arXiv subjects

Dan Zhao

Publications and source records attributed to Dan Zhao.

At least 37 records · Page 2Linked to original sources

arXiv2025

A Safety and Security Framework for Real-World Agentic Systems

This paper introduces a dynamic and actionable framework for securing agentic AI systems in enterprise deployment. We contend that safety and security are not merely fixed attributes of individual models but also emergent properties arising from the dynamic interactions among models, orchestrators, tools, and data within their operating environments. We propose a new way of identification of novel agentic risks through the lens of user safety. Although, for traditional LLMs and agentic models in isolation, safety and security has a clear separation, through the lens of safety in agentic systems, they appear to be connected. Building on this foundation, we define an operational agentic risk taxonomy that unifies traditional safety and security concerns with novel, uniquely agentic risks, including tool misuse, cascading action chains, and unintended control amplification among others. At the core of our approach is a dynamic agentic safety and security framework that operationalizes contextual agentic risk management by using auxiliary AI models and agents, with human oversight, to assist in contextual risk discovery, evaluation, and mitigation. We further address one of the most challenging aspects of safety and security of agentic systems: risk discovery through sandboxed, AI-driven red teaming. We demonstrate the framework effectiveness through a detailed case study of NVIDIA flagship agentic research assistant, AI-Q Research Assistant, showcasing practical, end-to-end safety and security evaluations in complex, enterprise-grade agentic workflows. This risk discovery phase finds novel agentic risks that are then contextually mitigated. We also release the dataset from our case study, containing traces of over 10,000 realistic attack and defense executions of the agentic workflow to help advance research in agentic safety.

cs.LG↗

arXiv2025

Lorentz Skew Scattering Nonreciprocal Magneto-Transport

In materials with broken inversion symmetry, nonreciprocal magneto-transport (NRMT) manifests as a bilinear dependence of charge conductivity on applied electric (E) and magnetic (B) fields. This phenomenon is deeply rooted in symmetry and electronic quantum geometry, holding promise for novel rectification and detector technologies. Existing experimental studies generally attribute NRMT to Zeeman-driven mechanisms and exhibit quadratic scaling with conductivity. Here, we report a previously unknown NRMT microscopic mechanism - Lorentz skew scattering (LSK) - revealed through the discovery of an unprecedented quartic scaling law of NRMT as well as quantitative agreement between theory and experiment in BiTeBr. LSK emerges from the interplay of Lorentz force and skew scattering, bridging classical field effect to quantum scattering effect on the Fermi surface. We demonstrate that the LSK dominates NRMT in BiTeBr, and elucidate that this dominance over other possible contributions stems from high mobility and strong Rashba splitting. The finding of LSK mechanism is of unique importance because it unveils the leading NRMT effect in high-mobility systems and suggests a universal principle towards strong NRMT by enhancing electronic relaxation time in topological materials, rendering a new designing idea for low-dissipation rectifiers and high-performance quantum electronics.

cond-mat.mes-hall↗

arXiv2025

Synchronization transitions and spike dynamics in a higher-order Kuramoto model with Lévy noise

Synchronization in various complex networks is significantly influenced by higher-order interactions combined with non-Gaussian stochastic perturbations, yet their mechanisms remain mainly unclear. In this paper, we systematically investigate the synchronization and spike dynamics in a higher-order Kuramoto model subjected to non-Gaussian Lévy noise. Using the mean order parameter, mean first-passage time, and basin stability, we identify clear boundaries distingusihing synchronization and incoherent states under different stability indexes and scale parameters of Lévy noise. For fixed coupling parameters, synchronization weakens as the stability index decreases, and even completely disappears when the scale parameter exceeds a critical threshold. By varying coupling parameters, we find significant dynamical phenomena including bifurcations and hysteresis. Lévy noise smooths the synchronization transitions and requires stronger coupling compared to Gaussian white noise. Furthermore, we define spikes and systematically investigate their statistical and spectral characteristics. The maximum number of spikes is observed at small scale parameter. A generalized spectral analysis further reveals burst-like structure via an edit distance algorithm. Moreover, a power-law distribution is observed in the large inter-window intervals of spikes, showing great memory effects. These findings deepen the understanding of synchronization and extreme events in complex networks driven by non-Gaussian Lévy noise.

cond-mat.stat-mech↗

arXiv2025

$\texttt{BluePrint}$: A Social Media User Dataset for LLM Persona Evaluation and Training

Large language models (LLMs) offer promising capabilities for simulating social media dynamics at scale, enabling studies that would be ethically or logistically challenging with human subjects. However, the field lacks standardized data resources for fine-tuning and evaluating LLMs as realistic social media agents. We address this gap by introducing SIMPACT, the SIMulation-oriented Persona and Action Capture Toolkit, a privacy respecting framework for constructing behaviorally-grounded social media datasets suitable for training agent models. We formulate next-action prediction as a task for training and evaluating LLM-based agents and introduce metrics at both the cluster and population levels to assess behavioral fidelity and stylistic realism. As a concrete implementation, we release BluePrint, a large-scale dataset built from public Bluesky data focused on political discourse. BluePrint clusters anonymized users into personas of aggregated behaviours, capturing authentic engagement patterns while safeguarding privacy through pseudonymization and removal of personally identifiable information. The dataset includes a sizable action set of 12 social media interaction types (likes, replies, reposts, etc.), each instance tied to the posting activity preceding it. This supports the development of agents that use context-dependence, not only in the language, but also in the interaction behaviours of social media to model social media users. By standardizing data and evaluation protocols, SIMPACT provides a foundation for advancing rigorous, ethically responsible social media simulations. BluePrint serves as both an evaluation benchmark for political discourse modeling and a template for building domain specific datasets to study challenges such as misinformation and polarization.

cs.CL↗

arXiv2025

Large Language Model Scaling Laws for Neural Quantum States in Quantum Chemistry

Scaling laws have been used to describe how large language model (LLM) performance scales with model size, training data size, or amount of computational resources. Motivated by the fact that neural quantum states (NQS) has increasingly adopted LLM-based components, we seek to understand NQS scaling laws, thereby shedding light on the scalability and optimal performance--resource trade-offs of NQS ansatze. In particular, we identify scaling laws that predict the performance, as measured by absolute error and V-score, for transformer-based NQS as a function of problem size in second-quantized quantum chemistry applications. By performing analogous compute-constrained optimization of the obtained parametric curves, we find that the relationship between model size and training time is highly dependent on loss metric and ansatz, and does not follow the approximately linear relationship found for language models.

cs.LG↗

arXiv2025

QCD corrections to Higgs boson pair production and decay to the $b\bar{b}τ^+τ^-$ final state

We present a comprehensive investigation of next-to-leading order (NLO) quantum chromodynamics (QCD) corrections to Higgs boson pair production and the subsequent decay into $b\bar{b}τ^+ τ^-$. We adopt the narrow-width approximation to separate the production and decay contributions and employ the dipole subtraction method to address infrared divergences inherent in perturbative QCD corrections. After applying a set of typical experimental cuts, we show that NLO QCD corrections to the decay process induce a significant reduction of the fiducial cross section and reshape important kinematic distributions at the 13.6 TeV LHC, such as the invariant mass of the Higgs boson pair and the transverse momentum of the leading $b$-jet. We also investigate the dependence of the kinematic distributions on the Higgs self-coupling and provide the signal acceptance as a function of the Higgs self-coupling modifier with full QCD corrections.

hep-ph↗

arXiv2025

Semantic-aware Graph-guided Behavior Sequences Generation with Large Language Models for Smart Homes

As smart homes become increasingly prevalent, intelligent models are widely used for tasks such as anomaly detection and behavior prediction. These models are typically trained on static datasets, making them brittle to behavioral drift caused by seasonal changes, lifestyle shifts, or evolving routines. However, collecting new behavior data for retraining is often impractical due to its slow pace, high cost, and privacy concerns. In this paper, we propose SmartGen, an LLM-based framework that synthesizes context-aware user behavior data to support continual adaptation of downstream smart home models. SmartGen consists of four key components. First, we design a Time and Semantic-aware Split module to divide long behavior sequences into manageable, semantically coherent subsequences under dual time-span constraints. Second, we propose Semantic-aware Sequence Compression to reduce input length while preserving representative semantics by clustering behavior mapping in latent space. Third, we introduce Graph-guided Sequence Synthesis, which constructs a behavior relationship graph and encodes frequent transitions into prompts, guiding the LLM to generate data aligned with contextual changes while retaining core behavior patterns. Finally, we design a Two-stage Outlier Filter to identify and remove implausible or semantically inconsistent outputs, aiming to improve the factual coherence and behavioral validity of the generated sequences. Experiments on three real-world datasets demonstrate that SmartGen significantly enhances model performance on anomaly detection and behavior prediction tasks under behavioral drift, with anomaly detection improving by 85.43% and behavior prediction by 70.51% on average. The code is available at https://github.com/horizonsinzqs/SmartGen.

cs.AI↗

arXiv2025

Improved constraints on Higgs boson self-couplings with quartic and cubic power dependencies of the cross section

Precise determination of the Higgs boson self-couplings is essential for understanding the mechanism underlying electroweak symmetry breaking. However, owing to the limited number of Higgs boson pair events at the LHC, only loose constraints have been established to date. Current constraints are based on the assumption that the cross section is a quadratic function of the trilinear Higgs self-coupling within the $κ$ framework. Incorporating higher-order quantum corrections from virtual Higgs bosons would significantly alter this functional form, introducing new quartic and cubic power dependencies on the trilinear Higgs self-coupling. To derive this new functional form, we propose a specialized renormalization procedure that tracks all Higgs self-couplings at each calculation step. Additionally, we introduce renormalization constants for coupling modifiers within the $κ$ framework to ensure the cancellation of all ultraviolet divergences. With new functional forms of the cross sections in both the gluon-gluon fusion and vector boson fusion channels, the upper limit of $κ_{λ_{\rm 3H}}=λ_{\rm 3H}/λ_{\rm 3H}^{\rm SM}$ set by the ATLAS (CMS) collaboration is reduced from 6.6 (6.49) to 5.5 (5.39). However, extracting a meaningful constraint on the quartic Higgs self-coupling $λ_{\rm 4H}$ from Higgs boson pair production data remains challenging. We also present the invariant mass distributions of the Higgs boson pair at different values of $κ_λ$, which could assist in setting optimal cuts for experimental analysis.

hep-ph↗

arXiv2025

Event-Triggered Resilient Consensus of Networked Euler-Lagrange Systems Under Byzantine Attacks

The resilient consensus problem is investigated in this paper for a class of networked Euler-Lagrange systems with event-triggered communication in the presence of Byzantine attacks. One challenge that we face in addressing the considered problem is the inapplicability of existing resilient decision algorithms designed for one-dimensional multi-agent systems. This is because the networked Euler-Lagrange systems fall into the category of multi-dimensional multi-agent systems with coupling among state vector components. To address this problem, we propose a new resilient decision algorithm. This algorithm constructs auxiliary variables related to the coordinative objectives for each normal agent, and transforms the considered resilient consensus problem into the consensus problem of the designed auxiliary variables. Furthermore, to relax the constraints imposed on Byzantine agent behavior patterns within continuous-time scenarios, the event-triggered communication scheme is adopted. Finally, the effectiveness of the proposed algorithm is demonstrated through case studies.

math.OC↗

arXiv2025

The Challenge of Teaching Reasoning to LLMs Without RL or Distillation

Reasoning-capable language models achieve state-of-the-art performance in diverse complex tasks by generating long, explicit Chain-of-Thought (CoT) traces. While recent works show that base models can acquire such reasoning traces via reinforcement learning or distillation from stronger models like DeepSeek-R1, previous works demonstrate that even short CoT prompting without fine-tuning is able to improve reasoning. We ask whether long CoT can be induced in a base model using only prompting or minimal tuning. Using just 20 long CoT examples from the reasoning model \texttt{QwQ-32B-Preview}, we lightly fine-tune the base model \texttt{Qwen2.5-32B}. The resulting model outperforms the much larger \texttt{Qwen2.5-Math-72B-Instruct}, showing that a handful of high-quality examples can unlock strong reasoning capabilities. We further explore using CoT data from non-reasoning models and human annotators, enhanced with prompt engineering, multi-pass editing, and structural guidance. However, neither matches the performance of reasoning model traces, suggesting that certain latent qualities of expert CoT are difficult to replicate. We analyze key properties of reasoning data, such as problem difficulty, diversity, and answer length, that influence reasoning distillation. While challenges remain, we are optimistic that carefully curated human-written CoT, even in small quantities, can activate reasoning behaviors in base models. We release our human-authored dataset across refinement stages and invite further investigation into what makes small-scale reasoning supervision so effective.

cs.AI↗

arXiv2025

Layer by Layer: Uncovering Hidden Representations in Language Models

From extracting features to generating text, the outputs of large language models (LLMs) typically rely on the final layers, following the conventional wisdom that earlier layers capture only low-level cues. However, our analysis shows that intermediate layers can encode even richer representations, often improving performance on a range of downstream tasks. To explain and quantify these hidden-layer properties, we propose a unified framework of representation quality metrics based on information theory, geometry, and invariance to input perturbations. Our framework highlights how each layer balances information compression and signal preservation, revealing why mid-depth embeddings can exceed the last layer's performance. Through extensive experiments on 32 text-embedding tasks across various architectures (transformers, state-space models) and domains (language, vision), we demonstrate that intermediate layers consistently provide stronger features, challenging the standard view on final-layer embeddings and opening new directions on using mid-layer representations for more robust and accurate representations.

cs.LG↗

arXiv2025

FEAT: Full-Dimensional Efficient Attention Transformer for Medical Video Generation

Synthesizing high-quality dynamic medical videos remains a significant challenge due to the need for modeling both spatial consistency and temporal dynamics. Existing Transformer-based approaches face critical limitations, including insufficient channel interactions, high computational complexity from self-attention, and coarse denoising guidance from timestep embeddings when handling varying noise levels. In this work, we propose FEAT, a full-dimensional efficient attention Transformer, which addresses these issues through three key innovations: (1) a unified paradigm with sequential spatial-temporal-channel attention mechanisms to capture global dependencies across all dimensions, (2) a linear-complexity design for attention mechanisms in each dimension, utilizing weighted key-value attention and global channel attention, and (3) a residual value guidance module that provides fine-grained pixel-level guidance to adapt to different noise levels. We evaluate FEAT on standard benchmarks and downstream tasks, demonstrating that FEAT-S, with only 23\% of the parameters of the state-of-the-art model Endora, achieves comparable or even superior performance. Furthermore, FEAT-L surpasses all comparison methods across multiple datasets, showcasing both superior effectiveness and scalability. Code is available at https://github.com/Yaziwel/FEAT.

cs.CV↗

arXiv2025

Origin of local magnetic exchange interaction in infiite-layer nickelates

Significant magnetic exchange interactions have been observed in infinite-layer nickelates RNiO2 (R = La, Pr, Nd), which exhibit unconventional superconductivity upon hole doping. Despite their structural and Fermi surface similarities to cuprates, infinite-layer nickelates possess a larger charge transfer gap, which influences their magnetic exchange interactions via oxygen. In this work, we performed 17O nuclear magnetic resonance (NMR) measurements on LaNiO2 and Sr-doped LaNiO2, revealing glassy spin dynamics originating from Ni-O planes. This indicates that infinite-layer nickelates are in proximity to magnetic ordering and that magnetic correlations play a crucial role in their physics. More importantly, our analysis of the Knight shift and hyperfine coupling of 17O nuclei revealed that the Ni-Ni superexchange interaction, mediated by the σ bond between the Ni-dx2-y2 and O-p orbitals, is one order of magnitude weaker than that in cuprates. This alone cannot account for the total magnetic exchange interaction observed in nickelates. First-principles many-body calculations indicate that an interstitial s orbital near the Fermi level, coupled with the Ni-d3z2-r2 orbital, significantly enhances the superexchange interaction. This contrasts with cuprates, where magnetic interactions are predominantly governed by Cu-dx2-y2 superexchange via oxygen. Our findings provide new insights into the distinct magnetic interactions in infinite-layer nickelates and their potential role in unconventional superconductivity.

cond-mat.supr-con↗

arXiv2025

FedORGP: Guiding Heterogeneous Federated Learning with Orthogonality Regularization on Global Prototypes

Federated Learning (FL) has emerged as an essential framework for distributed machine learning, especially with its potential for privacy-preserving data processing. However, existing FL frameworks struggle to address statistical and model heterogeneity, which severely impacts model performance. While Heterogeneous Federated Learning (HtFL) introduces prototype-based strategies to address the challenges, current approaches face limitations in achieving optimal separation of prototypes. This paper presents FedORGP, a novel HtFL algorithm designed to improve global prototype separation through orthogonality regularization, which not only encourages intra-class prototype similarity but also significantly expands the inter-class angular separation. With the guidance of the global prototype, each client keeps its embeddings aligned with the corresponding prototype in the feature space, promoting directional independence that integrates seamlessly with the cross-entropy (CE) loss. We provide theoretical proof of FedORGP's convergence under non-convex conditions. Extensive experiments demonstrate that FedORGP outperforms seven state-of-the-art baselines, achieving up to 10.12\% accuracy improvement in scenarios where statistical and model heterogeneity coexist.

cs.LG↗

arXiv2025

Higgs boson pair production and decay at NLO in QCD: the $b\bar{b}γγ$ final state

The Higgs boson pair production at the LHC provides a probe to the Higgs boson self-coupling. The higher-order QCD corrections in this process are sizable and must be taken into account in comparison with data. Due to the small cross section, it is necessary to consider at least one of the Higgs bosons decaying to bottom quarks. The QCD corrections to the decay processes would also be important in such cases. We present a full calculation of the total and differential cross sections for the $b\bar{b}γγ$ final state with next-to-leading order (NLO) QCD corrections. After applying typical kinematic cuts in the final state, we find that QCD NLO corrections in the decay decrease the LO result by $19\%$ and reduce the scale uncertainties by a factor of two. The QCD corrections to the invariant mass $m_{jjγγ}$ distribution, the transverse momentum spectra of the leading bottom quark jet and photon are significant and can not be approximated by a constant factor.

hep-ph↗

arXiv2025

Comparative study of the ansätze in quantum language models

Quantum language models are the alternative to classical language models, which borrow concepts and methods from quantum machine learning and computational linguistics. While several quantum natural language processing (QNLP) methods and frameworks exist for text classification and generation, there is a lack of systematic study to compare the performance across various ansätze, in terms of their hyperparameters and classical and quantum methods to implement them. Here, we evaluate the performance of quantum natural language processing models based on these ansätze at different levels in text classification tasks. We perform a comparative study and optimize the QNLP models by fine-tuning several critical hyperparameters. Our results demonstrate how the balance between simplification and expressivity affects model performance. This study provides extensive data to improve our understanding of QNLP models and opens the possibility of developing better QNLP algorithms.

quant-ph↗

arXiv2025

Pressure-enhanced spin-density-wave transition in double-layer nickelate $La_{3}Ni_{2}O_{7-δ}$

Recently, a signature of high-temperature superconductivity above the liquid nitrogen temperature (77 K) was reported for $La_{3}Ni_{2}O_{7-δ}$ under pressure. This finding immediately stimulated intense interest in the possible mechanism of high-$T_{c}$ superconductivity in double-layer nickelates. Notably, the pressure-dependent phase diagram inferred from transport measurements indicates that the superconductivity under high pressure emerges from the suppression of density-wave-like order at ambient pressure, which is similar to high-temperature superconductors. Here, nuclear magnetic resonance (NMR) spectroscopy of $^{139}La$ nuclei was performed to study the density-wave-like transition in a single crystal of $La_{3}Ni_{2}O_{7-δ}$. At high temperatures, two sets of sharp $^{139}La$ NMR peaks are clearly distinguishable from a broad background signals, which are ascribed to La(1) sites from two bilayer Ruddlesden-Popper phases with different oxygen vacancy $δ$. As the temperature decreases, the temperature-dependent $^{139}La$ NMR spectra and nuclear spin-lattice relaxation rate $(1/T_{1}$) for both La(1) sites provide evidence of spin-density-wave (SDW) ordering below the transition temperature ($T_{SDW}$), which is ~ 150 K. The anisotropic splitting in the NMR spectra suggests the formation of a possible double spin stripe with magnetic moments aligned along the c-axis. Furthermore, we studied the pressure-dependent SDW transition up to ~ 2.7 GPa. Surprisingly, the $T_{SDW}$ inferred from NMR measurements of both La(1) sites increases with increasing pressure, which is opposite to the results from previous transport measurements under pressure and suggests an intriguing phase diagram between superconductivity and SDW. All these results will be helpful for building a connection between superconductivity and magnetic interactions in double-layer nickelates.

cond-mat.supr-con↗

arXiv2025

VideoWebArena: Evaluating Long Context Multimodal Agents with Video Understanding Web Tasks

Videos are often used to learn or extract the necessary information to complete tasks in ways different than what text and static imagery alone can provide. However, many existing agent benchmarks neglect long-context video understanding, instead focusing on text or static image inputs. To bridge this gap, we introduce VideoWebArena (VideoWA), a benchmark for evaluating the capabilities of long-context multimodal agents for video understanding. VideoWA consists of 2,021 web agent tasks based on manually crafted video tutorials, which total almost four hours of content. For our benchmark, we define a taxonomy of long-context video-based agent tasks with two main areas of focus: skill retention and factual retention. While skill retention tasks evaluate whether an agent can use a given human demonstration to complete a task efficiently, the factual retention task evaluates whether an agent can retrieve instruction-relevant information from a video to complete a task. We find that the best model achieves 13.3% success on factual retention tasks and 45.8% on factual retention QA pairs, far below human performance at 73.9% and 79.3%, respectively. On skill retention tasks, long-context models perform worse with tutorials than without, exhibiting a 5% performance decrease in WebArena tasks and a 10.3% decrease in VisualWebArena tasks. Our work highlights the need to improve the agentic abilities of long-context multimodal models and provides a testbed for future development with long-context video agents.

cs.CV↗