Searcharxiv⌕ Search

arXiv subjects

Yue Xu

Publications and source records attributed to Yue Xu.

At least 19 recordsLinked to original sources
arXiv2026

ARASH: Adaptive Retrieval And Shot Selection for Tabular Prediction

Tabular prediction is a critical task across numerous applications. The recent success of large language models has sparked various approaches for adapting them to the tabular domain. A prevalent strategy involves training or fine-tuning specialized Tabular Foundation Models (TFMs) such as TabPFN. However, TFMs require substantial computational resources, and frequent model retraining is often impractical. In-context learning (ICL), specifically, few-shot prompting, offers a resource-efficient alternative to enhance performance. Yet, identifying the most relevant rows to serve as shots remains a challenge for tabular data. This paper introduces ARASH (Adaptive, query-specific Retrieval And Shot selection), a method that improves TFM efficiency by selecting optimal shots based on local neighborhood analysis within the training set. Our results demonstrate that ARASH reduces the prompt length and memory usage of TabPFN by 1261.5$\times$ and 2.56$\times$, respectively, while providing comparable accuracy.

cs.AI↗
arXiv2026

What to Remember, What to Reveal: Privacy-Aware Memory for Conversational Agents

Long-term memory enables personalized conversational agents to retain user information across sessions. However, existing memory architectures primarily optimize for utility while neglecting the risks of unnecessarily storing and reusing private attributes such as personally identifiable information (PII). Addressing privacy risks in personalized memory is challenging because simply removing sensitive values can undermine system utility. Therefore, privacy protection for memory agents should govern the full life cycle of sensitive values rather than only sanitizing individual records. To address this gap, we introduce Sanitized Privacy-Mapped Memory (SP-Mem), a privacy-aware memory architecture that decouples memory utility from exact private-value exposure. SP-Mem provides a full life-cycle privacy design that identifies and separates sensitive information from raw user inputs, stores sanitized content and exact private values in isolated structures, and selectively retrieves private values based on task requirements and user consent. We further introduce a privacy-aware memory benchmark that jointly evaluates response quality, privacy behavior, and inference cost. Extensive experiments across multiple LLM-based agents show that SP-Mem achieves stronger personalization while reducing unnecessary privacy exposure. Code and data are available at https://github.com/Jensassss/SP-Mem.

cs.CR↗
arXiv2026

Asymptotic independence of class-group 4-ranks in correlated pairs of imaginary quadratic fields

Fix a squarefree integer $d_0>1$, and let $d$ range over the positive squarefree integers coprime to $2d_0$. Although $\mathbb{Q}(\sqrt{-d})$ and $\mathbb{Q}(\sqrt{-d_0d})$ share all variable ramified primes, we prove that their class-group $4$-ranks are asymptotically independent. Over the subfamily $d\le X$, their joint distribution converges in total variation to the product of two copies of the Cohen--Lenstra--Gerth distribution, with error bounded by a negative power of $\log\log X$. We further conjecture that the corrected $2$-primary groups $2\operatorname{Cl}_{\mathbb{Q}(\sqrt{-d})}[2^\infty]$ and $2\operatorname{Cl}_{\mathbb{Q}(\sqrt{-d_0d})}[2^\infty]$ are asymptotically independent, each with the Cohen--Lenstra distribution. Suppose in addition that the class number of $\mathbb{Q}(\sqrt{d_0})$ is odd. For a density-one subset of this family, we prove that extension of ideals to $K(d)=\mathbb{Q}(\sqrt{d_0},\sqrt{-d})$ induces $4\operatorname{Cl}_{K(d)}[2^\infty]\cong 2\operatorname{Cl}_{\mathbb{Q}(\sqrt{-d})}[2^\infty]\oplus 2\operatorname{Cl}_{\mathbb{Q}(\sqrt{-d_0d})}[2^\infty]$. Together with this decomposition, the group-valued conjecture predicts that $4\operatorname{Cl}_{K(d)}[2^\infty]$ is distributed as the direct sum of two independent Cohen--Lenstra $2$-groups, giving a corrected Cohen--Lenstra--Martinet distribution for the biquadratic family. Unconditionally, the $8$-rank of $\operatorname{Cl}_{K(d)}$ has limiting distribution given by the convolution of two copies of the Cohen--Lenstra--Gerth distribution. The proof combines Smith's box method with quantitative truncated Gaussian-binomial moment inversion for diagonally coupled, fixed-width bordered R\'edei matrices.

math.NT↗
arXiv2026

Radio Core Size of Low-luminosity Active Galactic Nuclei under the MAD-jet model

After decades of efforts, there are now fruitful high-resolution radio observations of low-luminosity active galactic nuclei (LLAGNs), and the observed frequency has extended from $\sim$10 GHz up to $\sim$200 GHz. In this work, based on a model that combines a magnetically arrested disk (MAD) and a Blandford-Znajek-like jet, we carried out detailed analysis on size and location of the radio core of LLAGNs. The radio core size of nearby LLAGN M104 is re-visited based on this new model. We successfully reproduce a $size\propto\nu^{-1}$ scaling between 1 GHz and tens of GHz, if more than $50\%$ of electrons in jet follow a power-law (PL) distribution. We further confirm that, at high radio frequencies emission from MAD exceeds that from jet, and a flatter size-frequency slope is observed. The impact of PL electrons in MAD is also investigated. For those $L_{\rm bol}/L_{\rm Edd} \gtrsim (3-8)\times 10^{-6}$ LLAGNs and black hole binaries in their hard state, PL electrons are expected to be highly suppressed due to strong radiative cooling (so-called `synchrotron boiler' effect).

astro-ph.HE↗
arXiv2026

The Distinctive Evolution and Spectral Energy Distribution of Binary Massive Black Hole Accretion

Binary (super-)massive black holes (BHs) are expected to reside in the center of some galaxies. In this work, we re-visit accretion onto binary massive BHs, incorporating recent advances in both accretion theory and the mass transfer rate between the two massive BHs. We focus on relatively bright systems with an Eddington ratio of 0.1 for a binary with total BH mass $10^8\,M_\odot$, but consider a wide range of mass ratios $10^{-4} \le q \le 0.5$. The binary system consists of two mini-disks surrounding two individual BHs and a circumbinary disk surrounding the mass center of binary BHs. Depending on the mass ratio, the two mini-disks can be hot accretion flows, standard thin (cold) disks, or Slim disks. The radiative contributions from all three disks, each potentially in different accretion modes, are taken into account self-consistently. The spectral energy distributions of the binary BH system show universal ``notch'' features from the near-infrared to ultraviolet bands, caused by the gap or cavity in the accretion disk, consistent with previous studies. Binary with different mass ratios exhibit distinct spectral energy distribution properties, offering opportunities for testing (identifying candidates) with future broad band (infrared up to X-rays) observations. We also investigate the evolution of these binary systems, and find that, for systems with initial mass ratios $q \lesssim \text{a few} \times 10^{-3}$, the mass ratio evolves toward an equilibrium value $q \sim 10^{-3}$. For binary BH systems with a larger initial mass ratio, their mass ratio instead evolves toward unity.

astro-ph.HE↗
arXiv2026

Hypergraph Turan with bounded matching number

For a fixed graph $G$, an $r$-uniform hypergraph is said to contain a Berge-$G$ if there exists a bijection $f\colon E(G)\to E(\mathcal{H})$ for some subhypergraph $\mathcal{H}$ such that $e\subseteq f(e)$ for every $e\in E(G)$. Motivated by Alon and Frankl's study of Tur\'an problems under bounded matching constraints, we investigate the maximum number of edges in $r$-uniform Berge-$K_3$-free hypergraphs with matching number at most~$s$. We determine the exact Tur\'an numbers for the cases $r=3$ and $r=4$. For $r=3$ and $n \geq 3 s$, we prove that every $n$-vertex Berge- $K_3$-free 3-graph with matching number $s$ has at most $s(n-2 s)$ edges, and we characterize the unique extremal hypergraph attaining equality. For $r=4$ and $n \geq 4 s$, the maximum number of edges is $s\lfloor(n-2 s) / 2\rfloor$, except for the exceptional case $s=1$ and $n \equiv 1(\bmod 4)$, in which the bound is $(n-1) / 2$. As a corollary, our results recover the classical theorem of Gy\H{o}ri on Berge-$K_3$-free hypergraphs.

math.CO↗
arXiv2026

KnitID: Machine-Knitted RFID Antennas for Battery-Free Authentication, Localization and Interaction

Battery-free RFID systems offer a scalable and maintenance-free approach to interaction. We present KnitID, a machine-knitted textile RFID antenna design that enables on-body authentication, localization, and interaction. Unlike prior antenna designs, KnitID achieves a compact antenna form factor (60mm by 8mm) by integrating magnet wire into the unique loop-over-loop structure of machine knitting. This structure reduces the size of conventional loop antennas by around 90\%, while also providing 30\% longer sensing ranges than standard dipole designs with similar size on the human body. The compact form factor creates new opportunities to embed multiple RFID tags across the human body, enriching backscatter signals and supporting a broader range of battery-free on-body interactions. To demonstrate this capability, we build an interactive sleeve to support wearer authentication, spatial localization, and interaction detection. Through technical evaluations, we show the feasibility of KnitID to provide diverse and battery-free interactions on knitted user interfaces.

cs.HC↗
arXiv2026

From Passive Retrieval to Active Memory Navigation: Learning to Use Memory as a Structured Action Space

Long-term user memory is essential for personalized conversational agents, yet many memory systems still expose memory through passive retrieval interfaces, making the model a consumer of pre-selected evidence. We introduce NapMem, a framework for learning to use long-term user memory as a structured action space rather than passively retrieved context. NapMem organizes user history into a linked multi-granularity memory pyramid, where raw conversations, typed memory records, topic tracks, and user profiles are connected through provenance relations, and exposes these levels through memory tools. The agent is trained to select memory according to the query and intermediate evidence, allowing it to inspect different memory granularities before answering. Experiments on PersonaMem-v2, LongMemEval, and LoCoMo show that a NapMem agent trained with memory-tool reinforcement learning is competitive across diverse memory-intensive tasks, while evaluations on non-memory tasks suggest that the learned policy largely preserves general reasoning and tool-use abilities. Additional analyses examine storage, inference cost, tool-use behavior, and ablations over navigation, memory granularity, and RL training. Our results suggest that long-term user memory benefits from coupling structured storage with a learned policy for using memory at the appropriate granularity.

cs.AI↗
arXiv2026

TokenMem: Faithful Knowledge Injection for Frozen LLMs

Retrieval-augmented generation (RAG) enhances large language models (LLMs) with external knowledge, but suffers from knowledge conflicts: when retrieved information contradicts parametric memory, the shared self-attention pathway produces unpredictable outputs. We present TokenMem, a lightweight memory system that injects knowledge into frozen LLMs through a dedicated cross-attention channel, bypassing competition with parametric memory in the residual stream. TokenMem trains only a thin gating adapter ($\sim$3-7M parameters) via a two-phase curriculum: first learning general knowledge utilization, then strengthening faithful compliance under counterfactual knowledge. In controlled experiments on five models spanning three families (Qwen3-4B/8B/14B, LLaMA-3.1-8B, OLMo-3-7B), TokenMem achieves 69-70% Knowledge Compliance (KC) on counterfactual benchmarks, compared to 20-52% for vanilla RAG, a gap of up to 49 percentage points. Ablation studies show that the two-phase curriculum is critical: removing Phase 2 collapses KC to near-zero. Mechanistic analysis reveals that the gate adapter learns a conflict-aware, layer-specific injection strategy without explicit supervision.

cs.AI↗
arXiv2026

EgoGuide: Egocentric Guidance for Efficient Robot-Free Demonstration Collection and Learning

Robot learning from real-world demonstrations is currently constrained by data scaling. Universal Manipulation Interface (UMI) provides an efficient robot-free data collection interface, yet current UMI-style pipelines often collect redundant demonstrations and lack global scene context. To improve data efficiency, we present EgoGuide, a collection interface that records synchronized wrist and head/egocentric observations and couples them with online visual-geometric data quality guidance. We also introduce a Gated Egocentric Residual Policy for robust learning from a viewpoint-varying egocentric camera, allowing head/egocentric context to correct ambiguous local observations while preserving stable wrist-view control. Real-world experiments show that EgoGuide reduces the required number of data episodes and improves data efficiency. The residual policy further improves robustness under visual occlusion. Project Page: https://silicx.github.io/EgoGuide

cs.RO↗
arXiv2026

Tac-DINO: Learning Vision-Tactile Features with Patch Alignment

Touch is the primary medium through which humans interact with the environment. Currently, tactile learning mainly focuses on image-level pretraining or alignment. However, tactile signals correspond to local object contact, while research into scale alignment and holographic matching remains limited and proper datasets and benchmarks also lack. To bridge this gap, we first construct a data collection system to acquire a large-scale tactile dataset, with over 20 K tactile contacts from 505 real-world objects. Building on this dataset, we design a Vis-Tac Holographic Matching Benchmark to evaluate vision-tactile local-to-global alignment ability. Then we propose Vision-Tactile Patch Alignment (VTPA) methods for vision-tactile representation learning. Experiments demonstrate that these exceed the performance of methods without alignment and align with whole-object images.

cs.CV↗
arXiv2026

AetheRock: An Arm-Worn Robot Teaching System for Force-Guided Vision-Tactile Learning

Force and tactile sensing are indispensable in contact-rich manipulation. However, force-aware robot learning faces critical challenges due to the incompatible assembly of tactile and force sensors in handheld or wearable devices. To address these limitations, we first introduce AetheRock for gripper-force, vision, and tactile data collection, which is an arm-worn device featuring a modular and easily manufactured visuo-tactile sensor, GelSlim-MiniFab, at the fingertip, a resistive pressure sensor at the human finger contact region, a customized PCB module, and a wearable kit for comfortable and robust collection. Building on this, we propose ForceVT, a representation learning framework that uses force and vision to guide fidelity-agnostic tactile learning, enabling robust inference in any tactile situation. Real-world experiments show that AetheRock achieves qualified data efficiency and that ForceVT effectively alleviates inefficiencies when visuo-tactile sensors exhibit manufacturing and utilization inconsistencies. Overall, our work mitigates the limitations of gripper-force vision-tactile robot learning through innovative hardware design and algorithms.

cs.RO↗
arXiv2026

Polylogarithmic Bounds for Nested Cycles without Geometric Crossings

A problem of Erd\H{o}s asks for extremal conditions forcing edge-disjoint cycles with a prescribed nested structure. In the geometric version, the nesting is required to be noncrossing with respect to the cyclic orders. Fern\'andez, Kim, Kim and Liu proved that constant average degree forces two such cycles. We prove a polylogarithmic bound for the natural multi-layer version: for every fixed $k\ge 3$, every sufficiently large $n$-vertex graph with at least \[ C_k n(\log n)^{k-1}(\log\log n)^{k-3} \] edges contains $k$ pairwise edge-disjoint nested cycles without geometric crossings. The proof combines the robust sublinear expander framework of Alon, Buci\'c, Sauermann, Zakharov and Zamir with a controlled wrapping lemma that permits the layers to be built successively with controlled length.

math.CO↗
arXiv2026

Boundary-supported radial layering in Hoag-like ring galaxies

Clean Hoag-like ring galaxies are often characterized by an old compact central component, a depleted gap, and a detached outer ring. We identify a boundary-supported radial-layering mechanism in a shell-deformed Kepler control model. A compact inner boundary supplies the core state, while a localized effective shell deformation, interpreted as the reduced imprint of externally supplied material settled near a finite circularization radius, needs to create only an internal maximum and a subsequent outer minimum. These act as the gap barrier and ring-supporting well. The onset of this structure is organized by a saddle-node threshold of the critical-point equation. In a 10^4-point Monte Carlo scan, shell-localized boundary-supported candidates occupy finite parameter volume under the adopted priors, and none of the localized candidates contains an ordered interior minimum--maximum--minimum subsequence. The same branch gives a scale-free gap-to-ring interval overlapping representative ratios for Hoag's Object, UGC 4599, and PGC 1000714, but not for the environmentally processed comparison object JO171.

astro-ph.GA↗
arXiv2026

EVA: Editing for Versatile Alignment against Jailbreaks

Large Language Models (LLMs) and Vision Language Models (VLMs) have demonstrated impressive capabilities but remain vulnerable to jailbreaking attacks, where adversaries exploit textual or visual triggers to bypass safety guardrails. Recent defenses typically rely on safety fine-tuning or external filters to reduce the model's likelihood of producing harmful content. While effective to some extent, these methods often incur significant computational overheads and suffer from the safety utility trade-off, degrading the model's performance on benign tasks. To address these challenges, we propose EVA (Editing for Versatile Alignment against Jailbreaks), a novel framework that pioneers the application of direct model editing for safety alignment. EVA reframes safety alignment as a precise knowledge correction task. Instead of retraining massive parameters, EVA identifies and surgically edits specific neurons responsible for the model's susceptibility to harmful instructions, while leaving the vast majority of the model unchanged. By localizing the updates, EVA effectively neutralizes harmful behaviors without compromising the model's general reasoning capabilities. Extensive experiments demonstrate that EVA outperforms baselines in mitigating jailbreaks across both LLMs and VLMs, offering a precise and efficient solution for post-deployment safety alignment.

cs.CR↗
arXiv2026

Pyramid Self-Contrastive Learning for Single-shot Test-time Ultrasound Image Denoising

The inherent electronic and speckle noise complicates clinical interpretation of ultrasound images. Conventional denoising methods rely on explicit noise assumptions whose validity diminishes under composite noise conditions. Learning-based methods are usually pretrained in a limited image domain using a labeled dataset, which implies inevitable domain shift in complex in vivo environments. This study proposes a Pyramid Self-Contrastive Learning (PSCL) framework for test-time ultrasound image denoising without pretraining. Given multiple noisy samples from only one-shot imaging, PSCL disentangles anatomical similarity and noise randomness into separate pyramid latent spaces. The clean image is then decoded from the anatomy space while discarding the noise space. We first apply PSCL to synthetic aperture ultrasound (SAU), where an Aperture-to-Aperture loop serves as a self-supervised proxy task to ensure denoising fidelity. Simulation experiments, including noise levels from 0 to 30 dB and inclusion geometries from simple to complex, demonstrated improvements of 69.3% in SNR and 34.4% in CNR. The in vivo results showed 84.8% SNR and 25.7% CNR gains using only two aperture data of the heart in six echocardiographic views, liver, and kidney. PSCL delivers clear images across diverse imaging targets and configurations, paving the way for more reliable anatomical visualization without domain shift and pretraining costs.

cs.CV↗
arXiv2026

SEAR: Schema-Based Evaluation and Routing for LLM Gateways

Evaluating production LLM responses and routing requests across providers in LLM gateways requires fine-grained quality signals and operationally grounded decisions. To address this gap, we present SEAR, a schema-based evaluation and routing system for multi-model, multi-provider LLM gateways. SEAR defines an extensible relational schema covering both LLM evaluation signals (context, intent, response characteristics, issue attribution, and quality scores) and gateway operational metrics (latency, cost, throughput), with cross-table consistency links across around one hundred typed, SQL-queryable columns. To populate the evaluation signals reliably, SEAR proposes self-contained signal instructions, in-schema reasoning, and multi-stage generation that produces database-ready structured outputs. Because signals are derived through LLM reasoning rather than shallow classifiers, SEAR captures complex request semantics, enables human-interpretable routing explanations, and unifies evaluation and routing in a single query layer. Across thousands of production sessions, SEAR achieves strong signal accuracy on human-labeled data and supports practical routing decisions, including large cost reductions with comparable quality.

cs.DB↗
arXiv2026

Time-dependent photospheric radiative transfer in structured GRB jets: spectral evolution and polarization diagnostics

Photospheric emission from relativistic gamma-ray burst (GRB) jets is a promising mechanism for producing the Band-like spectra observed in the prompt phase, yet the connections between jet structure, dissipation location, and polarization signatures remain unclear.We investigate time-dependent photospheric radiation transfer in structured relativistic jets by coupling two-dimensional axisymmetric special relativistic hydrodynamic (SRHD) simulations with Monte Carlo photon propagation.Photon escape and subphotospheric dissipation are characterized using the residual line-of-sight optical depth tau_out evaluated along each photon trajectory, allowing a direction-dependent treatment of photon decoupling in structured jets. The radiative transfer includes Klein-Nishina Compton scattering and polarization evolution using the Mueller matrix formalism.We perform a systematic parameter study exploring the effects of viewing angle, electron-positron pair loading (Zpm), and the optical-depth window of subphotospheric dissipation. The model produces time-resolved spectra, peak-energy evolution Epk(t), Band parameters, polarization degree Pi(E,t), and last-scattering statistics.We find that jet angular structure and the geometry of the line-of-sight optical depth strongly regulate spectral evolution and polarization signatures. The dissipation depth and pair loading jointly control the stability of Epk, the formation of high-energy spectral tails, and the energy dependence of polarization. These results provide quantitative predictions for GRB prompt-emission spectra and polarization that can be tested with current and upcoming high-energy polarimeters.

astro-ph.HE↗