Searcharxiv⌕ Search

arXiv subjects

Qi Guo

Publications and source records attributed to Qi Guo.

At least 19 recordsLinked to original sources
arXiv2026

Inter-Agent Influence: Evaluating Persuasion, Deception, and Coercion in Multi-Agent Systems

The deployment of AI agents in multi-agent settings enables inter-agent influence, whereby agents can strategically steer each other's behaviour towards an intended outcome. Such capabilities pose novel risks, as malicious agents could direct others towards harmful actions. In this work we investigate three such capabilities - persuasion, deception, and coercion - across five realistic evaluation environments and a variety of agents instantiated with frontier language models. We observed significant inter-agent influence capabilities; in oversight environments, all tested models shifted overseer decisions from rejection to approval through persuasion, deception, or coercion. We also observed successful influence in non-hierarchical 'peer-to-peer' settings through redirection of a peer's safety research trajectory. As multi-agent deployments become increasingly commonplace, these results point to a pressing need for work on mitigating the risks from inter-agent influence. Code and data are available at https://github.com/Cooperative-AI-Foundation/inter-agent-influence.

cs.AI↗
arXiv2026

Predicting Primordial Non-Gaussianity Measurements in the CSST Spectroscopic Survey

Primordial non-Gaussianity (PNG) is a fundamental probe of the physics of the early Universe and inflation. Here we present a comprehensive study of the constraints on the local-type PNG parameter, $f_{\rm NL}$, for the spectroscopic galaxy survey of the upcoming Chinese Space-station Survey Telescope (CSST). Utilizing the high-resolution Jiutian N-body simulation suite, we construct realistic mock catalogs for emission line galaxies (ELGs) at three representative redshifts $z=0.3$, 0.6, and 0.9. The expected CSST observational characteristics are also considered, including redshift uncertainties and selection functions based on signal-to-noise ratios of emission lines. We develop a robust analysis framework for the redshift-space galaxy power spectrum and bispectrum that accounts for redshift-space distortions, scale-dependent bias, and nonlinear effects. For the parameter $p$ in the PNG bias $b_ϕ$ with $p=0.55$, we find $f_{\rm NL}=-4\pm43$ if only using the power spectrum data, and $f_{\rm NL}=3\pm41$ after combining the bispectrum data, reducing the uncertainty by $\sim$5%.For the $p=1$ case, the constraint results are $f_{\rm NL}=-3\pm86$ and $54^{+70}_{-80}$ for comparison. This work demonstrates the potential of the Stage IV surveys like CSST to probe inflationary physics, and highlights the importance of higher-order statistics in extracting information from large-scale structure surveys.

astro-ph.CO↗
arXiv2026

RTL-Interact: Training EDA-Interactive Models for Multi-Task RTL Design

Large language models offer a promising route to register-transfer-level (RTL) design, but reliable implementations require checking and revision based on electronic design automation (EDA) tools' feedback. Existing RTL-specialized models often lack learned interaction with such EDA tools or rely on separate generation and repair components. We introduce RTL-Interact, a multi-task post-training framework that trains a single RTL-specialized model to generate RTL, construct task-appropriate EDA checks, and revise implementations using tool feedback. The framework combines automatically constructed multi-task demonstrations with staged supervision from chain-of-thought to tool-integrated reasoning. We further introduce Pattern-Aware Policy Optimization (PAPO) for multi-task reinforcement learning. PAPO augments Group Relative Policy Optimization with correctness-gated rewards that favor concise successful checking while rewarding effective revision, and uses task grouping to preserve coverage across objectives. On CVDP, RTL-Interact-32B achieves 48.7% weighted Pass@1 and 68.9% Pass@5, surpassing its teacher model, DeepSeek-V3.2, while RTL-Interact-8B achieves 44.9% weighted Pass@1. Additionally, we show that PAPO can reduce average tool calls by 63.8% while improving Pass@1. Both models also outperform all RTL-specialized baselines on VerilogEval v2 and RTLLM v2, demonstrating the effectiveness of our method.

cs.AI↗
arXiv2026

Domain Decoupling Attack: Exploiting the Validation Gap Between Protective DNS and Shared Edge Routing

Command-and-control, data transfer, and censorship circumvention can hide in traffic to widely used services. DNS-bound access control grants permits based on resolved IP addresses, while shared edges route requests by tenant identity, creating a validation gap. An attacker can resolve an allowed anchor, obtain a permit for its shared edge address, and access a hidden tenant by presenting the tenant's name as both SNI and HTTP Host. To investigate this gap, we present the Domain Decoupling Attack (DDA), which exploits this gap in CDN and non-CDN hosting while preserving SNI--Host consistency. Across 1,069,048 domains, we conduct 18,025,068 successful probes. Overall, 95.8% of domains accept an external SNI, and 84.6% return content matching the named tenant. Exposure reaches 99.26% across 506,763 CDN domains and 92.75% across 562,285 non-CDN domains. The co-location graph places 666,413 exposed domains in cross-tenant components; 97.7% of 54,069 cross-tenant non-CDN addresses have a confirmed path to another tenant. None of the tested defenses bind tenant identity to the DNS-approved name. Cleartext inline IPS can block access based on SNI, but Encrypted Client Hello (ECH) hides it. The Windows Zero Trust DNS (ZTDNS) permit remains reusable in both cases, enabling the concealment of command-and-control and data-exfiltration traffic and facilitating censorship circumvention. The risk conceals command-and-control and data-exfiltration traffic and supports censorship.

cs.CR↗
arXiv2026

Constrain the Size of Broad Line Region in NGC 1068 Using Sub-millimeter Excess

Constraining the size of broad-line region (BLR) close to the accretion disk in active galactic nuclei (AGNs) is key to understand the central engine and evolution of AGN. We propose a new method to constrain the BLR size, based on millimeter to sub-millimeter spectral energy distribution (SED), from which a sub-millimeter excess may be found. Such excess can be used to determine the solid angle of the free-free emission region from BLR, as a region with high electron density and optically thick at sub-millimeter wavelength for free-free continuum. Millimeter to sub-millimeter SED analysis toward the central source S1 in a nearby type II AGN NGC 1068 was done, which gave an upper limit on the solid angle of BLR region as(4.94 +/- 0.61) *10^-6 arcsec^2, corresponding to a radius no more than 0.085 pc. Such a constraint on the BLR size is important for estimating the masses of central supermassive black holes in AGNs.

astro-ph.GA↗
arXiv2026

Inverse Garment Design with Visual Agents

We study the problem of inverse garment design, reconstructing simulation-ready 3D outfits with arbitrary layers from a single image, with visual agents. Unlike differentiable optimization that is time-consuming, or feed-forward methods limited by training data, we introduce DressAgent, a hierarchical agentic system that recovers garment sewing patterns through planning and closed-loop refinement grounded in physical simulation. To this end, we first present an enhanced representation based on GarmentCode for diverse styles and arbitrary dressing layers, and build the first multi-layer sewing pattern dataset with over 15K outfits for supervision and benchmarking. Inverse design then proceeds coarse to fine: In layer planning and coarse estimation, agents infer layer order and relations, peel each layer into its own view, and estimate coarse patterns. In fine-grained refinement, an agentic trio of Tailor, Critic, and Orchestrator refines patterns in a closed loop of editing, simulating, and judging against the reference with visual and measured evidence. Carefully designed tools and an evolving memory make this refinement converge on the target geometry. We also design an agentic generation method for high-fidelity textures. DressAgent outperforms feed-forward methods by a large margin, surpasses or matches differentiable optimization in a fraction of the time, and generalizes to complex layering, diverse poses, and body shapes.

cs.AI↗
arXiv2026

An LLM-powered Agentic Recommendation System for Connected TV Content Discovery

Recommendation systems, from traditional multi-stage to recent unified generative architectures, face challenges in incorporating diverse contextual signals, such as trending topics, breaking news, cultural events, and cross-surface user activities, into their ranking pipelines. These systems are designed to consume structured behavioral signals with consistent schemas, and lack the reasoning capability to naturally process unstructured or heterogeneously formatted contextual information. Incorporating such signals typically requires feature engineering, bespoke data pipelines, and carefully tuned heuristics. In this paper, we present an LLM-powered agentic recommendation system designed for Connected TV (CTV) content discovery that addresses these limitations. Our system leverages the reasoning capabilities of large language models to naturally process and synthesize diverse signals across varying schemas and structures, eliminating much of the manual integration inherent in traditional ranking and retrieval systems. Recognizing that current LLM-based solutions still fall short of traditional machine learning models in several recommendation tasks, including retrieval efficiency, personalization precision, and scalability, we adopt an agentic architecture that orchestrates specialized components, allowing each sub-task to be handled by the most suitable method, whether LLM-based or traditional ML. The main contribution of this work is our engineering approach to successfully overcoming the practical limitations of enabling LLM for recommendation, particularly inference latency. We share insights from our work and discuss the trade-offs and lessons learned in building a hybrid system that combines the flexibility of LLMs with the performance of established recommendation techniques.

cs.IR↗
arXiv2026

MHR-Router: Multimodal Heterogeneous Retrieval-Enhanced Adaptive Routing for VLM Inference

Vision-language models (VLMs) exhibit complementary capabilities, yet selecting the most suitable expert for a given image-question query remains challenging due to heterogeneous multimodal inputs and varying expert strengths. Existing routing approaches rely on text-based cues or static model profiles, failing to capture fine-grained visual evidence and expert-specific failure patterns. To address this, we propose MHR-Router, a multimodal heterogeneous retrieval-enhanced adaptive routing framework for VLM inference. It accurately dispatches each multimodal query to the most suitable VLM expert via local retrieval evidence. MHR-Router consists of two stages: (1) Multimodal Routing Representation, which fuses frozen multimodal features into a comprehensive query representation; and (2) Heterogeneous Retrieval-Enhanced Adaptive Routing, which retrieves expert-specific success and failure memories as local evidence and employs a shared scorer to adaptively select the optimal expert. Experiments show that MHR-Router effectively achieves heterogeneous modality routing and consistently improves accuracy over traditional baselines.

cs.LG↗
arXiv2026

MetaEncoder: Exploring the Limit of Bi-Encoders for Multimodal System One Decision Making with Natural Language Interface

System One models output constrained decisions and probability distributions rather than free-form text generation. While prevailing paradigms rely on structured schema objects to encode state, intent, and candidate choices, we revisit a fully natural language-based System One interface. In this framework, both the user request and each candidate option are expressed in natural language, supported by multimodal (image and video) auxiliary inputs. We introduce MetaEncoder, which fine-tunes a pre-trained Muse-Glimmer 30B decoder into an instruction-following decision-making encoder. To scale effectively across both small closed-set (< 256) and massive open-set (millions) candidate spaces, MetaEncoder employs a bi-encoder architecture trained via unidirectional contrastive learning for request-candidate alignment. We conduct extensive evaluations across 11 benchmark suites and 190 tasks spanning multimodal decision-making, understanding (closed-set) and retrieval (open-set), highlighting where MetaEncoder beats SOTA multimodal encoders, as well as its current limits on reasoning-intensive tasks.

cs.CL↗
arXiv2026

PersonaMem-v3: Toward Omni-Platform Personal Intelligence for Holistic User Understanding, Recommendation, and Agentic Tasks

Personal intelligence is becoming a central frontier for user-facing AI agents. To be helpful in everyday life, agents must understand users across the digital contexts where their preferences, intents, habits, social relationships, and needs unfold over time. Today's systems can personalize within individual apps or tasks, but personal intelligence as a whole remains under-measured: how agents build cross-context user understanding, support steerable recommendation systems, act proactively across platforms, and avoid over-personalization. We introduce PersonaMem-v3, a real-world-grounded benchmark and evaluation harness for omni-platform personal intelligence. PersonaMem-v3 is seeded from more than one million anonymized real-world engagement histories, most of which are implicit signals, and uses them to construct time-indexed user digital worlds across social media, chatbot, calendar, and AI-companion with preference evolvement over time. The benchmark brings personalization, LLM-powered recommendation, proactiveness, agentic tool use, and geo-temporal reasoning into one framework, anchored in psychology, social-linguistics, and user-behavior theories. It evaluates whether AI agents can infer holistic user understanding from cross-platform evidence, personalize responses, rerank recommendations on social media, follow user steering through natural language, and hold back when personalization would be inappropriate, repetitive, outdated, or unnecessary. PersonaMem-v3 points toward LLM-powered personal intelligent agents that work with existing scalable recommendation infrastructure while making personalization more interactive, agentic, and aligned with how real users experience their digital lives.

cs.CY↗
arXiv2026

GISTBench: Evaluating LLM User Understanding via Evidence-Based Interest Verification

We introduce GISTBench, a benchmark for evaluating Large Language Models' (LLMs) ability to understand users from their interaction histories in recommendation systems. Unlike traditional RecSys benchmarks that focus on item prediction accuracy, our benchmark evaluates how well LLMs can extract and verify user interests from engagement data. We propose two novel metric families: Interest Groundedness (IG), decomposed into precision and recall components to separately penalize hallucinated interest categories and reward coverage, and Interest Specificity (IS), which assesses the distinctiveness of verified LLM-predicted user profiles. We release a synthetic dataset constructed on real user interactions on a global short-form video platform. Our dataset contains both implicit and explicit engagement signals and rich textual descriptions. We validate our dataset fidelity against user surveys, and evaluate eight open-weight LLMs spanning 7B to 235B parameters, together with three proprietary frontier models (GPT-5, Claude 4.6, and Gemini 3.5 Flash). Our findings reveal performance bottlenecks in current LLMs, particularly their limited ability to accurately count and attribute engagement signals across heterogeneous interaction types.

cs.AI↗
arXiv2026

Cross-correlating Squared kSZ and HI Intensity Fields: A Map-level ACT--MeerKAT Forecast

The kinetic Sunyaev--Zel'dovich (kSZ) effect probes the line-of-sight momentum of free electrons, but its blackbody spectrum and velocity-dependent sign make it difficult to isolate and cause conventional cross-correlations with density tracers to vanish. In this work, we investigate the cross-correlation of the squared kSZ field with the projected squared HI intensity field using mock ACT and MeerKAT observations constructed from a Jiutian-1G simulation lightcone. We generate halo-based kSZ maps and MeerKAT-like HI brightness-temperature cubes over a $13.5^\circ\times13.5^\circ$ field. The simulations incorporate ACT beam smoothing and effective internal-linear-combination noise, together with the frequency-dependent MeerKAT beam, scanning strategy, map-making, and inhomogeneous thermal noise. We analyze signal-only, idealized amplitude-reconstructed, and Wiener-filtered kSZ maps. The resulting squared-field cross-correlations have diagnostic cumulative signal-to-noise ratios of approximately $3$-$4$. Our map-level results demonstrate the detectability of the ${\rm kSZ}^2\times {\rm HI}^2$ correlation in future joint ACT--MeerKAT observations.

astro-ph.CO↗
arXiv2026

KernelZero: Co-Evolving Proposer and Coder for Continuously Improved GPU Kernel Generation

High-performance GPU kernels are essential to modern machine learning systems, yet automatically generating kernels that are both correct and efficient remains challenging. Existing LLM-based approaches face two major limitations: the scarcity of high-quality training data aligned with the model's current capabilities, and the inherent trade-off between kernel correctness and performance. To address these challenges, we propose KernelZero, a co-evolution framework that continuously improves GPU kernel generation through two specialized models: a Proposer that generates Torch modules from API sets and a Coder that translates them into CUDA or Triton kernels. KernelZero uses a frontier-driven module generation mechanism to continuously produce capability-aligned training modules based on the Coder's current weaknesses. It further introduces Correctness-Aware Group Relative Policy Optimization (CA-GRPO), which optimizes performance only after correctness becomes sufficiently reliable. By alternating the optimization of the Proposer and Coder, KernelZero forms an automatic curriculum that enables targeted and training-efficient capability improvement. Empirically, KernelZero-7B surpasses Claude-4.5-Sonnet on CUDA and DeepSeek-V4-Pro on Triton. On KernelBench Level 1 and 2, it achieves CUDA pass@1 scores of 75.8% and 69.6%, respectively, with pass@10 reaching 100% and 97%. On Triton, it achieves pass@1 scores of 77.2% and 72.5%, respectively.

cs.LG↗
arXiv2026

Uniqueness and Nondegeneracy of Ground State for the $L^2$-Critical Boson Star Equation

We prove uniqueness, up to phase rotations and translations, of the ground state for the $L^2$-critical boson star equation \[ (\sqrt{-Δ}+1)Q =\bigl(|x|^{-1}* |Q|^2\bigr)Q, \qquad x\in\mathbb R^3. \] We also determine the kernels of the real and imaginary linearized operators. As a consequence, we show the Lieb-Yau uniqueness conjecture holds when the mass is sufficiently close to the Chandrasekhar mass.

math.AP↗
arXiv2026

TTE-Flash: Accelerating Reasoning-based Multimodal Representations via Think-Then-Embed Tokens

Recent research has demonstrated that Universal Multimodal Embedding (UME) benefits significantly from Chain-of-Thought (CoT) reasoning. In this paradigm, a generative model produces explicit reasoning traces for a multimodal query, with the final representation extracted from an embedding token attending to both the query and the reasoning. Despite its effectiveness, the computational overhead of generating explicit CoT traces is often prohibitive. In this work, we propose replacing explicit CoT with latent think tokens, which are interpreted as latent variables that can produce explicit CoT traces as observed variables. By optimizing think tokens using CoT generation loss and subsequent embedding tokens using contrastive loss, we produce high-performance, reasoning-aware representations at a constant inference cost. Our study investigates two key architectural designs: 1) how think and embeddings tokens should be extracted from the same LLM backbone. 2) how the tokens should be trained as two dependent tasks. We introduce TTE-Flash-2B, a reasoning-aware multimodal representation model that outperforms its explicit-CoT counterpart on the MMEB-v2 benchmark, while producing latent think tokens that are interpretable both textually and visually. Furthermore, zero-shot evaluation across 15 video datasets reveals scaling behavior as the number of think tokens increases.

cs.AI↗
arXiv2026

Uniqueness, Nondegeneracy, and Asymptotic Expansions of Ground States for the Dirac-Coulomb System

We study ground states of the following Dirac-Coulomb system $$\mathscr{D}_cu_c- \left(\abs{x}^{-1}\ast \abs{u_c}^2\right)u_c=ω_cu_c,$$ under the constraint $\norm[L^2]{u_c}^2=1.$For sufficiently large \(c\), we prove uniqueness of ground states modulo translations, phase transformations and spinorial rotations, and determine the kernel of the linearized operator at ground state. We also obtain asymptotic expansions of ground state in the nonrelativistic limit regime.

math.AP↗
arXiv2026

SplitLite: Low-Rank Residual Compression for Split Learning

Federated fine-tuning of on-device large language models (LLMs) faces a significant computing burden. To overcome this limitation, split learning (SL) has emerged as a promising solution, which offloads the primary training workload to a powerful server. However, SL requires exchanging high-dimensional activations and gradients between clients and the server, resulting in prohibitive communication costs. To overcome this challenge, we propose SplitLite, a communication-efficient split federated LoRA fine-tuning method that exploits the low effective rank structure of consecutive-epoch activation and gradient residuals. Our key finding is that, when LoRA uses rank $r$ updates in parameter space, the activation and gradient residuals of the same data sample between adjacent epochs also exhibit effective rank-$2r$ and rank-$4r$ structures, respectively. By revealing this property, SplitLite transmits only quantized truncated singular value decomposition (SVD) residual factors, thereby significantly reducing both activation uplink and gradient downlink traffic. Extensive experiments on the GLUE benchmark across a series of advanced on-device LLMs demonstrate that our method reduces activation uplink communication costs by up to 93.5\% and total communication costs by up to 83.7\%, without performance degradation.

cs.LG↗
arXiv2026

PIFFLE: Characterizing the Foreground Contributions from 4 Decades in Halo Mass to the FRB20230907D Dispersion Measure

We characterize the foreground environment of FRB20230907D, localized to a galaxy at $z=0.464$, which has an observed dispersion measure of ${\rm DM}_{\rm obs}=1031~{\rm pc~cm^{-3}}$. At its redshift, FRB20230907D lies above the Macquart relation, the expected relation between cosmological dispersion measure and the source redshift, indicating a substantial excess DM along this line of sight. We use Subaru/PFS and SDSS spectroscopy, published group catalogs, Rubin/LSST imaging, and eROSITA X-ray data to characterize the foreground structures that may account for this excess. A friends-of-friends search identifies a massive foreground system at $z\simeq0.09$ with $M_{200}\simeq5.2\times10^{14}~M_\odot$, while low redshift catalogs reveal an additional group at $z\simeq0.02565$. Assuming that the halo gas follows a modified-NFW halo density profile, we estimate observer frame contributions of $150^{+110}_{-70}~{\rm pc~cm^{-3}}$ and $80^{+60}_{-40}~{\rm pc~cm^{-3}}$ from these systems, respectively. Together with the Milky Way, diffuse intergalactic medium, Virgo cluster, M49 group, and host galaxy contributions, these foreground structures can account for the excess dispersion measure of FRB20230907D within uncertainties. This highlights the importance of dense foreground spectroscopy and multi-wavelength data.

astro-ph.CO↗