Search arXiv⌕ Search

arXiv subjects

Kai Wu

Publications and source records attributed to Kai Wu.

At least 19 recordsLinked to original sources
arXiv2026

RSIHub: A Composable Framework for Agent-Evolution Methods

Agent-evolution methods use execution feedback to improve prompts, skills, tools, and orchestration, but their implementations often rebuild the same experimental machinery. We present RSIHub, an open-source framework for building, comparing, and extending these methods on shared infrastructure. Recipes compose reusable operators and declare mutable resources; RSIHub supplies execution, evaluation, recovery, and lineage. Researchers can focus on evolution policy while evaluation stays outside the mutable surface and results remain tied to candidates and evaluation conditions. We instantiate AHE, Hyperagents, A-Evolve, and GEPA without modifying the framework core, keeping method-specific logic in recipes and operators. We evaluate them with MiniSWE and Codex on Terminal Bench 2 and $τ^3$ Banking under fixed model and evaluator configurations. The largest observed overall pass@1 gain is 13.5 percentage points, while method rankings vary across settings and optimization gains do not always transfer to held-out tasks. RSIHub provides a common protocol for comparing evolution policies, inspectable candidate histories, and reusable components for developing new methods.

cs.AI↗
arXiv2026

Solving Large-Scale Parameter Inference in Controllable Nuclear Fusion via Coherent Ising Machines and Block Coordinate Descent

Bayesian inference enables self-consistent plasma parameter inversion in tokamak integrated data analysis by fusing multi-source diagnostics. However, under nonlinear Gaussian likelihoods, finding the maximum a posteriori (MAP) estimate typically relies on sampling methods whose high computational overhead precludes real-time plasma control. To address this bottleneck, we propose a Bayesian inversion framework using a Coherent Ising Machine (CIM)-a quantum computing architecture based on coupled degenerate optical parametric oscillators (DOPOs). With native all-to-all connectivity, the CIM directly maps the negative log-posterior potential onto a spin Hamiltonian and leverages quantum vacuum fluctuations for rapid state-space search during pitchfork bifurcation. Validated on plasma current profile inversion for the EAST tokamak, this method achieves a single-run solution time of approximately 1 ms. Although current qubit limits necessitate a block-wise iterative scheme that prevents full real-time execution today, rapidly scaling quantum hardware offers a clear pathway toward real-time tokamak control.

quant-ph↗
arXiv2026

CasePlay: Self-Play Reinforcement Learning from Case Reports for Medical Reasoning

Reinforcement learning (RL) is a promising way to improve the clinical reasoning abilities of large language models (LLMs), but current pipelines often rely on expert-labeled questions or costly synthetic-data generation. Meanwhile, the medical literature offers a wealth of high-quality unlabeled documents, including clinical case reports that trace the full clinical course from initial presentation to final treatment decisions. We study whether document-grounded self-play can convert these reports into effective RL training signals. We identify two challenges that limit existing self-play methods in this setting: knowledge-point fixation}, where the proposer repeatedly asks about a narrow set of salient clues in a source report, and difficulty-quality mismatch, where appropriately difficult questions may still be poorly formed or shallow, limiting their value for reasoner training. To address these challenges, we propose CasePlay, a self-play framework with two key components: (a) a Knowledge-Conditioned Proposer, which anchors question generation to case-specific knowledge points so that questions draw on a broader range of evidence from the source report; and (b) a Rubric-Judge Reasoner, which augments answer-accuracy feedback with rubric-based quality feedback on clue integration, reasoning depth, and option quality, guiding the proposer toward reasoning-intensive questions that better support reasoner training. Extensive experiments across seven medically related benchmarks demonstrate that CasePlay outperforms existing self-play baselines and improves the SFT warm-start model by 3.1 points on average, highlighted by a 6.3-point gain on LiveClin. This approach offers an effective path toward continually improving medical reasoning ability through self-play RL.

cs.LG↗
arXiv2026

PACE: Pareto-Adaptive Compression for Efficient Native MLLMs

Native multimodal large language models (native MLLMs) jointly encode visual and linguistic sequences within a unified autoregressive Transformer, emerging as a promising paradigm for multimodal understanding. Unlike conventional ViT-based MLLMs relying on external vision encoders, ViT-free native MLLMs tokenize pixel patches directly, potentially suffering severe visual token redundancy. A pilot study confirms this: unlike ViT-based models that degrade under compression, ViT-free MLLMs tolerate aggressive compression with near-lossless or even improved performance. Yet existing compression methods target ViT-based settings, leaving native MLLMs, which stand to gain the most, unexplored. To close this gap, we introduce an analysis pipeline centered on token discriminability and cross-dimensional heterogeneous redundancy, revealing that (i) redundancy is distributed independently and heterogeneously along height and width axes, and (ii) early-layer visual tokens are indiscriminable and uniformly redundant. Guided by these findings, we propose PACE (Pareto-Adaptive Compression for Efficient Native MLLMs), the first visual token compression framework tailored for native MLLMs with Multimodal-RoPE. Leveraging dimensional decoupling of positional encoding, PACE decomposes attention into two orthogonal spatial saliency estimates along height and width, gated by a positional attention signal, yielding direction-aware importance with negligible overhead. It further casts token budget allocation as Pareto optimization over performance preservation, inference efficiency, and spatial coverage, jointly identifying optimal compression depths and retention ratios. Experiments show PACE maintains near-lossless performance at only 25%-50% visual token retention, even improving several benchmarks, while delivering over 2x FLOPs reduction.

cs.CV↗
arXiv2026

Agentic Semantic Sensing for Resource-Adaptive AI-RAN

Semantic sensing (SemS) acquires task-relevant information rather than reconstructing complete physical information. Existing SemS formulations typically operate open loop: sensing configurations and observation schedules are fixed before inference and cannot respond to evolving task-level evidence. We propose Agentic SemS, a closed-loop framework for AI-enabled radio access networks (AI-RANs) that controls sensing within a communication-feasible profile set. A profile-conditioned causal Transformer updates the semantic belief from streaming observations, while key-value caching enables efficient state updates across profile changes without repeatedly processing the complete history. A semantic utility network estimates the task-level benefit of acquiring the next observation block under each feasible profile after accounting for sensing cost. The resulting continuation utilities jointly support next-profile selection and semantic early exit, adapting sensing configuration and duration to evolving evidence. The expected semantic gain is further related to conditional mutual information, providing a value-of-information interpretation of continued online sensing. Experiments on Widar3.0 with six emulated sensing profiles show that, in comparison with full-sequence High, the resource-efficient Agentic setting reduces normalized cumulative sensing cost by 25.33% while achieving 85.79% Macro-F1. At the same utility checkpoint, semantic early exit provides a further 12.35% cost reduction over adaptive sensing without early exit, with a 0.97-percentage-point Macro-F1 decrease.

cs.AI↗
arXiv2026

EdgeAgent: Orchestrating On-Device LLM inference for End-User Multi-Agent Systems on CPU-GPU Unified Memory Architectures

Emerging multi-agent LLMs demand privacy-preserving edge deployment, yet current inference systems struggle with these collaborative workflows. Specifically, the memory-bound decode phase causes severe bus contention on unified memory architectures (UMA), paralyzing naive CPU-GPU co-execution. Furthermore, speculative decoding in multi-agent workloads faces extreme variance in drafting difficulty, alternating between complex reasoning and predictable structured generation. Compounded by frequent tool-induced stalls, this highly fragmented execution severely underutilizes hardware and defeats traditional static batching. We present EdgeAgent, a cross-layer inference system explicitly co-designed for edge UMA and multi-agent workloads. At the micro-architectural level, it bypasses rigid graph-compiler constraints to enable zero-copy UMA-aware tensor parallelism, utilizing asymmetric memory layouts to fully saturate both CPU and GPU compute units. At the scheduling level, it dynamically allocates draft budgets based on real-time sequence predictability to bound bandwidth waste. Concurrently, an asynchronous suspend-and-yield mechanism actively evicts stalled agents, ensuring continuous hardware saturation during unpredictable tool invocations. Extensive evaluations on an Apple M4 SoC demonstrate that the UMA-aware execution alone contributes a 1.29x speedup over batched speculative decoding. Adding the agent-aware scheduling lifts the full EdgeAgent system to a 1.77x speedup under extreme tool-use latencies.

cs.DC↗
arXiv2026

Simulating tidal disruption events in nuclear star clusters

Context. Nuclear star clusters (NSCs) and supermassive black holes (SMBHs) often coexist at the centers of galaxies, engaging in complex dynamical interactions that influence SMBH growth. Stars that approach the SMBH too closely are tidally disrupted, and a fraction of their debris is accreted onto the black hole. Such events may contribute to SMBH growth in gas-deficient galaxies. Aims. Our work aims to improve the theoretical framework for understanding the mass accretion of stellar debris following tidal disruption events (TDEs) through numerical simulations. We perform direct N-body simulations of an NSC with a central SMBH embedded in a fully self-consistent dynamical model to test a new TDE accretion prescription in a set of pilot simulations. Methods. The previous TDE prescription used a fixed accretion radius and assumed that the entire stellar mass was accreted after disruption. In the new partial-mass accretion model, the mass added to the SMBH depends on the orbital eccentricity and penetration factor of the disrupted star. Results. We find that most TDEs involving main-sequence stars are marginally eccentric or marginally hyperbolic and contribute approximately half of the stellar mass to SMBH growth, consistent with recent hydrodynamical simulations. In all simulated models, the TDE-driven mass accretion rate exhibits an early peak followed by a rapid decline. Models with higher initial SMBH masses exhibit a faster decline in the accretion rate, consistent with more rapid loss-cone depletion and inefficient replenishment

astro-ph.GA↗
arXiv2026

Boundary and Intra-Segment Learning for Partial Audio Deepfake Localization

Partial audio deepfakes manipulate only selected speech regions, making them difficult to be localized. Existing methods exploit boundary cues for partial deepfake localization, but primarily focus on identifying boundary positions rather than modeling the feature changes that characterize authenticity transitions. Meanwhile, the internal characteristics of continuous bona fide and spoofed segments remain underexplored. In this paper, we propose Boundary and Intra-Segment Learning (BISL), which introduces boundary learning to model feature differences between adjacent frames and distinguish authenticity transitions from general acoustic variations. In addition, intra-segment learning captures the overall characteristics of continuous bona fide and spoofed segments while enhancing feature consistency within each segment. By jointly learning frame, boundary, and segment information, BISL enables more effective fine-grained partial audio deepfake localization. Experiments on multiple localization benchmarks show that BISL achieves an EER of 2.52\% and an F1-score of 97.40\% on PartialSpoof, outperforming the compared methods, while maintaining competitive performance on HAD and improved cross-dataset performance on LPS. The code will be made publicly available upon acceptance.

cs.SD↗
arXiv2026

Shape- and Cation-Engineered Ferrite Nanoparticles for Enhanced Theranostic Performance in Ovarian Cancer Tumor-Mimicking Phantom

Magnetic hyperthermia integrated with magnetic resonance imaging (MRI) requires nanoparticles that combine strong heating, T2 contrast, and hemocompatibility after systemic administration. Here, we prepared citrate-stabilized ferrite nanoparticles (Fe3O4, Co0.6Fe2.4O4, Zn0.3Fe2.7O4, and Zn0.35Mn0.25Fe2.4O4) in spherical (8-10.5 nm) and quasi-cubic (9-12.5 nm) forms, along with 35 nm Fe3O4 cubes. By coupling morphology engineering with spinel-lattice cation substitution, we tuned saturation magnetization, coercivity, T2-weighted signal attenuation, and heat dissipation in ovarian tumor-mimicking phantoms under clinically compatible alternating magnetic fields. The 35 nm Fe3O4 cube and quasi-cubic Zn0.35Mn0.25Fe2.4O4 generated the highest phantom temperature rises, whereas smaller spherical particles produced softer heating. All formulations were cytocompatible in SKOV3 cells, with >96% viability at 24 h. However, hemocompatibility distinguished the leading candidates: 35 nm Fe3O4 cubes induced red blood cell deformation despite minimal hemolysis, while 9 nm quasi-cubic Zn0.35Mn0.25Fe2.4O4 showed no change versus PBS up to 1,000 μg mL-1. Perfusable GelMA models under physiological flow further confirmed biocompatibility. MRI phantoms showed that Fe3O4 cubes produced extended signal voids that may obscure boundaries. Overall, 9.5 nm quasi-cubic Zn0.35Mn0.25Fe2.4O4 is the leading intravenous candidate, combining strong heating, T2 contrast, and hemocompatibility, whereas large Fe3O4 cubes are better suited for localized intratumoral hyperthermia.

physics.app-ph↗
arXiv2026

Magnetic Particle Spectroscopy for Detecting Cell-Associated Zinc Ferrite Nanoparticles and Probing Their Relaxation Dynamics

Magnetic particle spectroscopy (MPS) enables sensitive detection of magnetic nanoparticles (MNPs) and characterization of their dynamic magnetization through higher-order harmonics. Here, we synthesized citrate-functionalized, 30 nm cubic Zn0.4Fe2.6O4 (ZFO) MNPs and investigated their association with SKOV3 ovarian cancer cells using MPS. The ZFO MNPs exhibited a crystalline spinel structure, a mean hydrodynamic diameter of 39.4 nm, strong room-temperature magnetization, low coercivity, and citrate-associated surface functional groups. Live/Dead imaging indicated good cytocompatibility after 24 h exposure at concentrations up to 0.5 mg/mL, while bright-field microscopy showed concentration-dependent cell-associated nanoparticle accumulation. Two MPS drive fields were compared using ZFO MNPs dispersed in DI water and agar as relatively unrestricted and strongly confined reference states. The 7.75 kHz, 20 mT condition retained more higher-order harmonics than 11.37 kHz and 10 mT and produced a larger and order-dependent spectral separation between the two states; it was therefore selected for the cellular measurements. After ZFO exposure and removal of unbound nanoparticles, MPS detected cell-associated MNPs in samples containing 0.1, 1, and 2x10^6 SKOV3 cells. The 3rd-, 5th-, and 7th-harmonic amplitudes increased proportionally with cell number, with power-law exponents of 1.03-1.09. We also report that most normalized harmonic ratios from the 1- and 2x10^6-cell samples fell between the water and agar references, indicating partial restriction of Brownian rotation in the cellular environment. These findings demonstrate that MPS can detect cell-associated MNPs while providing complementary spectral information about their ensemble-averaged physical confinement, supporting its application in magnetic cell labeling and cell-tracking studies.

physics.app-ph↗
arXiv2026

Enabling High-Bandwidth Flash for Generative Recommendation Serving with Write-Aware KV Cache Policy

Generative recommendation (GR) systems increasingly leverage user-level KV cache reuse to avoid recomputing long user histories. However, the growing KV cache capacity and bandwidth requirements introduce new challenges for memory system. High-Bandwidth Flash (HBF) provides a promising solution by offering substantially higher capacity than HBM while approaching HBM-class read bandwidth, enabling larger scale KV cache retention and improved serving throughput. Yet conventional Least-Recently-Used (LRU) KV cache management tightly couples KV cache writes with cache misses, generating excessive write traffic that rapidly exhausts flash endurance. In this work, we evaluate a write-aware KV cache policy based on admission-controlled LRU-K for HBF-based GR serving. By filtering low-reuse users before cache admission, LRU-K decouples KV cache writes from misses and significantly reduces unnecessary writes. We develop an analytical model to characterize GR serving performance, KV cache write traffic, and HBF lifetime, and evaluate performance across diverse memory systems and GR workloads. Our results show that HBF-based systems achieve 3.8 to 4.7 times higher throughput than HBM-only systems. Moreover, LRU-K extends HBF lifetime from about one year under conventional LRU to over six years with a moderate K=10, while maintaining comparable or even slightly improved throughput. These results highlight the importance of write aware KV cache policy for sustainable HBF-based GR serving.

cs.AR↗
arXiv2026

The transverse matter Hamiltonian

Enrico Fermi in 1932 used classical Gauss equation to derive the Coulomb density--density interaction from the longitudinal electromagnetic potential. In this work we extend the Fermi procedure to the transverse component of the vector potential. By using a fully quantum canonical transformation, we replace the transverse vector potential with a current--current and current--current--density interactions. The transformed Hamiltonian is, then, projected in the fermionic space providing a matter--only, totally incoherent and gauge respecting Hamiltonian. The removal of coherence will avoid the breakdown of perturbation theory predicted by Haag's theorem, and make possible to introduce, in the transformed space, the diagrammatic approach. After discussing the implications of the transverse interactions on the current theories based on the Fermi procedure, we conclude by showing how the transverse Hamiltonian provides the quantum origin of the electromagnetic longitudinal--transverse splitting of phonons and other elemental excitations.

math-ph↗
arXiv2026

Multi-UE Networked Sensing: A New Paradigm for 6G Perceptive Mobile Networks

Networked sensing, which jointly exploits observations from multiple distributed nodes, is essential for unlocking the full sensing potential of integrated sensing and communications (ISAC). This article introduces multi-UE sensing, a new networked sensing paradigm for future perceptive mobile networks that exploits the correlated sensing observations naturally arising from distributed user equipment devices (UEs) interacting with common targets. Representative uplink, downlink, and hybrid sensing architectures are presented, together with a multi-view signal processing framework encompassing synchronization, correlation-aware parameter estimation, and sensing fusion. Key open challenges, including correlation modelling, target association, sensing information compression, and communication-sensing co-optimization, are also discussed.

eess.SP↗
arXiv2026

Rainfall Sensing via Mobile Communication Signals

Rainfall monitoring is important for hydrological observation, disaster warning, and environmental sensing, but conventional rain gauges and weather radars suffer from sparse deployment and high infrastructure costs. This paper proposes PMN-RainSense, a rainfall sensing framework using sub-6-GHz mobile communication signals that supports practical single-antenna deployment. Unlike attenuation-based approaches, which are unreliable at sub-6 GHz because rain-induced attenuation over short mobile access links is only on the order of hundredths of a decibel, the proposed framework exploits fine-grained dynamics. A spectral-temporal channel state information (CSI) compensation method suppresses packet-wise timing and phase distortions while preserving sensing-relevant information. Rainfall-sensitive features are extracted from the delay-Doppler domain to mitigate environmental interference, with angle-domain filtering as an optional extension for multi-antenna receivers. Under bandwidth and antenna constraints, rainfall-correlated Doppler fluctuations serve as the dominant sensing signature, while Doppler-domain normalization improves robustness across links and deployments. Controlled WiFi experiments demonstrate rainfall-associated Doppler broadening and achieve a three-class classification accuracy of 95.48% using a random forest classifier. Long-Term Evolution (LTE) CSI measurements collected from cellular base stations over 11 carrier frequencies from 0.763 to 2.68 GHz yield a mean absolute error (MAE) of 0.25-0.27 mm/h for rainfall intensity estimation using a one-dimensional convolutional network.

eess.SP↗
arXiv2026

Pretrained Optimization Model for Zero-Shot Black Box Optimization

Zero-shot optimization involves optimizing a target task that was not seen during training, aiming to provide the optimal solution without or with minimal adjustments to the optimizer. It is crucial to ensure reliable and robust performance in various applications. Current optimizers often struggle with zero-shot optimization and require intricate hyperparameter tuning to adapt to new tasks. To address this, we propose a Pretrained Optimization Model (POM) that leverages knowledge gained from optimizing diverse tasks, offering efficient solutions to zero-shot optimization through direct application or fine-tuning with few-shot samples. Evaluation on the BBOB benchmark and two robot control tasks demonstrates that POM outperforms state-of-the-art black-box optimization methods, especially for high-dimensional tasks. Fine-tuning POM with a small number of samples and budget yields significant performance improvements. Moreover, POM demonstrates robust generalization across diverse task distributions, dimensions, population sizes, and optimization horizons. For code implementation, see https://github.com/ninja-wm/POM/.

cs.NE↗
arXiv2026

MIRA: Medical Image Reflection for Agentic Diagnosis

Medical visual agents can use tools to inspect images and retrieve external knowledge, but indiscriminate tool use may introduce noisy or misleading evidence. Reliable diagnosis therefore requires not only acquiring additional observations, but also verifying whether tool actions are necessary and whether the resulting evidence supports the current hypothesis. We introduce MIRA (Medical Image Reflection for Agentic Diagnosis), a medical visual diagnostic framework for autonomous evidence search and reflective verification. MIRA dynamically invokes image-processing operations, including zooming, grounding, pointing, rotation, and measurement, as well as web search, while evaluating the relevance and consistency of the acquired evidence. We develop MIRA through a two-stage training strategy. First, a tool-augmented Monte Carlo Tree Search data engine explores diverse diagnostic hypotheses and jointly verifies visual grounding accuracy and semantic consistency to construct supervised fine-tuning trajectories. Second, reinforcement learning further improves decision-making through online reflective principle evolution: failure cases are distilled into candidate principles, and only principles that improve held-out rollout rewards are retained. Across nine medical visual reasoning benchmarks, MIRA achieves an average score of 64.73, improving its Qwen3-VL-8B backbone by 7.44 points. It also increases useful tool-use judgments from 56.2% to 73.8% and reduces harmful judgments from 8.9% to 1.6%. Qualitative analyses show that MIRA can re-examine evidence, correct premature conclusions, and adapt its tool-use strategy. Project page: https://MIRA-VL.github.io/

cs.CV↗
arXiv2026

MetaView: Monocular Novel View Synthesis with Scale-Aware Implicit Geometry Priors

Current visual generation models are capable of producing high-quality content, yet they lack a coherent perception of the spatial structure. Existing generative novel view synthesis methods typically introduce explicit geometry priors, which enforce spatial consistency but inherently restrict generalization in large view changes. In contrast, recent interactive generative methods favor implicit scene modeling, offering greater flexibility at the cost of precise camera control and geometry consistency. In this paper, we propose MetaView, a diffusion-based monocular novel view synthesis framework that enables rendering under large view changes from a single image. Our key insight is to combine implicit geometry modeling with minimal yet essential explicit 3D cues: we incorporate implicit geometry priors from a feed-forward geometry perception network to regularize structure without imposing restrictive reconstruction pipelines, while leveraging metric depth to anchor the generation to a metric scale. This design allows MetaView to achieve both geometry consistency and precise controllability. Extensive experiments demonstrate that, under challenging monocular large viewpoint changes, MetaView significantly outperforms existing methods and exhibits superior generalization. Our code is publicly available at https://github.com/KlingAIResearch/MetaView.

cs.CV↗
arXiv2026

Boosting Deepresearch and LongContext Ability with Self-Generated Deepresearch Rollouts Traces

Deepresearch (DR) agents interact with real-world web environments through multi-turn search and visit, causing their contexts to grow rapidly over time. We observe that, even after DR Agentic Reinforcement Learning (DR-RL), 61.6% of the model's remaining prediction errors can still be attributed to insufficient long-context understanding, including longcontext hallucination and failures in cross-document evidence integration. It motivates us to further break the bottleneck of DR-RL by strengthening the model's long-context ability. However, effective LongContext training requires more than simply increasing context length. To bridge the data gap, we propose `DR Rollouts to LongContext-QA (DR-to-Long)'. The method repurposes DR-RL trajectories, which naturally contain search histories, visited webpages, evidence snippets, and final-answer supervision. It then replaces the compact snippets and webpage summaries in each trajectory with the full contents of their corresponding URLs, producing substantially longer multi-document contexts while preserving the original evidence relationships. Building on DR-to-Long, we introduce DLD (DR -> LongQA -> DR)-RL. DLD-RL first performs a short DR-RL stage to collect rollout trajectories, which are then converted into LongQA instances at zero annotation cost. The model is subsequently optimized with LongQA-RL to strengthen LongContext ability, followed by full DR-RL to continue improving its DR capability. Experiments show that DLD-RL outperforms standard DR-RL by 7.3% on three Deepresearch benchmarks and improves performance by 13.5% on three long-context benchmarks.

cs.CL↗