Moreover, TAI3 generalizes well to stronger target models using smaller LLMs for test generation, and adapts to evolving APIs across domains.

cs.SE↗

arXiv2025

Equipping Retrieval-Augmented Large Language Models with Document Structure Awareness

While large language models (LLMs) demonstrate impressive capabilities, their reliance on parametric knowledge often leads to factual inaccuracies. Retrieval-Augmented Generation (RAG) mitigates this by leveraging external documents, yet existing approaches treat retrieved passages as isolated chunks, ignoring valuable structure that is crucial for document organization. Motivated by this gap, we propose Retrieve-DocumentRoute-Read (RDR2), a novel framework that explicitly incorporates structural information throughout the RAG process. RDR2 employs an LLM-based router to dynamically navigate document structure trees, jointly evaluating content relevance and hierarchical relationships to assemble optimal evidence. Our key innovation lies in formulating document routing as a trainable task, with automatic action curation and structure-aware passage selection inspired by human reading strategies. Through comprehensive evaluation on five challenging datasets, RDR2 achieves state-of-the-art performance, demonstrating that explicit structural awareness significantly enhances RAG systems' ability to acquire and utilize knowledge, particularly in complex scenarios requiring multi-document synthesis.

cs.CL↗

arXiv2025

Triaxial relativistic Hartree-Bogoliubov theory in continuum for exotic nuclei

A triaxial relativistic Hartree-Bogoliubov theory in continuum (TRHBc) has been developed to incorporate triaxial deformation, pairing correlations, and continuum effects in a fully microscopic and self-consistent way, aiming for a reliable description of triaxial exotic nuclei with extreme neutron-to-proton ratios. The TRHBc formalism is presented in detail, and its numerical implementation is benchmarked against the results from the axially deformed relativistic Hartree-Bogoliubov theory in continuum and the TRHB theory in harmonic oscillator expansion. The TRHBc theory is applied to investigate the aluminum isotopes systematically, and the available data are well reproduced for the binding energies, one- and two-neutron separation energies, and charge radii. The nuclei near the one-neutron drip line, $^{40}$Al and $^{42}$Al, are found to be triaxially deformed with one-neutron separation energies below 1 MeV. Possible neutron halos in the triaxial nuclei $^{40}$Al and $^{42}$Al are explored by examining the single-particle levels around the Fermi surface, including their composition and contribution to the total neutron density. The existence of neutron halos in $^{40}$Al and $^{42}$Al is also supported by the halo scale, which is comparable to other halo nuclei well-established previously. More importantly, the mechanism for the halo formation in $^{40}$Al is revealed to be the triaxial deformation, which results in the decoupling of the halo orbitals from those of the core.

nucl-th↗

arXiv2025

FinSearchComp: Towards a Realistic, Expert-Level Evaluation of Financial Search and Reasoning

Search has emerged as core infrastructure for LLM-based agents and is widely viewed as critical on the path toward more general intelligence. Finance is a particularly demanding proving ground: analysts routinely conduct complex, multi-step searches over time-sensitive, domain-specific data, making it ideal for assessing both search proficiency and knowledge-grounded reasoning. Yet no existing open financial datasets evaluate data searching capability of end-to-end agents, largely because constructing realistic, complicated tasks requires deep financial expertise and time-sensitive data is hard to evaluate. We present FinSearchComp, the first fully open-source agent benchmark for realistic, open-domain financial search and reasoning. FinSearchComp comprises three tasks -- Time-Sensitive Data Fetching, Simple Historical Lookup, and Complex Historical Investigation -- closely reproduce real-world financial analyst workflows. To ensure difficulty and reliability, we engage 70 professional financial experts for annotation and implement a rigorous multi-stage quality-assurance pipeline. The benchmark includes 635 questions spanning global and Greater China markets, and we evaluate 21 models (products) on it. Grok 4 (web) tops the global subset, approaching expert-level accuracy. DouBao (web) leads on the Greater China subset. Experimental analyses show that equipping agents with web search and financial plugins substantially improves results on FinSearchComp, and the country origin of models and tools impact performance significantly.By aligning with realistic analyst tasks and providing end-to-end evaluation, FinSearchComp offers a professional, high-difficulty testbed for complex financial search and reasoning.

cs.LG↗

arXiv2025

MARS-Bench: A Multi-turn Athletic Real-world Scenario Benchmark for Dialogue Evaluation

Large Language Models (\textbf{LLMs}), e.g. ChatGPT, have been widely adopted in real-world dialogue applications. However, LLMs' robustness, especially in handling long complex dialogue sessions, including frequent motivation transfer, sophisticated cross-turn dependency, is criticized all along. Nevertheless, no existing benchmarks can fully reflect these weaknesses. We present \textbf{MARS-Bench}, a \textbf{M}ulti-turn \textbf{A}thletic \textbf{R}eal-world \textbf{S}cenario Dialogue \textbf{Bench}mark, designed to remedy the gap. MARS-Bench is constructed from play-by-play text commentary so to feature realistic dialogues specifically designed to evaluate three critical aspects of multi-turn conversations: Ultra Multi-turn, Interactive Multi-turn, and Cross-turn Tasks. Extensive experiments on MARS-Bench also reveal that closed-source LLMs significantly outperform open-source alternatives, explicit reasoning significantly boosts LLMs' robustness on handling long complex dialogue sessions, and LLMs indeed face significant challenges when handling motivation transfer and sophisticated cross-turn dependency. Moreover, we provide mechanistic interpretability on how attention sinks due to special tokens lead to LLMs' performance degradation when handling long complex dialogue sessions based on attention visualization experiment in Qwen2.5-7B-Instruction.

cs.CL↗