arXiv subjects

Jiayi Wen

Publications and source records attributed to Jiayi Wen.

12 recordsLinked to original sources

arXiv2026

REVOLVE: An Automated Closed-Loop Framework for Evolving Robot Manipulation with Minimal Human Intervention

Recent advances in data-driven robot manipulation policies have substantially improved task execution and generalization. However, real-world deployment still relies heavily on humans for failure assessment, correction, and environment reset, while models often fail to continually learn from failures and corrective experience. We present REVOLVE (Robot Evolving via Orchestrated Loops, Verification, and Experience), an automated closed-loop framework for evolving robot manipulation with minimal human intervention. Built on a unified software platform, REVOLVE integrates data collection, policy training and deployment, failure recovery, and continual learning into a single closed-loop workflow. Its Automated Reset and Correction (ARC) architecture automatically resets the environment and intervenes to correct policy failures. Dual-Loop Evolution (DLE) continually improves the manipulation policy and agent by feeding real-world interaction and failure--correction data back into policy learning and using an external mismatch memory to refine agent judgments. Experiments across four real-world manipulation tasks show that, after five iterations, REVOLVE improves average policy success rate by 18.5% and agent judgment accuracy by 8.5%, while reducing human effort in data collection and deployment testing by 94.4% and 95.1%, respectively. These results demonstrate that REVOLVE transforms real-world deployment into a closed-loop learning process that continually accumulates and uses execution experience, enabling continual evolution of both the policy and supervisory model with substantially less human intervention.

cs.RO↗

arXiv2026

FEWT: Frequency-Enhanced Wavelet-based Transformer for Multimodal Wheeled Bimanual Manipulation

Embodied intelligence bridges the physical world and information spaces, with robots demonstrating immense potential through imitation learning algorithms. In this study, a custom-built wheeled bimanual robotic platform equipped with an exoskeleton-style teleoperation system was utilized to realize intuitive remote manipulation and the efficient collection of anthropomorphic action data. To overcome the representation mismatch between spatial visual semantics and localized high-frequency physical dynamics, we propose a lightweight frequency-aligned imitation-learning framework, termed the Frequency-Enhanced Wavelet-based Transformer (FEWT). FEWT integrates two primary modules: Frequency-Enhanced Efficient Multi-Scale Attention (FE-EMA) and Time-Series Discrete Wavelet Transform (TS-DWT) to explicitly extract and align multi-scale features, improving the compatibility between spatial visual representations and temporal-frequency recalibration. Crucially, for real-world deployment, this framework is further extended into a multimodal system by seamlessly integrating a self-developed Smart Tactile Fabric (STF) sensor into the physical end-effectors, providing local contact-stress information that complements proprioceptive and chassis-motion cues in the shared multimodal representation. Experimental evaluations demonstrate that the core FEWT architecture significantly improves the success rate over the widely used Action Chunking with Transformers baseline, particularly during the most challenging phases of simulated bimanual insertion tasks. Furthermore, in complex real-world mobile and desktop manipulation tasks, the full STF-enhanced system effectively adapts to microscopic dynamic perturbations, yielding substantial performance enhancements.

cs.RO↗

arXiv2026

On some Grothendieck expansions

The orthogonal and symplectic groups act on the complete flag variety with finitely many orbits. The orthogonal Grothendieck polynomials $\mathfrak{G}^{\mathsf{O}}_z$ and symplectic Grothendieck polynomials $\mathfrak{G}^{\mathsf{Sp}}_z$ are distinguished representatives for the $K$-theory classes of the corresponding orbit closures. There is a simple formula to expand $\mathfrak{G}^{\mathsf{Sp}}_z$ as a linear combination of Grothendieck polynomials $\mathfrak{G}^{(β)}_w$, which represent the $K$-theory classes of Schubert varieties. Although the constructions of $\mathfrak{G}^{\mathsf{Sp}}_z$ and $\mathfrak{G}^{\mathsf{O}}_z$ are similar, finding the $\mathfrak{G}^{(β)}$-expansion of $\mathfrak{G}^{\mathsf{O}}_z$ or even computing $\mathfrak{G}^{\mathsf{O}}_z$ is much harder. If $z$ is vexillary then $\mathfrak{G}^{\mathsf{O}}_z$ has a nonnegative $\mathfrak{G}^{(β)}$-expansion, but the associated coefficients are mostly unknown. This paper derives several new formulas for $\mathfrak{G}^{\mathsf{O}}_z$ and its $\mathfrak{G}^{(β)}$-expansion when $z$ is vexillary. Among other applications, we prove that the latter expansion has a nontrivial stability property.

math.CO↗

arXiv2026

RoboMatch: A Unified Mobile-Manipulation Teleoperation Platform with Auto-Matching Network Architecture for Long-Horizon Tasks

This paper presents RoboMatch, a novel unified teleoperation platform for mobile manipulation with an auto-matching network architecture, designed to tackle long-horizon tasks in dynamic environments. Our system enhances teleoperation performance, data collection efficiency, task accuracy, and operational stability. The core of RoboMatch is a cockpit-style control interface that enables synchronous operation of the mobile base and dual arms, significantly improving control precision and data collection. Moreover, we introduce the Proprioceptive-Visual Enhanced Diffusion Policy (PVE-DP), which leverages Discrete Wavelet Transform (DWT) for multi-scale visual feature extraction and integrates high-precision IMUs at the end-effector to enrich proprioceptive feedback, substantially boosting fine manipulation performance. Furthermore, we propose an Auto-Matching Network (AMN) architecture that decomposes long-horizon tasks into logical sequences and dynamically assigns lightweight pre-trained models for distributed inference. Experimental results demonstrate that our approach improves data collection efficiency by over 20%, increases task success rates by 20-30% with PVE-DP, and enhances long-horizon inference performance by approximately 40% with AMN, offering a robust solution for complex manipulation tasks. Project website: https://robomatch.github.io

cs.RO↗

arXiv2025

PrivDFS: Private Inference via Distributed Feature Sharing against Data Reconstruction Attacks

In this paper, we introduce PrivDFS, a distributed feature-sharing framework for input-private inference in image classification. A single holistic intermediate representation in split inference gives diffusion-based Data Reconstruction Attacks (DRAs) sufficient signal to reconstruct the input with high fidelity. PrivDFS restructures this vulnerability by fragmenting the representation and processing the fragments independently across a majority-honest set of servers. As a result, each branch observes only an incomplete and reconstruction-insufficient view of the input. To realize this, PrivDFS employs learnable binary masks that partition the intermediate representation into sparse and largely non-overlapping feature shares, each processed by a separate server, while a lightweight fusion module aggregates their predictions on the client. This design preserves full task accuracy when all branches are combined, yet sharply limits the reconstructive power available to any individual server. PrivDFS applies seamlessly to both ResNet-based CNNs and Vision Transformers. Across CIFAR-10/100, CelebA, and ImageNet-1K, PrivDFS induces a pronounced collapse in DRA performance, e.g., on CIFAR-10, PSNR drops from 23.25 -> 12.72 and SSIM from 0.963 -> 0.260, while maintaining accuracy within 1% of non-private split inference. These results establish structural feature partitioning as a practical and architecture-agnostic approach to reducing reconstructive leakage in cloud-based vision inference.

cs.LG↗

arXiv2025

A Few Words Can Distort Graphs: Knowledge Poisoning Attacks on Graph-based Retrieval-Augmented Generation of Large Language Models

Graph-based Retrieval-Augmented Generation (GraphRAG) has recently emerged as a promising paradigm for enhancing large language models (LLMs) by converting raw text into structured knowledge graphs, improving both accuracy and explainability. However, GraphRAG relies on LLMs to extract knowledge from raw text during graph construction, and this process can be maliciously manipulated to implant misleading information. Targeting this attack surface, we propose two knowledge poisoning attacks (KPAs) and demonstrate that modifying only a few words in the source text can significantly change the constructed graph, poison the GraphRAG, and severely mislead downstream reasoning. The first attack, named Targeted KPA (TKPA), utilizes graph-theoretic analysis to locate vulnerable nodes in the generated graphs and rewrites the corresponding narratives with LLMs, achieving precise control over specific question-answering (QA) outcomes with a success rate of 93.1\%, while keeping the poisoned text fluent and natural. The second attack, named Universal KPA (UKPA), exploits linguistic cues such as pronouns and dependency relations to disrupt the structural integrity of the generated graph by altering globally influential words. With fewer than 0.05\% of full text modified, the QA accuracy collapses from 95\% to 50\%. Furthermore, experiments show that state-of-the-art defense methods fail to detect these attacks, highlighting that securing GraphRAG pipelines against knowledge poisoning remains largely unexplored.

cs.CL↗

arXiv2025

Assortative Marriage and Geographic Sorting

Between 1980 and 2000, the U.S. experienced a significant rise in geographic sorting and educational homogamy, with college graduates increasingly concentrating in high-skill cities and marrying similarly educated spouses. We develop and estimate a spatial equilibrium model with local labor, housing, and marriage markets, incorporating a marriage matching framework with transferable utility. Using the model, we estimate trends in assortative preferences, quantify the interplay between marital and geographic sorting, and assess their combined impact on household inequality. Welfare analyses show that after accounting for marriage, the college well-being gap grew substantially more than the college wage gap.

econ.EM↗

arXiv2024

Insuring Long-Term Care in Developing Countries: The Interaction between Formal and Informal Insurance

Does public insurance reduce uninsured long-term care (LTC) risks in developing countries, where informal insurance predominates? This paper exploits the rollout of LTC insurance in China around 2016 to examine the impact of public LTC insurance on healthy workers' labor supply, a critical self-insurance channel. We find that workers eligible for public LTC insurance were less likely to engage in labor work and worked fewer weeks annually following the policy change, suggesting a mitigation of uninsured risks. However, these impacts were insignificant among those with strong informal insurance coverage. Parallel changes in anticipated formal care use corroborate these findings. While our results reveal that public LTC insurance provides limited additional risk-sharing when informal insurance predominates, they also underscore its growing importance.

econ.GN↗

arXiv2024

Demazure crystals for flagged key polynomials

One definition of key polynomials is as the weight generating functions of key tableaux. Assaf and Schilling introduced a crystal structure on key tableaux and related it to Morse--Schilling crystals on reduced factorizations for permutations via weak Edelman--Greene insertion. In this paper, we consider generalizations of both crystals depending on a flag. We extend weak EG insertion to a bijection between our flagged objects and show that the recording tableau gives a crystal isomorphism. As an application, we show that flagged key tableaux have a natural Demazure crystal structure, whose characters recover Reiner and Shimozono's flagged key polynomials.

math.CO↗

arXiv2024

A New Testing Method for Justification Bias Using High-Frequency Data of Health and Employment

Justification bias, wherein retirees may report poorer health to rationalize their retirement, poses a major concern to the widely-used measure of self-assessed health in retirement studies. This paper introduces a novel method for testing the presence of this bias in the spirit of regression discontinuity. The underlying idea is that any sudden shift in self-assessed health immediately following retirement is more likely attributable to the bias. Our strategy is facilitated by a unique high-frequency data that offers monthly, in contrast to the typical biennial, information on employment, self-assessed health, and objective health conditions. Across a wider post-retirement time frame, we observe a decline in self-assessed health, potentially stemming from both justification bias and changes in actual health. However, this adverse effect diminishes with shorter intervals, indicating no evidence of such bias. Our method also validates a widely-used indirect testing approach.

econ.GN↗

arXiv2023

Occupational Retirement and Pension Reform: The Roles of Physical and Cognitive Health

Despite increasing cognitive demands of jobs, knowledge about the role of health in retirement has centered on its physical dimensions. This paper estimates a dynamic programming model of retirement that incorporates multiple health dimensions, allowing differential effects on labor supply across occupations. Results show that the effect of cognitive health surges exponentially after age 65, and it explains a notable share of employment declines in cognitively demanding occupations. Under pension reforms, physical constraint mainly impedes manual workers from delaying retirement, whereas cognitive constraint dampens the response of clerical and professional workers. Multidimensional health thus unevenly exacerbate welfare losses across occupations.

econ.GN↗

arXiv2023

Parental Health Penalty on Adult Children's Employment: Gender Difference and Long-Term Consequence

This paper examines the long-term gender-specific impacts of parental health shocks on adult children's employment in China. We build up an inter-temporal cooperative framework to analyze household work decisions in response to parental health deterioration. Then employing an event-study approach, we establish a causal link between parental health shocks and a notable decline in female employment rates. Male employment, however, remains largely unaffected. This negative impact shows no abatement up to eight years that are observable by the sample. These findings indicate the consequence of "growing old before getting rich" for developing countries.

econ.GN↗