arXiv subjects

Yi Qi

Publications and source records attributed to Yi Qi.

At least 19 recordsLinked to original sources

arXiv2026

Skill Use or Skill Theater? Evaluating the Reasoning Backroom in Skill-Augmented Language Agents

Reusable skills are becoming a standard interface for extending language agents with task procedures. Yet evaluators usually infer skill use from visible reasoning or the agent's own attribution. These signals show what the agent appears to use, not whether the skill changed its decision. We ask whether skill-augmented agents exhibit a \textbf{Reasoning Backroom}, a systematic gap between stated skill use and intervention-measured influence. We introduce BACKTRACE, an evaluation framework that pairs each skill-conditioned answer with a matched no-skill counterfactual, intervenes on skill meaning, wording, identity, content, and assignment, and elicits attribution only after the answer is committed. We instantiate the framework as BACKROOMBench, a verified testbed spanning controlled logic and competition mathematics, multiple skill conditions, single-agent and multi-agent settings, and diverse model families. Our evaluation reveals a pervasive provenance failure. Across models and domains, stated skill use often remains stable while causal reliance and signed utility vary, producing both silent uptake and performative use. Behavioral effects follow procedural content more reliably than displayed skill identity, whereas stated attributions respond strongly to artifact availability. Observational detectors based on direct skill-use claims, text mentions, trace similarity, and an LLM judge do not identify which decisions actually depend on the skill. In multi-agent systems, skill influence can survive communication even after its source is lost, while no-skill teams still name skills and sources that were never supplied. These findings establish the Reasoning Backroom as a general AI provenance problem whose audit requires intervention.

cs.AI↗

arXiv2026

Generalized Hyperbolic Conical Circle Packings associated with Finite Polygonal Decompositions of Surfaces with Boundary

Let S be a compact topological surface with finitely many genus and finitely many holes and let D be a polygonal decomposition of S. In this paper, we consider the generalized hyperbolic conical circle packings associated with D. We first show that the boundary value problem has a unique solution k by prescribing total geodesic curvatures of generalized hyperbolic conical circles centered at interior vertices and geodesic curvatures of generalized hyperbolic conical circles centered at boundary vertices. Then we show that such a solution k can be obtained by taking a limit of the packings inductively modified by Thurston's algorithm via an arbitrarily chosen initial generalized hyperbolic conical circle packing associated with D and with given boundary values. Thirdly, we develop the so-called discrete Schwarz-Pick lemma for the solution packing k on D.

math.GT↗

arXiv2026

Perron's method and spherical ideal circle patterns with prescribed total geodesic curvatures

In this paper, we apply the classical Perron method to give a proof of the existence and uniqueness/rigidity result of a circle pattern on a closed surface equipped with conical spherical metric when prescribed measures of the angles of intersecting circles stay in the range (0,{\pi}/2] and total geodesic curvatures are assigned to the circles, which is recently obtained in [3] via Colin de Verdi\`ere's variation method. Then we show the convergence of Thurston's algorithm, which adjusts the geodesic curvatures of circles one by one based on the prescribed values for total geodesic curvatures of the circles, to the desired circle pattern in the setting of the result.

math.GT↗

arXiv2026

ProGRank: Probe-Gradient Reranking to Defend Dense-Retriever RAG from Corpus Poisoning

Retrieval-Augmented Generation (RAG) improves large language model applications by grounding generation in retrieved evidence, but also introduces corpus poisoning as a new attack surface. In this setting, an adversary injects or edits passages so that they enter the Top-$K$ results for target queries and influence downstream generation. Existing defences often rely on content filtering, auxiliary models, or generator-side reasoning, which complicates deployment. We propose ProGRank, a post hoc, training-free retriever-side defence for dense-retriever RAG. ProGRank stress-tests each query--passage pair under mild randomized perturbations, extracts probe gradients from a small fixed parameter subset, and derives two instability signals: representational consistency and dispersion risk. It then combines these signals with a score gate for reranking. ProGRank preserves the original passage content, requires no retraining, and supports a surrogate-based variant when the deployed retriever is unavailable. Experiments across datasets, retrievers, attacks, and retrieval-stage and end-to-end settings show that ProGRank improves robustness and maintains a favorable robustness--utility trade-off, including under adaptive evasive attacks.

cs.AI↗

arXiv2025

The asymptoticity of extremal length in Teichm\"uller space

We study the asymptotic behavior of extremal length along Teichm\"uller rays. Specifically, we determine the limit of extremal length along a Teichm\"uller ray and obtain an explicit expression for this limit, which complements a related formula established by Cormac Walsh. Building on this result and Kerckhoff's formula, we establish a formula for the limiting Teichm\"uller distance between two points moving along arbitrary pairs of Teichm\"uller rays. Furthermore, we derive a necessary and sufficient condition for two Teichm\"uller rays to be asymptotic. Finally, by shifting the initial points of the Teichm\"uller rays along their associated Teichm\"uller geodesics, we show that the minimum of the limiting Teichm\"uller distance coincides with the detour metric between the endpoints of the rays on the horofunction boundary.

math.CV↗

arXiv2025

The asymptoticity of pairs of Teichm\"uller rays

In this paper, we study the limit of Teichm\"uller distance between two points along a pair of Teichm\"uller rays. We obtain an explicit formula for the limiting Teichm\"uller distance when the vertical measured foliations of the quadratic differentials are finite sums of weighted simple closed curves and uniquely ergodic measures. The limit is expressed in terms of ratios of the corresponding moduli and the Teichm\"uller distance between the limit surfaces when the vertical measured foliations are absolutely continuous. Consequently, two Teichm\"uller rays are asymptotic if and only if their vertical measured foliations are modularly equivalent and their limit surfaces coincide, which implies a main result of Masur on the asymptoticity of Teichm\"uller rays determined by uniquely ergodic quadratic differentials. Furthermore, we prove that the infimum of the limiting Teichm\"uller distances can be represented in terms of the distance between the limit surfaces of the Teichm\"uller rays and the detour metric of their endpoints on the Gardiner-Masur boundary, when the initial points of the rays vary along the Teichm\"uller geodesics.

math.CV↗

arXiv2025

Observation of a Knotted Electron Diffusion Region in Earth's Magnetotail Reconnection

Magnetic reconnection is a fundamental plasma process that alters the magnetic field topology and releases magnetic energy. Most numerical simulations and spacecraft observations assume a two-dimensional diffusion region, with the electron diffusion region (EDR) embedded in the same plane as the ion diffusion region (IDR) and a uniform guide field throughout. Using observations from Magnetospheric Multiscale (MMS) mission, we report a non-coplanar, knotted EDR in Earth's magnetotail current sheet. The reconnection plane of the knotted EDR deviates by approximately 38{\deg} from that of the IDR, with the guide field exhibiting both a 38{\deg} directional shift and a twofold increase in amplitude. Moreover, the Hall magnetic field is bipolar in the EDR but quadrupolar in the IDR, indicating different Hall current structures at electron and ion scales. These observations highlight the importance of three-dimensional effects and illustrate the complexity of multiscale coupling between the EDR and IDR during reconnection studies.1

physics.space-ph↗

arXiv2025

TAIJI: Textual Anchoring for Immunizing Jailbreak Images in Vision Language Models

Vision Language Models (VLMs) have demonstrated impressive inference capabilities, but remain vulnerable to jailbreak attacks that can induce harmful or unethical responses. Existing defence methods are predominantly white-box approaches that require access to model parameters and extensive modifications, making them costly and impractical for many real-world scenarios. Although some black-box defences have been proposed, they often impose input constraints or require multiple queries, limiting their effectiveness in safety-critical tasks such as autonomous driving. To address these challenges, we propose a novel black-box defence framework called \textbf{T}extual \textbf{A}nchoring for \textbf{I}mmunizing \textbf{J}ailbreak \textbf{I}mages (\textbf{TAIJI}). TAIJI leverages key phrase-based textual anchoring to enhance the model's ability to assess and mitigate the harmful content embedded within both visual and textual prompts. Unlike existing methods, TAIJI operates effectively with a single query during inference, while preserving the VLM's performance on benign tasks. Extensive experiments demonstrate that TAIJI significantly enhances the safety and reliability of VLMs, providing a practical and efficient solution for real-world deployment.

cs.CV↗

arXiv2025

Fiber-based Ultra-High Speed Diffuse Speckle Contrast Analysis System for Deep Blood Flow Sensing Using a Large SPAD Camera

Diffuse speckle contrast analysis (DSCA), also called speckle contrast optical spectroscopy(SCOS), has emerged as a groundbreaking optical imaging technique for tracking dynamic biological processes, including blood flow and tissue perfusion. Recent advancements in single-photon avalanche diode (SPAD) cameras have unlocked exceptional capabilities in sensitivity, time resolution, and high frame rate imaging. Despite this, the application of large-format SPAD arrays in speckle contrast analysis is still relatively uncommon. In this study, we introduce a pioneering use of a large format SPAD camera for DSCA. By harnessing the camera's high temporal resolution and photon detection efficiency, we significantly enhance the accuracy and robustness of speckle contrast measurements. Our experimental results demonstrate the system's remarkable ability to capture rapid temporal variations over a broad field of view, enabling detailed spatiotemporal analysis. Through simulations, phantom experiments, and in vivo studies, we validate the approach's potential for a wide range of biomedical applications, such as cuff occlusion tests and functional tissue monitoring. This work highlights the transformative impact of large SPAD cameras on DSCA, paving the way for new breakthroughs in optical imaging.

physics.ins-det↗

arXiv2024

Safeguarding Large Language Models: A Survey

In the burgeoning field of Large Language Models (LLMs), developing a robust safety mechanism, colloquially known as "safeguards" or "guardrails", has become imperative to ensure the ethical use of LLMs within prescribed boundaries. This article provides a systematic literature review on the current status of this critical mechanism. It discusses its major challenges and how it can be enhanced into a comprehensive mechanism dealing with ethical issues in various contexts. First, the paper elucidates the current landscape of safeguarding mechanisms that major LLM service providers and the open-source community employ. This is followed by the techniques to evaluate, analyze, and enhance some (un)desirable properties that a guardrail might want to enforce, such as hallucinations, fairness, privacy, and so on. Based on them, we review techniques to circumvent these controls (i.e., attacks), to defend the attacks, and to reinforce the guardrails. While the techniques mentioned above represent the current status and the active research trends, we also discuss several challenges that cannot be easily dealt with by the methods and present our vision on how to implement a comprehensive guardrail through the full consideration of multi-disciplinary approach, neural-symbolic method, and systems development lifecycle.

cs.CR↗

arXiv2024

Building Guardrails for Large Language Models

As Large Language Models (LLMs) become more integrated into our daily lives, it is crucial to identify and mitigate their risks, especially when the risks can have profound impacts on human users and societies. Guardrails, which filter the inputs or outputs of LLMs, have emerged as a core safeguarding technology. This position paper takes a deep look at current open-source solutions (Llama Guard, Nvidia NeMo, Guardrails AI), and discusses the challenges and the road towards building more complete solutions. Drawing on robust evidence from previous research, we advocate for a systematic approach to construct guardrails for LLMs, based on comprehensive consideration of diverse contexts across various LLMs applications. We propose employing socio-technical methods through collaboration with a multi-disciplinary team to pinpoint precise technical requirements, exploring advanced neural-symbolic implementations to embrace the complexity of the requirements, and developing verification and testing to ensure the utmost quality of the final product.

cs.CL↗

arXiv2024

Direct Training Needs Regularisation: Anytime Optimal Inference Spiking Neural Network

Spiking Neural Network (SNN) is acknowledged as the next generation of Artificial Neural Network (ANN) and hold great promise in effectively processing spatial-temporal information. However, the choice of timestep becomes crucial as it significantly impacts the accuracy of the neural network training. Specifically, a smaller timestep indicates better performance in efficient computing, resulting in reduced latency and operations. While, using a small timestep may lead to low accuracy due to insufficient information presentation with few spikes. This observation motivates us to develop an SNN that is more reliable for adaptive timestep by introducing a novel regularisation technique, namely Spatial-Temporal Regulariser (STR). Our approach regulates the ratio between the strength of spikes and membrane potential at each timestep. This effectively balances spatial and temporal performance during training, ultimately resulting in an Anytime Optimal Inference (AOI) SNN. Through extensive experiments on frame-based and event-based datasets, our method, in combination with cutoff based on softmax output, achieves state-of-the-art performance in terms of both latency and accuracy. Notably, with STR and cutoff, SNN achieves 2.14 to 2.89 faster in inference compared to the pre-configured timestep with near-zero accuracy drop of 0.50% to 0.64% over the event-based datasets. Code available: https://github.com/Dengyu-Wu/AOI-SNN-Regularisation

cs.NE↗

arXiv2024

Combinatorial p-th Calabi Flows for Total Geodesic Curvatures in hyperbolic background geometry

In hyperbolic background geometry, we investigate a generalized circle packing (including circles, horocycles and hypercycles) with conical singularities on a surface with boundary, which has a total geodesic curvature on each generalized circle of this circle packing and a discrete Gaussian curvature on the center of each dual circle. The purpose of this paper is to find this type of circle packings with prescribed total geodesic curvatures on generalized circles and discrete Gaussian curvatures on centers of dual circles. To achieve this goal, we firstly establish existence and rigidity on this type of circle packings by the variational principle. Secondly, for $p>1$, we introduce combinatorial $p$-th Calabi flows to find the circle packing with prescribed total geodesic curvatures on generalized circles and discrete Gaussian curvatures on centers of dual circles for the first time.

math.DG↗

arXiv2023

Hyperbolic Circle Packings and Total Geodesic Curvatures on Surfaces with Boundary

This paper investigates a generalized hyperbolic circle packing (including circles, horocycles or hypercycles) with respect to the total geodesic curvatures on the surface with boundary. We mainly focus on the existence and rigidity of circle packing whose contact graph is the $1$-skeleton of a finite polygonal cellular decomposition, which is analogous to the construction of Bobenko and Springborn [4]. Motivated by Colin de Verdière's method [6], we introduce the variational principle for generalized hyperbolic circle packings on polygons. By analyzing limit behaviours of generalized circle packings on polygons, we give an existence and rigidity for the generalized hyperbolic circle packing with conical singularities regarding the total geodesic curvature on each vertex of the contact graph. As a consequence, we introduce the combinatoral Ricci flow to find a desired circle packing with a prescribed total geodesic curvature on each vertex of the contact graph.

math.GT↗

arXiv2023

A Survey of Safety and Trustworthiness of Large Language Models through the Lens of Verification and Validation

Large Language Models (LLMs) have exploded a new heatwave of AI for their ability to engage end-users in human-level conversations with detailed and articulate answers across many knowledge domains. In response to their fast adoption in many industrial applications, this survey concerns their safety and trustworthiness. First, we review known vulnerabilities and limitations of the LLMs, categorising them into inherent issues, attacks, and unintended bugs. Then, we consider if and how the Verification and Validation (V&V) techniques, which have been widely developed for traditional software and deep learning models such as convolutional neural networks as independent processes to check the alignment of their implementations against the specifications, can be integrated and further extended throughout the lifecycle of the LLMs to provide rigorous analysis to the safety and trustworthiness of LLMs and their applications. Specifically, we consider four complementary techniques: falsification and evaluation, verification, runtime monitoring, and regulations and ethical use. In total, 370+ references are considered to support the quick understanding of the safety and trustworthiness issues from the perspective of V&V. While intensive research has been conducted to identify the safety and trustworthiness issues, rigorous yet practical methods are called for to ensure the alignment of LLMs with safety and trustworthiness requirements.

cs.AI↗

arXiv2023

Extended magnetic reconnection in kinetic plasma turbulence

Magnetic reconnection and plasma turbulence are ubiquitous processes important for laboratory, space and astrophysical plasmas. Reconnection has been suggested to play an important role in the energetics and dynamics of turbulence by observations, simulations and theory for two decades. The fundamental properties of reconnection at kinetic scales, essential to understanding the general problem of reconnection in magnetized turbulence, remain largely unknown at present. Here we present an application of the magnetic flux transport method that can accurately identify reconnection in turbulence to a three-dimensional simulation. Contrary to ideas that reconnection in turbulence would be patchy and unpredictable, highly extended reconnection X-lines, on the same order of magnitude as the system size, form at kinetic scales. Extended X-lines develop through bi-directional reconnection spreading. They satisfy critical balance characteristic of turbulence, which predicts the X-line extent at a given scale. These results present a picture of fundamentally extended reconnection in kinetic-scale turbulence.

physics.plasm-ph↗

arXiv2023

STPA for Learning-Enabled Systems: A Survey and A New Practice

Systems Theoretic Process Analysis (STPA) is a systematic approach for hazard analysis that has been used across many industrial sectors including transportation, energy, and defense. The unstoppable trend of using Machine Learning (ML) in safety-critical systems has led to the pressing need of extending STPA to Learning-Enabled Systems (LESs). Although works have been carried out on various example LESs, without a systematic review, it is unclear how effective and generalisable the extended STPA methods are, and whether further improvements can be made. To this end, we present a systematic survey of 31 papers, summarising them from five perspectives (attributes of concern, objects under study, modifications, derivatives and processes being modelled). Furthermore, we identify room for improvement and accordingly introduce DeepSTPA, which enhances STPA from two aspects that are missing from the state-of-the-practice: (i) Control loop structures are explicitly extended to identify hazards from the data-driven development process spanning the ML lifecycle; (ii) Fine-grained functionalities are modelled at the layer-wise levels of ML models to detect root causes. We demonstrate and compare DeepSTPA and STPA through a case study on an autonomous emergency braking system.

cs.SE↗

arXiv2023

Safety Analysis in the Era of Large Language Models: A Case Study of STPA using ChatGPT

Can safety analysis make use of Large Language Models (LLMs)? A case study explores Systems Theoretic Process Analysis (STPA) applied to Automatic Emergency Brake (AEB) and Electricity Demand Side Management (DSM) systems using ChatGPT. We investigate how collaboration schemes, input semantic complexity, and prompt guidelines influence STPA results. Comparative results show that using ChatGPT without human intervention may be inadequate due to reliability related issues, but with careful design, it may outperform human experts. No statistically significant differences are found when varying the input semantic complexity or using common prompt guidelines, which suggests the necessity for developing domain-specific prompt engineering. We also highlight future challenges, including concerns about LLM trustworthiness and the necessity for standardisation and regulation in this domain.

cs.CL↗