arXiv subjects

Xi Shi

Publications and source records attributed to Xi Shi.

13 recordsLinked to original sources

arXiv2026

KVShareArena: KV-Cache Reuse Across Contexts and Model Checkpoints

LLM serving systems already reuse KV caches, but only when the reused text sits at the very start of the prompt. Two growing workloads break this condition: a retrieval-augmented generation server assembles a different set of retrieved chunks for every query, and a multi-agent coordinator reads reports written by other agents. Reused inside a new prompt, a cache carries the wrong positions and never attended to the other sources. The cache may also have been written by a different checkpoint of the same model family, which changes the stored values. Repair methods for such caches have appeared in three separate communities, each measured on its own terms, and existing benchmarks test only exact-prefix reuse, where nothing is lost. KVShareArena benchmarks KV-cache reuse across prompt contexts and model checkpoints on retrieved chunks and agent reports. It scores every method by the fraction of the gap it recovers between no cache and full recomputation, and charges compute, memory, and per-request latency with the cache in hand, reporting the one-time cost of building a cache separately. We find that correcting positions, which needs no recomputation, is enough until a question needs several sources at once. There, only methods that pay, by re-encoding part of the cache or by training, recover half to two thirds of the gap; unrepaired caches can be worse than no cache. Cache-compression methods that are harmless on a single prompt fall significantly behind position correction on freshly written agent reports. These patterns hold across three model boards. When a different checkpoint wrote the cache, training-free methods are barely affected, while an adapter trained on one checkpoint's caches loses quality. Harness, frozen querysets, and cost accounting ship as a pip package with an automated submission workflow and a public leaderboard.

cs.CL↗

arXiv2026

Learning Latency-Aware Orchestration for Multi-Agent Systems

Multi-agent systems (MAS) coordinate multiple LLM-powered agents through structured workflows, gaining reasoning power but incurring high inference latency from multi-step execution and repeated model invocations. Existing orchestration methods primarily optimize task performance and inference cost, leaving latency largely unaddressed. In MAS, end-to-end latency is governed by the critical execution path, so reducing total cost alone does not reliably reduce latency. Moreover, optimizing latency while preserving accuracy remains non-trivial: naive latency optimization can misassign operator-level credit and degrade task accuracy. To address this gap, we propose Latency-Aware Multi-agent System (LAMaS), a latency-aware orchestration framework for learning-based multi-agent systems. LAMaS addresses this challenge at two levels: at training time, it learns latency-aware execution graphs through constrained optimization with critical-path-aware credit assignment; at inference time, since a graph committed at training time cannot exploit runtime evidence, it complements graph construction with a lightweight controller that adaptively eliminates redundant future agent interactions as execution unfolds. Experiments on four benchmarks show that LAMaS achieves the best latency among evaluated learning-based MAS baselines, reducing end-to-end latency by over 50\% while maintaining competitive or better accuracy. LAMaS is also modular and transfers to other MAS with minimal changes, consistently yielding latency reductions.

cs.MA↗

arXiv2026

Anomalous Non-Hermitian Topological Anderson Insulator

Strong disorder drives conventional Hermitian systems into Anderson insulating states, suppressing all topological phases. Here, we unveil symmetry-protected, anomalous topological phases in the strong disorder limit of a non-Hermitian system, characterized by a scale-invariant merging of zero-energy modes. Using the maximally symmetric Jx lattice as an ideal platform and introducing specifically engineered (ABBA-type) symmetry-preserving non-Hermitian disorder, we observe a sequence of disorder-induced phase transitions: from a trivial insulator into and through a non-Hermitian topological Anderson insulator (TAI) phase, culminating in a stable anomalous non-Hermitian TAI phase characterized by a quantized polarization P_x \approx 0.25. Within this anomalous phase protected by the mobility gap, the zero-energy modes exhibit a distinct (N/2)-mode coalescence that scales with system size. Our findings demonstrate that non-Hermitian disorder engineered to preserve symmetry can induce and protect novel topological order inaccessible to conventional Hermitian disorder, thereby advancing the fundamental understanding of topological phenomena mediated by the interplay of disorder and non-Hermiticity.

cond-mat.mes-hall↗

arXiv2026

Learning Latency-Aware Orchestration for Multi-Agent Systems

Multi-agent systems (MAS) coordinate multiple LLM-powered agents through structured workflows, gaining reasoning power but incurring high inference latency from multi-step execution and repeated model invocations. Existing orchestration methods primarily optimize task performance and inference cost, leaving latency largely unaddressed. In MAS, end-to-end latency is governed by the \textit{critical execution path}, so reducing total cost alone does not reliably reduce latency. Moreover, optimizing latency while preserving accuracy remains non-trivial: naive latency optimization can misassign operator-level credit and degrade task accuracy. To address this gap, we propose \textbf{L}atency-\textbf{A}ware \textbf{M}ulti-\textbf{a}gent \textbf{S}ystem (\textbf{LAMaS}), a latency-aware orchestration framework for learning-based multi-agent systems. LAMaS addresses this challenge at two levels: at \emph{training time}, it learns latency-aware execution graphs through constrained optimization with critical-path-aware credit assignment; at \emph{inference time}, since a graph committed at training time cannot exploit runtime evidence, it complements graph construction with a lightweight controller that adaptively eliminates redundant future agent interactions as execution unfolds. Experiments on four benchmarks show that LAMaS achieves the best latency among evaluated learning-based MAS baselines, reducing end-to-end latency by over 50% while maintaining competitive or better accuracy. LAMaS is also modular and transfers to other MAS with minimal changes, consistently yielding latency reductions.

cs.MA↗

arXiv2025

Hidden in Plain Sight: Evaluation of the Deception Detection Capabilities of LLMs in Multimodal Settings

Detecting deception in an increasingly digital world is both a critical and challenging task. In this study, we present a comprehensive evaluation of the automated deception detection capabilities of Large Language Models (LLMs) and Large Multimodal Models (LMMs) across diverse domains. We assess the performance of both open-source and commercial LLMs on three distinct datasets: real life trial interviews (RLTD), instructed deception in interpersonal scenarios (MU3D), and deceptive reviews (OpSpam). We systematically analyze the effectiveness of different experimental setups for deception detection, including zero-shot and few-shot approaches with random or similarity-based in-context example selection. Our results show that fine-tuned LLMs achieve state-of-the-art performance on textual deception detection tasks, while LMMs struggle to fully leverage cross-modal cues. Additionally, we analyze the impact of auxiliary features, such as non-verbal gestures and video summaries, and examine the effectiveness of different prompting strategies, including direct label generation and chain-of-thought reasoning. Our findings provide key insights into how LLMs process and interpret deceptive cues across modalities, highlighting their potential and limitations in real-world deception detection applications.

cs.CL↗

arXiv2025

LegalCore: A Dataset for Event Coreference Resolution in Legal Documents

Recognizing events and their coreferential mentions in a document is essential for understanding semantic meanings of text. The existing research on event coreference resolution is mostly limited to news articles. In this paper, we present the first dataset for the legal domain, LegalCore, which has been annotated with comprehensive event and event coreference information. The legal contract documents we annotated in this dataset are several times longer than news articles, with an average length of around 25k tokens per document. The annotations show that legal documents have dense event mentions and feature both short-distance and super long-distance coreference links between event mentions. We further benchmark mainstream Large Language Models (LLMs) on this dataset for both event detection and event coreference resolution tasks, and find that this dataset poses significant challenges for state-of-the-art open-source and proprietary LLMs, which perform significantly worse than a supervised baseline. We will publish the dataset as well as the code.

cs.CL↗

arXiv2024

DHGS: Decoupled Hybrid Gaussian Splatting for Driving Scene

Existing Gaussian splatting methods often fall short in achieving satisfactory novel view synthesis in driving scenes, primarily due to the absence of crafty designs and geometric constraints for the involved elements. This paper introduces a novel neural rendering method termed Decoupled Hybrid Gaussian Splatting (DHGS), targeting at promoting the rendering quality of novel view synthesis for static driving scenes. The novelty of this work lies in the decoupled and hybrid pixel-level blender for road and non-road layers, without the conventional unified differentiable rendering logic for the entire scene. Still, consistency and continuity in superimposition are preserved through the proposed depth-ordered hybrid rendering strategy. Additionally, an implicit road representation comprised of a Signed Distance Function (SDF) is trained to supervise the road surface with subtle geometric attributes. Accompanied by the use of auxiliary transmittance loss and consistency loss, novel images with imperceptible boundary and elevated fidelity are ultimately obtained. Substantial experiments on the Waymo dataset prove that DHGS outperforms the state-of-the-art methods. The project page where more video evidences are given is: https://ironbrotherstyle.github.io/dhgs_web.

cs.CV↗

arXiv2023

Fluctuation-induced dispersion forces on thin DNA films

In this work, the calculation of Casimir forces across thin DNA films is carried out based on the Lifshitz theory. The variations of Casimir forces due to the DNA thicknesses, volume fractions of containing water, covering media and substrates are investigated. For a DNA film suspended in the air or water, the Casimir force is attractive, and its magnitude increases with decreasing the thickness of DNA films and the water volume fraction. For DNA films deposited on a dielectric(silica) substrate, the Casimir force is attractive for the air environment. However, the Casimir force shows unusual features in a water environment. Under specific conditions, switching signs of the Casimir force from attractive to repulsive can be achieved by increasing the DNA-film thickness. Finally, the Casimir force for DNA films deposited on a metallic substrate are investigated. The Casimir force is dominant by the repulsive interactions at a small DNA-film thickness for both the air and water environment. In a water environment, the Casimir force turns out to be attractive at a large DNA-film thickness, and a stable Casimir equilibrium can be found. In addition to the adhesion stability, our finding could be applicable to the problems of condensation and de-condensation of DNA, due to the fluctuation-induced dispersion forces.

physics.bio-ph↗

arXiv2020

Tunable Casimir equilibria with phase change materials: from quantum trapping to its release

A stable suspension of nanoscale particles due to the Casimir force is of great interest for many applications such as sensing, non-contract nano-machines. However, the suspension properties are difficult to change once the devices are fabricated. Vanadium dioxide (VO$_2$) is a phase change material, which undergoes a transition from a low-temperature insulating phase to a high-temperature metallic phase around a temperature of 340 K. In this work, we study Casimir forces between a nanoplate (gold or Teflon) and a layered structure containing a VO$_2$ film. It is found that stable Casimir suspensions of nanoplates can be realized in a liquid environment, and the equilibrium distances are determined, not only by the layer thicknesses but also by the matter phases of VO$_2$. Under proper designs, a switch from quantum trapping of the gold nanoplate ("on" state) to its release ("off" state) as a result of the metal-to-insulator transition of VO$_2$, is revealed. On the other hand, the quantum trapping and release of a Teflon nanoplate is found under the insulator-to-metal transition of VO$_2 $. Our findings offer the possibility of designing switchable devices for applications in micro-and nano-electromechanical systems.

physics.app-ph↗

arXiv2018

Magnetically tunable multi-band near-field radiative heat transfer between two graphene sheets

Near-field radiative heat transfer (NFRHT) is strongly related with many applications such as near-field imaging, thermos-photovoltaics and thermal circuit devices. The active control of NFRHT is of great interest since it provides a degree of tunability by external means. In this work, a magnetically tunable multi-band NFRHT is revealed in a system of two suspended graphene sheets at room temperature. It is found that the single-band spectra for B=0 split into multi-band spectra under an external magnetic field. Dual-band spectra can be realized for a modest magnetic field (e.g., B=4 T). One band is determined by intra-band transitions in the classical regime, which undergoes a blue shift as the chemical potential increases. Meanwhile, the other band is contributed by inter-Landau-level transitions in the quantum regime, which is robust against the change of chemical potentials. For a strong magnetic field (e.g., B=15 T), there is an additional band with the resonant peak appearing at near-zero frequency (microwave regime), stemming from the magneto-plasmon zero modes. The great enhancement of NFRHT at such low frequency has not been found in any previous systems yet. This work may pave a way for multi-band thermal information transfer based on atomically thin graphene sheets.

cond-mat.mes-hall↗

arXiv2018

Prediction of interface states in liquid surface waves with one-dimensional modulation

We theoretically studied the interface states of liquid surface waves propagating through the heterojunctions formed by a bottom with one-dimensional periodic undulations. Via considering the periodic structure as a homogeneous one, our systematic study shows that the signs of the effective depth and gravitational acceleration are opposite within the band gaps no matter the structure is symmetric or asymmetric. Those effective parameters can be used to predict the interface states which could amplify the amplitudes of liquid surface waves. These phenomena provide new opportunities to control the localization of water-wave energy.

physics.class-ph↗

arXiv2015

Dynamically tunable Fano resonance in periodically asymmetric graphene nanodisk pair

We present a dynamically frequency tunable Fano resonance planar device composed of periodically asymmetric graphene nanodisk pair for the mid-infrared region. There are two kinds of modes in this structure, that is, the symmetric mode and the antisymmetric mode. The resonance coupling between the symmetric and antisymmetric modes creates a classical Fano resonance. Both of the Fano resonance amplitude and frequency of the structure can be dynamically controlled by varying the Fermi energy of graphene. Resonance transition in the structure is studied to reveal the physical mechanism behind the dynamically tunable Fano resonance. The features of the Fano resonant graphene nanostructures should have promising applications in tunable THz filters, switches, and modulators.

physics.optics↗

arXiv2012

Transfer matrix method for optics in graphene layers

A transfer matrix method is developed for optical calculations of non-interacting graphene layers. Within the framework of this method, optical properties such as reflection, transmission and absorption for single-, double- and multi-layer graphene are studied. We also apply the method to structures consisting of periodically arranged graphene layers, revealing well-defined photonic band structures and even photonic bandgaps. Finally, we discuss graphene plasmons and introduce a simple way to tune the plasmon dispersion.

cond-mat.mes-hall↗