arXiv subjects

Shixin Wu

Publications and source records attributed to Shixin Wu.

4 recordsLinked to original sources

arXiv2026

VFEAgent: A Multimodal Agent Framework for End-to-End Automated Finite Element Analysis

Finite Element Analysis (FEA) serves as the cornerstone of modern engineering design. However, its workflow is inherently complex and relies heavily on domain expertise. Although recent efforts have integrated Large Language Models (LLMs) into FEA, existing approaches face limitations in handling multimodal inputs and executing complex tasks. To address these limitations, we propose VFEAgent, an end-to-end multi-agent system designed to automate FEA modeling and simulation directly from input images and problem descriptions. Our methodology integrates two core components: (1) a multimodal vision-language multi-agent pipeline that employs ReAct-driven reasoning to extract structured FEA specifications from heterogeneous inputs and (2) a verification-first code synthesis framework, incorporating robust self-debugging and fallback mechanisms to ensure executability and physical validity. We systematically evaluated the system across various engineering mechanics scenarios. The results demonstrate that VFEAgent achieves a high success rate in generating complete and physically valid simulations, outperforming LLM-based baseline methods in reliability and correctness. These findings validate the feasibility of automating the complete FEA workflow, highlighting the framework's potential to liberate engineers from tedious manual analysis.

cs.AI↗

arXiv2026

Universal Weakly Fault-Tolerant Quantum Computation via Code Switching in the [[8,3,2]] Code

Code-switching offers a route to universal, fault-tolerant quantum computation by circumventing the limitation implied by the Eastin-Knill theorem against a universal transversal gate set within a single quantum code. Here, we present a fault-tolerant code-switching protocol between two versions of the $[[8, 3, 2]]$ code. One version supports weakly fault-tolerant single-qubit Clifford gates, while the other supports a logical $\overline{\mathrm{CCZ}}$ gate via transversal $T/T^\dagger$ together with logical $\overline{\mathrm{CZ}}$, $\overline{\mathrm{CNOT}}$, and $\overline{\mathrm{SWAP}}$ gates. Because both codes have distance 2, the protocol operates in a postselected, error-detecting regime: single faults lead to detectable outcomes, and accepted runs exhibit quadratic suppression of logical error rates. This yields a universal scheme for postselected fault-tolerant computation. We validate the protocol numerically through simulations of state preparation, code switching, and a three-logical-qubit implementation of Grover's search.

quant-ph↗

arXiv2025

TTF-VLA: Temporal Token Fusion via Pixel-Attention Integration for Vision-Language-Action Models

Vision-Language-Action (VLA) models process visual inputs independently at each timestep, discarding valuable temporal information inherent in robotic manipulation tasks. This frame-by-frame processing makes models vulnerable to visual noise while ignoring the substantial coherence between consecutive frames in manipulation sequences. We propose Temporal Token Fusion (TTF), a training-free approach that intelligently integrates historical and current visual representations to enhance VLA inference quality. Our method employs dual-dimension detection combining efficient grayscale pixel difference analysis with attention-based semantic relevance assessment, enabling selective temporal token fusion through hard fusion strategies and keyframe anchoring to prevent error accumulation. Comprehensive experiments across LIBERO, SimplerEnv, and real robot tasks demonstrate consistent improvements: 4.0 percentage points average on LIBERO (72.4\% vs 68.4\% baseline), cross-environment validation on SimplerEnv (4.8\% relative improvement), and 8.7\% relative improvement on real robot tasks. Our approach proves model-agnostic, working across OpenVLA and VLA-Cache architectures. Notably, TTF reveals that selective Query matrix reuse in attention mechanisms enhances rather than compromises performance, suggesting promising directions for direct KQV matrix reuse strategies that achieve computational acceleration while improving task success rates.

cs.CV↗

arXiv2025

Bias-tailored single-shot quantum LDPC codes

Quantum hardware rarely suffers equal amounts of bit-flip ($X$) and phase-flip ($Z$) errors; one type is often much more common than the other. A code that is ``bias-tailored'' can exploit this imbalance, lowering the fault-tolerance overhead. A complementary idea, called "single-shot" error correction, aims to recover from data errors and noisy measurements in a single round of stabilizer readout, avoiding slow repetition cycles. In this work, we combine these two ideas and build a hierarchy of new quantum codes. The full construction starts from the syndrome-encoded hypergraph product code and then tailors it to the dominant error type. The resulting code keeps the single-shot guarantee for every noise model while boosting the threshold whenever $X$ and $Z$ errors are asymmetric. By removing carefully chosen blocks of stabilizers we obtain two trimmed variants. The first, called the simplified code, cuts the physical-qubit count by $1/6$ and halves the number of stabilizer measurements, yet its minimum distance grows quadratically compared to the standard design and its biased noise threshold is unchanged. The second, called the reduced code, achieves the same hardware savings but trades away single-shot protection for purely $X$ or purely $Z$ noise; instead it remains single-shot under balanced, or depolarizing, noise. In settings where strongly biased noise is likely, either trimmed code offers a less resource-intensive alternative to the full construction. As a concrete illustration, we lift the two-dimensional XZZX surface code to a three-dimensional cubic lattice and show that this ``3D XZZX'' code is an explicit member of the simplified family. Taken together, these bias-tailored single-shot codes provide an adjustable set of code design alternatives, allowing tradeoffs between hardware overhead and noise types.

quant-ph↗