Search arXiv⌕ Search

arXiv subjects

Yi Xu

Publications and source records attributed to Yi Xu.

At least 73 records · Page 4Linked to original sources
arXiv2026

Idleness is Relative: Exploiting Tool-Call Idle Windows for Offloading in Agentic Systems with MORI

Modern LLM serving systems increasingly host agentic workloads, whose sessions issue tens of model invocations interleaved with tool calls, accumulating KV cache that can be reused across steps. As requests' total KV cache size easily exceeds GPU HBM capacity, researchers offload them to CPU DRAM. However, tool-call durations span orders of magnitude, and the cost of transferring KV cache between tiers makes it impractical to re-place entries on every call. We observe that agentic programs exhibit a two-phase structure: busy phases of rapid short tool calls and idle phases dominated by long-running calls. Current eviction policies such as LRU fail to capture this property. A binary busy/idle label also falls short because the ratio of busy to idle programs may not match the hardware's GPU-to-CPU capacity ratio. When it does not, one tier sits underutilized while the other is oversubscribed, wasting memory or forcing unnecessary evictions. We present MORI, an agent serving system that solves the above problem. Our key insight is that idleness is a continuous, relative spectrum. MORI ranks all active programs by idleness, assigns the busiest to GPU HBM and the most idle to CPU DRAM, dynamically shifts the partition boundary to match hardware capacity, and enforces admission control at each memory tier. Evaluated on real coding agent workloads collected from Claude Code across four GPU and model pairs, MORI delivers 20--71% higher throughput and 18--43% lower TTFT than the best baseline with offloading.

cs.OS↗
arXiv2026

GRLoc: Geometric Representation Regression for Visual Localization

Absolute Pose Regression (APR) has emerged as a compelling paradigm for visual localization. However, APR models typically operate as black boxes, directly regressing a 6-DoF pose from a query image, which can lead to memorizing training views rather than understanding 3D scene geometry. In this work, we propose a geometrically-grounded alternative. Inspired by novel view synthesis, which renders images from intermediate geometric representations, we reformulate APR as its inverse that regresses the underlying 3D representations directly from the image, and we name this paradigm Geometric Representation Regression (GRR). Our model explicitly predicts two disentangled geometric representations in the world coordinate system: (1) a raymap's directions to estimate camera rotation, and (2) a corresponding pointmap to estimate camera translation. The final camera pose is then recovered from these geometric components using a differentiable deterministic solver. This disentangled approach, which separates the learned visual-to-geometry mapping from the final pose calculation, introduces a strong geometric prior into the network. We find that the explicit decoupling of rotation and translation predictions measurably boosts performance. We demonstrate state-of-the-art performance on 7-Scenes and Cambridge Landmarks datasets, validating that modeling the inverse rendering process is a more robust path toward generalizable absolute pose estimation.

cs.CV↗
arXiv2026

ECHO: Efficient Chest X-ray Report Generation with One-step Block Diffusion

Chest X-ray report generation (CXR-RG) has the potential to substantially alleviate radiologists' workload. However, conventional autoregressive vision--language models (VLMs) suffer from high inference latency due to sequential token decoding. Diffusion-based models offer a promising alternative through parallel generation, but they still require multiple denoising iterations. Compressing multi-step denoising to a single step could further reduce latency, but often degrades textual coherence due to the mean-field bias introduced by token-factorized denoisers. To address this challenge, we propose \textbf{ECHO}, an efficient diffusion-based VLM (dVLM) for chest X-ray report generation. ECHO enables stable one-step-per-block inference via a novel Direct Conditional Distillation (DCD) framework, which mitigates the mean-field limitation by constructing unfactorized supervision from on-policy diffusion trajectories to encode joint token dependencies. In addition, we introduce a Response-Asymmetric Diffusion (RAD) training strategy that further improves training efficiency while maintaining model effectiveness. Extensive experiments demonstrate that ECHO surpasses state-of-the-art autoregressive methods, improving RaTE and SemScore by \textbf{64.33\%} and \textbf{60.58\%} respectively, while achieving up to \textbf{$8\times$} inference speedup with negligible degradation in clinical accuracy.

cs.LG↗
arXiv2026

DREAMS. JWST Spectroscopy of a $z=8.3$ Galaxy with an ALMA Dust Continuum Detection: Early Dust, Very High $T_{\rm dust}$, and a Multi-wavelength [OIII] Ratio Discrepancy

We present a deep DREAMS JWST/NIRSpec MSA medium-grating spectrum of MACS0416-Y1, a galaxy at $z=8.312$ with the highest-redshift ALMA dust continuum detection to date, in order to characterize its properties together with archival IFU and ALMA data. The deep NIRSpec spectrum reveals a broad H$β$ line with a width of $\sim1100$ km s$^{-1}$. We interpret it as a broad-line AGN whose line diagnostics are consistent with AGN activity across its clumpy structure, given the absence of little red dot signatures. MACS0416-Y1 clearly shows [OIII]4363 emission, suggesting a moderately low metallicity of $12+\log(\mathrm{O/H})=7.86^{+0.09}_{-0.08}$ ($0.15~Z_\odot$). The combination of [CII]158$μ$m and dust continuum emission indicates low dust mass ratios of $\log (M_{\rm dust}/M_{\rm gas})=-3.60^{+0.29}_{-0.22}$ and $\log (M_{\rm dust}/M_{\rm metal})=-0.95^{+0.29}_{-0.20}$. Because the metallicity of MACS0416-Y1 is around the critical metallicity of $0.1\textrm{-}0.2~Z_\odot$, the system is expected to undergo dust growth, explaining these low dust mass ratios as well as its small dust mass, $M_{\rm dust}\sim10^6~M_\odot$. The intense UV radiation from the AGN may contribute to a high dust temperature of $T_{\rm dust}\simeq 91^{+62}_{-35}$ K, boosting the dust-continuum emission above the ALMA detection limit despite the small $M_{\rm dust}$ at $z>8$. We find a very high total flux ratio of [OIII]88$μ$m/[OIII]5007 = $0.26 \pm 0.06$ in MACS0416-Y1, above predictions from single ionized nebular models at any electron density. This discrepancy suggests that the [OIII]88$μ$m and [OIII]5007 trace largely distinct regions, with the optical line suppressed in dusty nebulae, and thus requires careful interpretation when combining optical and infrared emission lines in JWST+ALMA studies.

astro-ph.GA↗
arXiv2026

HyperEyes: Dual-Grained Efficiency-Aware Reinforcement Learning for Parallel Multimodal Search Agents

Existing multimodal search agents process target entities sequentially, issuing one tool call per entity and accumulating redundant interaction rounds whenever a query decomposes into independent sub-retrievals. We argue that effective multimodal agents should search wider rather than longer: dispatching multiple grounded queries concurrently within a round. To this end, we present HyperEyes, a parallel multimodal search agent that fuses visual grounding and retrieval into a single atomic action, enabling concurrent search across multiple entities while treating inference efficiency as a first-class training objective. HyperEyes is trained in two stages. For cold-start supervision, we develop a Parallel-Amenable Data Synthesis Pipeline covering visual multi-entity and textual multi-constraint queries, curating efficiency-oriented trajectories via Progressive Rejection Sampling. Building on this, our central contribution, a Dual-Grained Efficiency-Aware Reinforcement Learning framework, operates at two levels. At the macro level, we propose TRACE (Tool-use Reference-Adaptive Cost Efficiency), a trajectory-level reward whose reference is monotonically tightened during training to suppress superfluous tool calls without restricting genuine multi-hop search. At the micro level, we adapt On-Policy Distillation to inject dense token-level corrective signals from an external teacher on failed rollouts, mitigating the credit-assignment deficiency of sparse outcome rewards. Since existing benchmarks evaluate accuracy as the sole metric, omitting inference cost, we introduce IMEB, a human-curated benchmark of 300 instances that jointly evaluates search capability and efficiency. Across six benchmarks, HyperEyes-30B surpasses the strongest comparable open-source agent by 9.9% in accuracy with 5.3x fewer tool-call rounds on average.

cs.LG↗
arXiv2026

Vulnerability Identification by Harnessing Inter-connected Multi-Source Information

The utilization of third-party open-source libraries is widespread in modern software development. Due to the dependency relationships, vulnerabilities within open-source libraries pose significant security threats to downstream software. However, the library vulnerabilities are usually implicitly reported and patched, without explicit notification to dependent software, leaving the downstream software vulnerable to potential attacks. Existing research efforts primarily focus on identifying vulnerability patches according to bug reports, commit messages, or code changes, overlooking the rich semantic connections among various sources of information. In this paper, our main insight is that various sources of information, including the vulnerability descriptions (e.g., bug reports) and its fixing strategies (e.g., commit messages and code changes), are highly interconnected. They express the high-level semantic information about the symptom, root cause and fixing strategies of the bugs. Hence, we propose an approach that involves training an AI model to integrate multiple sources, thus enhancing the effectiveness of vulnerability identification and vulnerability type classification. We introduce VPFinder, a tool that utilizes multi-head attention mechanisms to extract high-level semantic information from diverse sources. Evaluation results demonstrate that VPFinder achieves remarkable 0.941 F1-score in vulnerability identification task and 0.610 F1-score in vulnerability type classification task, outperforming state-of-the-art approaches by 5.4%.

cs.SE↗
arXiv2026

FluSplat: Sparse-View 3D Editing without Test-Time Optimization

Recent advances in text-guided image editing and 3D Gaussian Splatting (3DGS) have enabled high-quality 3D scene manipulation. However, existing pipelines rely on iterative edit-and-fit optimization at test time, alternating between 2D diffusion editing and 3D reconstruction. This process is computationally expensive, scene-specific, and prone to cross-view inconsistencies. We propose a feed-forward framework for cross-view consistent 3D scene editing from sparse views. Instead of enforcing consistency through iterative 3D refinement, we introduce a cross-view regularization scheme in the image domain during training. By jointly supervising multi-view edits with geometric alignment constraints, our model produces view-consistent results without per-scene optimization at inference. The edited views are then lifted into 3D via a feedforward 3DGS model, yielding a coherent 3DGS representation in a single forward pass. Experiments demonstrate competitive editing fidelity and substantially improved cross-view consistency compared to optimization-based methods, while reducing inference time by orders of magnitude.

cs.CV↗
arXiv2026

Joint Optimization of Handoff and Video Rate in LEO Satellite Networks

Low Earth Orbit (LEO) satellite communication is a promising approach to providing Internet connectivity to users in many remote areas. As videos are likely to account for most traffic in the LEO satellite network, as in the rest of the Internet, this work introduces a novel video-aware mobility management framework tailored for LEO satellite networks. Utilizing simulation models alongside real-world datasets, we show the importance of handoff strategy and throughput prediction algorithms in single-user and multi-user video streaming scenarios. Motivated by these observations, we propose a set of novel algorithms that can jointly choose the satellite and video bitrate to optimize the Quality of Experience (QoE). We first develop Model Predictive Control (MPC) and Reinforcement Learning (RL) based algorithms for a single user, and then extend them to accommodate multiple competing users that may share the same satellite. We introduce centralized training and distributed inference for our RL design, enabling a distributed policy informed by a global perspective. We demonstrate the effectiveness of our proposed models using trace-driven simulation and testbed experiments. We share our code and data with the research community.

cs.NI↗
arXiv2026

Generalized spectral Turán problems for disjoint cliques

The generalized Turán number $\text{ex}(n, H, F)$ denotes the maximum number of copies of $H$ in an $n$-vertex $F$-free graph. Let $kK_{r+1}$ be the disjoint union of $k$ copies of the complete graph $K_{r+1}$. Recently, Gerbner determined $\text{ex}(n, K_{t},kK_{r+1})$ for all sufficiently large $n$. In this paper, we study a spectral analogue of this problem via the $t$-clique tensor of a graph. We prove that if an $n$-vertex $kK_{r+1}$-free graph $G$ maximizes the $t$-clique spectral radius, then for sufficiently large $n$, $G$ is the join of a complete graph $K_{k-1}$ and the $r$-partite Turán graph $T_{r}(n-k+1)$. This establishes a spectral counterpart of Gerbner's Theorem. Moreover, in the case $t=2$, our result recovers a theorem of Ni, Wang, and Kang on the maximum spectral radius of $kK_{r+1}$-free graphs.

math.CO↗
arXiv2026

Sema Code: Decoupling AI Coding Agents into Programmable, Embeddable Infrastructure

AI coding agents have become central to developer workflows, yet every existing solution locks its reasoning capabilities within a specific delivery form, such as a CLI, IDE plugin, or web application. This limitation creates systemic barriers when enterprises attempt to reuse these capabilities across heterogeneous engineering environments. To address this challenge, we present Sema Code, an open AI coding framework built on the principle of being embeddable, pluggable, and framework-first. Sema Code completely decouples the core agent engine from all client layers, publishing it as a standalone npm library that any runtime can drive programmatically. Built around this architecture, we designed eight key mechanisms: multi-tenant engine isolation, FIFO input queuing with safe session reconstruction, adaptive context compression, multi-agent collaborative scheduling, intelligent Todo-based process management, four-layer asynchronous permission control, three-tier ecosystem integration spanning MCP, Skills, and Plugins, and a background task framework with separated execution and observation privileges. These mechanisms collectively address the engineering challenges of transforming a complex agent engine into a shared, programmable core. Demonstrating its architectural versatility, the same Sema Core engine simultaneously powers a VSCode extension and a multi-channel messaging gateway, which we name SemaClaw, to unify agent interactions across platforms such as Telegram and Feishu. These represent two fundamentally different product forms sharing an identical reasoning kernel, differing only at the client layer.

cs.SE↗
arXiv2026

SemaClaw: A Step Towards General-Purpose Personal AI Agents through Harness Engineering

The rise of OpenClaw in early 2026 marks the moment when millions of users began deploying personal AI agents into their daily lives, delegating tasks ranging from travel planning to multi-step research. This scale of adoption signals that two parallel arcs of development have reached an inflection point. First is a paradigm shift in AI engineering, evolving from prompt and context engineering to harness engineering-designing the complete infrastructure necessary to transform unconstrained agents into controllable, auditable, and production-reliable systems. As model capabilities converge, this harness layer is becoming the primary site of architectural differentiation. Second is the evolution of human-agent interaction from discrete tasks toward a persistent, contextually aware collaborative relationship, which demands open, trustworthy and extensible harness infrastructure. We present SemaClaw, an open-source multi-agent application framework that addresses these shifts by taking a step towards general-purpose personal AI agents through harness engineering. Our primary contributions include a DAG-based two-phase hybrid agent team orchestration method, a PermissionBridge behavioral safety system, a three-tier context management architecture, and an agentic wiki skill for automated personal knowledge base construction.

cs.AI↗
arXiv2026

Trust Your Memory: Verifiable Control of Smart Homes through Reinforcement Learning with Multi-dimensional Rewards

Large Language Models (LLMs) have become a key foundation for enabling personalized smart home experiences. While existing studies have explored how smart home assistants understand user queries to control devices in real time, their ability to perform memory-driven device control remains challenging from both evaluation and methodological perspectives. In terms of evaluation, existing benchmarks either focus on immediate device control or general open-domain memory retrieval tasks, and therefore cannot effectively evaluate a model's ability to perform memory-driven device control. Methodologically, while memory-driven device control can be approached using Reinforcement Learning, conventional RL methods generally rely on outcome-based supervision (i.e., whether the final task is achieved). This lack of intermediate feedback can lead to sub-optimal performance or local failures in fine-grained memory management tasks (adding, updating, deleting, and utilizing). To address these issues, we first release MemHomeLife, built from anonymized real-world long-term user interaction logs. To enable more fine-grained evaluation of different memory-related subtasks, we further construct MemHome, the first benchmark designed to systematically evaluate memory-driven device control in smart home scenarios.

cs.AI↗
arXiv2026

Simultaneous ground-state cooling of six mechanical modes of two levitated nanoparticles

Ground-state cooling is a prerequisite for exploring macroscopic quantum effects in mechanical motion of massive objects. Here we construct a polarization-angle-controllable coupled cavity-levitated-nanoparticle system in which two nanoparticles trapped by individual tweezers are coupled to a single-mode field in a cavity. We also study the simultaneous ground-state cooling of six mechanical displacement modes of the two levitated nanoparticles through the coherent scattering mechanism. By deriving the Hamiltonian of the system and performing the linearization, we obtain a linearized seven-mode Hamiltonian, which can exhibit the coupling structure and cooling mechanism. We confirm the physical condition for the appearance of dark modes, which will suppress the simultaneous ground-state cooling of these mechanical modes. We also find that, by properly tuning the polarization angle $θ$ between the cavity field and the optical tweezer fields, the coupling channels can be controlled on demand and simultaneous ground-state cooling of these six motional modes of the two nanoparticles can be realized. Our work paves the way for generation and manipulation of collective macroscopic quantum effects in multiple levitated nanoparticles.

quant-ph↗
arXiv2026

IQ-LUT: interpolated and quantized LUT for efficient image super-resolution

Lookup table (LUT) methods demonstrate considerable potential in accelerating image super-resolution inference. However, pursuing higher image quality through larger receptive fields and bit-depth triggers exponential growth in the LUT's index space, creating a storage bottleneck that limits deployment on resource-constrained devices. We introduce IQ-LUT, which achieves a reduction in LUT size while simultaneously enhancing super-resolution quality. First, we integrate interpolation and quantization into the single-input, multiple-output ECNN, which dramatically reduces the index space and thereby the overall LUT size. Second, the integration of residual learning mitigates the dependence on LUT bit-depth, which facilitates training stability and prioritizes the reconstruction of fine-grained details for superior visual quality. Finally, guided by knowledge distillation, our non-uniform quantization process optimizes the quantization levels, thereby reducing storage while also compensating for quantization loss. Extensive benchmarking demonstrates our approach substantially reduces storage costs (by up to 50x compared to ECNN) while achieving superior super-resolution quality.

cs.CV↗
arXiv2026

The Depth Ceiling: On the Limits of Large Language Models in Discovering Latent Planning

The viability of chain-of-thought (CoT) monitoring hinges on models being unable to reason effectively in their latent representations. Yet little is known about the limits of such latent reasoning in LLMs. We test these limits by studying whether models can discover multi-step planning strategies without supervision on intermediate steps and execute them latently, within a single forward pass. Using graph path-finding tasks that precisely control the number of required latent planning steps, we uncover a striking limitation unresolved by massive scaling: tiny transformers trained from scratch discover strategies requiring up to three latent steps, fine-tuned GPT-4o and Qwen3-32B reach five, and GPT-5.4 attains seven under few-shot prompting. Although the maximum latent planning depth models can learn during training is five, the discovered strategy generalizes up to eight latent steps at test-time. This reveals a dissociation between the ability to discover a latent strategy under final-answer supervision alone and the ability to execute it once discovered. If similar limits hold more broadly, strategies requiring multiple coordinated latent planning steps may need to be explicitly taught or externalized, lending credence to CoT monitoring.

cs.LG↗
arXiv2026

Biquadratic SOS Rank and Augmented Zarankiewicz Number

This paper introduces the concepts of the augmented Zarankiewicz number $z_A(m,n)$ and the limited augmented Zarankiewicz number $z_L(m,n)$, which are natural combinatorial extensions of the classical Zarankiewicz number. These numbers arise from augmented bipartite graphs that may contain both standard edges (1-edges) and pairs of edges representing squares of binomials (2-edges). The main theoretical result establishes the inequality chain $\operatorname{BSR}(m, n) \geq z_A(m, n) \geq z_L(m, n) \geq z(m, n)$, linking the maximum biquadratic sum-of-squares (SOS) rank to these extremal graph parameters. We determine the exact values of $z_L(m, n)$ for the cases $(m,2)$, $(3,3)$, $(4, 3)$ and $(4,4)$, and provide new lower bounds for the cases $(5,3)$, $(5,4)$, and $(5,5)$. These results yield improved lower bounds for the maximum SOS rank of biquadratic forms, demonstrating that $z_L(m,n)$ can exceed the classical Zarankiewicz number, thereby offering a refined combinatorial perspective on the SOS rank problem.

math.OC↗
arXiv2026

Towards High-Consistency Embodied World Model with Multi-View Trajectory Videos

Embodied world models aim to predict and interact with the physical world through visual observations and actions. However, existing models struggle to accurately translate low-level actions (e.g., joint positions) into precise robotic movements in predicted frames, leading to inconsistencies with real-world physical interactions. To address these limitations, we propose MTV-World, an embodied world model that introduces Multi-view Trajectory-Video control for precise visuomotor prediction. Specifically, instead of directly using low-level actions for control, we employ trajectory videos obtained through camera intrinsic and extrinsic parameters and Cartesian-space transformation as control signals. However, projecting 3D raw actions onto 2D images inevitably causes a loss of spatial information, making a single view insufficient for accurate interaction modeling. To overcome this, we introduce a multi-view framework that compensates for spatial information loss and ensures high-consistency with physical world. MTV-World forecasts future frames based on multi-view trajectory videos as input and conditioning on an initial frame per view. Furthermore, to systematically evaluate both robotic motion precision and object interaction accuracy, we develop an auto-evaluation pipeline leveraging multimodal large models and referring video object segmentation models. To measure spatial consistency, we formulate it as an object location matching problem and adopt the Jaccard Index as the evaluation metric. Extensive experiments demonstrate that MTV-World achieves precise control execution and accurate physical interaction modeling in complex dual-arm scenarios.

cs.RO↗
arXiv2026

FAST3DIS: Feed-forward Anchored Scene Transformer for 3D Instance Segmentation

While recent feed-forward 3D reconstruction models provide a strong geometric foundation for scene understanding, extending them to 3D instance segmentation typically relies on a disjointed "lift-and-cluster" paradigm. Grouping dense pixel-wise embeddings via non-differentiable clustering scales poorly with the number of views and disconnects representation learning from the final segmentation objective. In this paper, we present a Feed-forward Anchored Scene Transformer for 3D Instance Segmentation (FAST3DIS), an end-to-end approach that effectively bypasses post-hoc clustering. We introduce a 3D-anchored, query-based Transformer architecture built upon a foundational depth backbone, adapted efficiently to learn instance-specific semantics while retaining its zero-shot geometric priors. We formulate a learned 3D anchor generator coupled with an anchor-sampling cross-attention mechanism for view-consistent 3D instance segmentation. By projecting 3D object queries directly into multi-view feature maps, our method samples context efficiently. Furthermore, we introduce a dual-level regularization strategy, that couples multi-view contrastive learning with a dynamically scheduled spatial overlap penalty to explicitly prevent query collisions and ensure precise instance boundaries. Experiments on complex indoor 3D datasets demonstrate that our approach achieves competitive segmentation accuracy with significantly improved memory scalability and inference speed over state-of-the-art clustering-based methods.

cs.CV↗