Searcharxiv⌕ Search

arXiv subjects

Kai Yan

Publications and source records attributed to Kai Yan.

At least 19 recordsLinked to original sources
arXiv2026

DiffSequence-Q: Diffusion Design with Quantum Assistance for Biologically Active Proteins

Designing protein sequences that satisfy both secondary structure requirements and solubility targets remains challenging. We present DiffSequence-Q, a hybrid multinomial diffusion model conditioned on secondary structure at each residue and a continuous solubility target. To introduce quantum auxiliary supervision during training, we employ a quantum neural network (QNN) or quantum Kolmogorov-Arnold network (QKAN) as an auxiliary solubility prediction head over shared Transformer representations. The auxiliary prediction loss is jointly optimized with the diffusion objective, allowing quantum auxiliary supervision to influence the learned denoising model. Once trained, protein sequences are generated entirely through classical reverse diffusion. On the CATH benchmark, DiffSequence-Q achieved the highest secondary structure recovery among the four evaluated models for complete sequence generation. Evaluation with a separately trained solubility prediction model showed that the best QNN and QKAN configurations achieved mean predicted solubility scores 14.0% and 22.7% higher than the multilayer perceptron (MLP) baseline, respectively. Experimental characterization further demonstrated biological activity across three distinct protein families. Selected defensins exhibited pronounced inhibitory activity against Aspergillus niger, while xylanases and short-chain dehydrogenases exhibited substantial enzymatic activity. Together, these results support DiffSequence-Q as a framework for integrating structural requirements and solubility targets in protein generation, with experimentally demonstrated biological activity.

quant-ph↗
arXiv2026

A Control Line Multiplexing Framework for Scalable Quantum Computing

Universal fault-tolerant superconducting quantum computers are expected to require millions of physical qubits; yet extending the prevailing dedicated-control-line paradigm to this scale would exceed the wiring density and cooling-power budget of dilution refrigerators. Microwave frequency-division multiplexing (FDM) revolutionized classical communications by enabling many independent channels to share common transmission infrastructure. Translating this principle to superconducting quantum control, however, requires an all-microwave architecture and experimental validation of frequency-multiplexed gate control. Recent work in quantum error correction has demonstrated microwave-based readout, single-qubit gates and leakage suppression, whereas two-qubit gates remain predominantly driven by unipolar pulses and thus constitute the missing element of an all-microwave architecture. Here, we propose a periodically extensible hybrid multiplexing framework in which all gate operations are frequency-division multiplexed using polychromatic microwave pulses and band-pass filters, thereby reducing the required control-line count by one to two orders of magnitude. In a 1:4 demonstration, we achieve average errors of $(6.4\pm0.2)\times10^{-3}$ for parallel FDM CZ gates and $(10.6\pm0.2)\times10^{-4}$ for parallel FDM $XY(π/2)$ gates, without statistically significant multiplexing-induced degradation. Together with on-chip DC sources, this FDM-based gate-control scheme completes the last missing module in a scalable multiplexed-control architecture toward universal fault-tolerant quantum computing.

quant-ph↗
arXiv2026

Vision-Language-Action Autonomous Driving Agent with Language-based Memory

Vision-Language-Action (VLA) foundation models have recently emerged as one of the prevailing solutions for autonomous driving, as they can utilize knowledge acquired during vision-language pretraining for accurate and interpretable driving. However, VLAs can take only a limited number of frames as visual input due to the high token cost of an image, which is problematic for memory-dependent tasks such as determining the arrival order at all-way stops and long-horizon driving scene understanding. Existing solutions use latent vector memories accessed through cross-attention, which are neither interpretable nor portable. In this paper, we propose AD-Memo, a general-purpose VLA driving agent with language-based memory. The agent outputs memory as an extension of its Chain-of-Thought (CoT) to record surrounding objects critical to driving; this memory becomes part of the agent's future input. We curate memory-based datasets and train VLAs with a two-stage recipe: Supervised Fine-Tuning (SFT) and \textit{Da Capo}, a novel semi-closed-loop Reinforcement Learning (RL) algorithm which uses trajectory-level advantage for memory and step-level advantage for driving, leading to better credit assignment. Across scenarios such as all-way stops and general driving, AD-Memo improves driving quality, enables better question answering on driving scenes, and produces plug-and-play memory for other models.

cs.CV↗
arXiv2026

Geometry as Address: Routing Attention to Visual Memory for Long-Horizon Camera-Controlled Video Generation

Long-horizon camera-controlled video generation requires recovering previously observed content from an ever-growing visual history. Existing approaches either search historical context implicitly or reconstruct it into persistent 3D memory, facing inefficient memory access or accumulated geometric errors. Our key insight is that geometry need not explain the scene--it only needs to determine where visual memory should be read from, while attention decides what should be recovered. Based on this insight, we introduce GEAR, a Geometry-Enabled Attention Routing framework that uses geometry as an explicit token-level address for visual memory. Rather than fusing historical observations into a persistent global 3D representation, GEAR retains them as frame latents and uses per-frame geometry only to establish token-level correspondences with target views, thereby avoiding persistent error accumulation from global fusion. Guided by these correspondences, Geometric Correspondence Attention (GCA) selectively injects geometrically matched historical features into noisy target patches during denoising. We further introduce an Invisible Octree to accumulate visibility evidence and reject geometrically plausible but occluded correspondences. Extensive experiments demonstrate that GEAR achieves state-of-the-art visual quality, precise camera control, and revisit consistency, enabling minute-long video generation along challenging trajectories.

cs.CV↗
arXiv2026

Global dynamics and asymptotic stability for a nonlocal LANS-$α$-type system

This paper is devoted to the global dynamics of a nonlocal Lagrangian-averaged Navier-Stokes-$α$ (LANS-$α$)-type system (also known as the viscous Euler-Poincaré system) in one, two, and three spatial dimensions. In the unforced periodic setting with zero mean, we establish global well-posedness for the initial-value problem and prove that the corresponding solution semigroup is dissipative and possesses a compact global attractor consisting of the singleton $\{0\}$. Furthermore, we obtain the following global-in-time algebraic asymptotic stability estimate: \[ \|u(t,\cdot)\|_{H^3}\lesssim t^{-1/2},\qquad \forall\, t>0, \] which is derived from higher-order energy inequalities combined with a uniform Gronwall-type argument. Our results provide a complete description of the long-time behavior of the unforced periodic dynamics for this nonlocal LANS-$α$-type system.

math.AP↗
arXiv2026

ReFigBench: Benchmarking Scientific Figure Reconstruction as Editable PowerPoint Artifacts

Multimodal coding agents are expected to turn visual inputs into usable artifacts, and they act through a harness, the layer of tools, context management, and execution environment around the model. Existing evaluations often isolate short tool calls, API traces, or screenshot resemblance, and a low score under these proxies cannot say whether the model saw poorly, planned poorly, or was failed by its harness. We study scientific overview figure reconstruction, an agent task in which a source image must become an editable PowerPoint slide that preserves text, topology, layout, and native document structure. We introduce ReFigBench, a benchmark and evaluation framework built on 1,000 real overview figures retrieved from arXiv papers with full provenance. Coding agents from four model families reconstruct every figure under two workflows, direct code generation and a specialized PPTX workflow, and the strongest model runs inside two commercial harnesses, yielding ten configurations. Evaluation combines deterministic artifact checks, repeated automated scoring by judges from two model families, and blinded human comparisons. Perception remains a bottleneck that iterative rendering only partly repays. Whether workflow effort converts into quality depends on the model together with its harness, since the same model gains from the specialized workflow inside one harness and loses inside the other, and the harness shifts scores even under an identical direct prompt. The specialized workflow erases native connectors in every configuration, human judges still prefer its renderings in most matchups, and even the strongest agent falls short of the rubric ceiling. These results expose the tension between fidelity and editability as the central challenge for practical multimodal document agents.

cs.CL↗
arXiv2026

Real-Time Neural Hair G-Buffer Anti-Aliasing

We propose a lightweight real-time method for reconstructing strand-based hair G-Buffers from severely undersampled rasterized inputs. Our pipeline first applies neural spatial reconstruction and temporal accumulation to recover hair coverage, i.e., fractional hair visibility within a pixel, and tangent. It then uses a tangent-guided reconstruction step to complete the position, which is subsequently used for physically based deferred hair shading. We evaluate our method across a diverse set of hairstyles, including straight, wavy, afro, and ponytail styles, under both static and dynamic scenarios. Our method achieves higher hair reconstruction quality than general industrial neural reconstruction solutions such as DLSS and FSR.

cs.GR↗
arXiv2026

Scenix: Sparse-View 3D Scene Reconstruction via Executable Scene Programs

Synthesizing a structured and editable 3D indoor scene from a few uncalibrated RGB views requires more than generating high-quality individual assets: a system must infer the room structure, associate objects across incomplete observations, and recover a globally consistent spatial configuration. Previous methods mainly focus on 3D scene generation with text input or require continuous visual inputs with additional priors, \ e.g., human-annotated masks or accurate 3D layouts, which makes these methods labor demanding and hard to apply in general cases. We present \textsc{Scenix}, a sparse-view 3D scene reconstruction framework via executable scene programs, a structured representation that can be directly instantiated into editable 3D scenes. Given sparse views, \textsc{Scenix} predicts executable scene programs through perception-grounded asset instantiation and closed-loop spatial refinement. % We present \method, a framework that predicts an executable scene representation from sparse views and realizes it through perception-grounded asset instantiation and closed-loop spatial refinement. To support this task, we construct \dataset, a dataset of approximately 110,000 synthetic and real indoor scenes with multiview imagery, room structures, object-centric descriptions, and metric spatial annotations. We further introduce observation-consistent supervision that aligns each target scene with the visual evidence available in its input views. Experiments on held-out \textsc{XScene} scenes, real indoor images, and out-of-distribution SpatialGen cases evaluate structured scene prediction, object grounding, and spatial refinement.

cs.CV↗
arXiv2026

On a two-dimensional Camassa-Holm-Zakharov-Kuznetsov equation

This paper is devoted to a new two-dimensional nonlinear dispersive wave model named as the Camassa-Holm-Zakharov-Kuznetsov (CH-ZK) equation which combines the nonlinear structure of the Camassa-Holm equation with the transverse Laplacian dispersion of the Zakharov-Kuznetsov equation. We first establish the local well-posedness of its Cauchy problem in a suitable Sobolev space and derive a blow-up criterion for strong solutions. Then the finite-time blow-up strong solutions for the CH-ZK equation have been constructed. Moreover, we prove a unique continuation property for the solutions to the CH-ZK equation. Finally, we investigate the existence of both peaked and smooth solitary waves, and obtain a rigidity theorem for the traveling wave solutions according to the magnitude of wave speed.

math.AP↗
arXiv2026

Non-Markovian Cosmic-Ray Pitch-Angle Transport from Mirror Interactions

Cosmic-ray pitch-angle transport in magnetohydrodynamic (MHD) turbulence is governed by the interplay between magnetic mirroring and gyroresonant scattering. We develop a guiding-center (GC) Langevin model with explicit mirror drift and gyroresonant diffusion to describe the pitch angle evolution. This model accurately captures our test-particle simulation results in three-dimensional MHD turbulence, driven both solenoidally and compressively. We find that magnetic mirroring can drive anomalous pitch-angle diffusion at large pitch angles (including $90^\circ$) with non-Markovian memory effects, which arises from trapping of particles in magnetic wells. Gyroresonant scattering controls the escape rate from these wells. Across $M_{\rm A}$, large-pitch-angle particles are jointly regulated by mirror trapping and gyroresonant escape, exhibiting a transition from anomalous to normal diffusive pitch-angle transport as scattering strengthens, whereas small-pitch-angle particles remain gyroresonance-dominated and diffusive throughout. The pitch angle transport is found to be dominated by the compressible perturbations with marginal influence from Alfvén modes. In compressible turbulence with realistic damping accounted for, transit time damping (TTD) treatment fully recovers mirror interactions.

astro-ph.HE↗
arXiv2026

Sharp well-posedness and ill-posedness of the Camassa-Holm equation in critical Triebel-Lizorkin spaces

This paper is devoted to the sharp well-posedness and ill-posedness of the Cauchy problem for the Camassa-Holm (CH) equation in critical Triebel-Lizorkin spaces $F^{1+\frac{1}{p}}_{p,q}(\mathbb{R})$ with $(p,q)\in[1,\infty)\times[1,\infty]$ or $p=q=\infty$. On the one hand, we establish the local well-posedness in the sense of Hadamard in $F^2_{1,q}(\mathbb{R})$ for $1\leq q<\infty$ via Lagrangian coordinate transformation. On the other hand, by means of smooth atomic decomposition, strong ill-posedness is then proved in $F^{1+\frac{1}{p}}_{p,q}(\mathbb{R})$ with $(p,q)\in(1,\infty)\times[1,\infty]$ or $p=q=\infty$ in the sense of norm inflation, which in particular yields the ill-posedness of CH in critical Sobolev spaces $W^{1+\frac{1}{p},p}(\mathbb{R})$ with $1<p<\infty$, and provides a new perspective on the ill-posedness of CH in $H^{\frac{3}{2}}(\mathbb{R})$.

math.AP↗
arXiv2026

On a rod-Kadomtsev-Petviashvili shallow water equation in two dimensions

In this paper, we derive a new two-dimensional rod-Kadomtsev-Petviashvili (rod-KP) equation from the incompressible and irrotational three-dimensional Euler equation under the shallow water scaling. We establish the local well-posedness of the Cauchy problem in a suitable Sobolev space and derive the blow-up criterion for strong solutions via the energy method. Then, by exploring two appropriate conservation laws of the rod-KP equation, we are able to construct its global strong solution when the physical dimensionless parameter $σ=0$, and on the other hand produce the finite-time blow up solutions under some certain conditions when $σ\neq 0$. Furthermore, we present a uniqueness continuation property for the solutions. Finally, we investigate the existence of traveling-wave solutions in order to highlight the influence of weak transverse effects on wave stability, and we also exhibit the symmetry of solitary waves in the propagation direction.

math.AP↗
arXiv2026

Transformer refined quantum sampling for strongly correlated electronic structure

Although quantum computing offers a promising solution for strongly correlated system simulation, existing algorithms face significant bottlenecks on current noisy intermediate-scale quantum (NISQ) devices. Here, we introduce QiankunNet-QSCI, a hybrid quantum-classical framework that addresses this challenge by combining efficient quantum-sampling with a transformer neural network. An efficient unitary selected configuration Interaction (USCI) ansatz especially designed for quantum sampling is proposed to identify the most chemically significant electronic configurations on the Zuchongzhi 3.1 quantum processor. Subsequently, the transformer model QiankunNet learns from these sparse yet critical quantum data to infer and reconstruct the complete electronic wavefunction with high fidelity. Simulation of the challenging 40-qubit [2Fe-2S] ferredoxin active center achieves chemical accuracy. Simulation of the nitrogenase P-cluster in a 114-electron 73-orbital active space also reaches 12 milli-Hartree-level agreement with the best density matrix renormalization group (DMRG) result. QiankunNet-QSCI thus offers a practical route to accurate quantum-assisted electronic structure calculations on current devices.

quant-ph↗
arXiv2026

Boosting Reinforcement Learning with Verifiable Rewards via Randomly Selected Few-Shot Guidance

Reinforcement Learning with Verifiable Rewards (RLVR) has achieved great success in developing Large Language Models (LLMs) with chain-of-thought rollouts for many tasks such as math and coding. Nevertheless, RLVR struggles with sample efficiency on difficult problems where correct rollouts are hard to generate. Prior works propose to address this issue via demonstration-guided RLVR, i.e., to conduct Supervised FineTuning (SFT) when RL fails; however, SFT often requires a lot of data, which can be expensive to acquire. In this paper, we propose FEST, a FEw-ShoT demonstration-guided RLVR algorithm. It attains compelling results with only 128 demonstrations randomly selected from an SFT dataset. We find that three components are vital for the success: supervised signal, on-policy signal, and decaying weights on the few-shot SFT dataset to prevent overfitting from multiple-epoch training. On several benchmarks, FEST outperforms baselines with magnitudes less SFT data, even matching their performance with full dataset.

cs.LG↗
arXiv2026

Commutator estimates and their applications to the transport-type equations

In this paper, we derive new commutator estimates in the Triebel-Lizorkin spaces by employing Bony's para-product decomposition, the Nikol'skij representation, and the Fefferman-Stein vector-valued maximal function. These estimates are then applied to develop a general theory for transport equations. Although analogous results are already available in the setting of Besov spaces, the methods developed there do not carry over directly to the Triebel-Lizorkin case. Our approach works for Triebel-Lizorkin spaces and, as a byproduct, also yields the corresponding results in Besov spaces. All proofs are presented in a unified manner that applies to both scales of function spaces, thereby extending and sharpening previous results on transport equations in these frameworks. Furthermore, the general theory we obtain is widely applicable to evolution equations, including incompressible and compressible ideal fluid flows, shallow water waves, and related models. As an illustration, we consider the two-component Euler-Poincaré system. Using the theoretical framework developed herein, we establish its local well-posedness and a blow-up criterion in both sub-critical and critical Triebel-Lizorkin spaces.

math.AP↗
arXiv2026

Autonomous Laparoscope Control through Unified Mechanics-Based Representation of Multimodal Intraoperative Information

Laparoscope-holding robots can provide surgeons with a stable laparoscopic field of view (FOV) and reduce the burden on human assistants. To maintain an ideal intraoperative FOV, the robot must continuously adjust the laparoscope pose according to intraoperative information. However, intraoperative multimodal signals, such as position, force/torque, and images, differ markedly in physical meaning and units, making it difficult to build a unified representation and to generate control commands that can be used directly for laparoscope control. To address this issue, we propose a laparoscope-holding robot control method based on unified mechanics modeling of multimodal information. First, we design mapping strategies for multiple intraoperative sources, including position, force/torque, and images, and unify them into an equivalent-wrench representation in the operational space. Then, using a task-priority scheme, we inject the wrenches into the task space and the null space, respectively, and synthesize laparoscope control commands via task-priority projection, thereby achieving consistent representation and coordinated fusion of multimodal information within a single framework. Finally, taking the intraoperative remote center of motion (RCM) position, force/torque sensor readings, and laparoscopic images as examples, we construct an RCM-constraint wrench to enforce the RCM geometric constraint and reduce the contact force at the trocar site, a laparoscope-manipulation wrench to enable compliant dragging, and an instrument-tracking wrench to achieve autonomous visual tracking of the instruments. Experiments on a surgical phantom and in vivo porcine trials demonstrate that the proposed method supports multi-task operation, including compliant laparoscope manipulation and autonomous instrument tracking, while maintaining the RCM constraint and reducing sustained trocar-site loading.

cs.RO↗
arXiv2026

XRZero-G0: Pushing the Frontier of Dexterous Robotic Manipulation with Interfaces, Quality and Ratios

The acquisition of high-quality, action-aligned demonstration data remains a fundamental bottleneck in scaling foundation models for dexterous robot manipulation. Although robot-free human demonstrations (e.g., the UMI paradigm) offer a scalable alternative to traditional teleoperation, current systems are constrained by sub-optimal hardware ergonomics, open-loop workflows, and a lack of systematic data-mixing strategies. To address these limitations, we present XRZero-G0, a hardware-software co-designed system for embodied data collection and policy learning. The system features an ergonomic, virtual reality interface equipped with a top-view camera and dual specialized grippers to directly improve collection efficiency. To ensure dataset reliability, we propose a closed-loop collection, inspection, training, and evaluation pipeline for non-proprioceptive data. This workflow achieves an 85% data validity rate and establishes a transparent mechanism for quality control. Furthermore, we investigate the empirical scaling behaviors and optimal mixing ratios of robot-free data. Extensive experiments indicate that combining a minimal volume of real-robot data with large-scale robot-free data (e.g., a 10:1 ratio) achieves performance comparable to exclusively real-robot datasets, while reducing acquisition costs by a factor of twenty. Utilizing XRZero-G0, we construct a 2,000-hour robot-free dataset that enables zero-shot cross-embodiment transfer to a target physical robot, demonstrating a highly scalable methodology for generalized real-world manipulation.Our project repository: https://github.com/X-Square-Robot/XRZero-G0

cs.RO↗