arXiv subjects

Zhiyan Li

Publications and source records attributed to Zhiyan Li.

5 recordsLinked to original sources

arXiv2026

Mirror: A Multi-Agent System for AI-Assisted Ethics Review

Ethics review is a foundational mechanism of modern research governance, yet contemporary systems face increasing strain as ethical risks arise as structural consequences of large-scale, interdisciplinary scientific practice. The demand for consistent and defensible decisions under heterogeneous risk profiles exposes limitations in institutional review capacity rather than in the legitimacy of ethics oversight. Recent advances in large language models (LLMs) offer new opportunities to support ethics review, but their direct application remains limited by insufficient ethical reasoning capability, weak integration with regulatory structures, and strict privacy constraints on authentic review materials. In this work, we introduce Mirror, an agentic framework for AI-assisted ethical review that integrates ethical reasoning, structured rule interpretation, and multi-agent deliberation within a unified architecture. At its core is EthicsLLM, a foundational model fine-tuned on EthicsQA, a specialized dataset of 41K question-chain-of-thought-answer triples distilled from authoritative ethics and regulatory corpora. EthicsLLM provides detailed normative and regulatory understanding, enabling Mirror to operate in two complementary modes. Mirror-ER (expedited Review) automates expedited review through an executable rule base that supports efficient and transparent compliance checks for minimal-risk studies. Mirror-CR (Committee Review) simulates full-board deliberation through coordinated interactions among expert agents, an ethics secretary agent, and a principal investigator agent, producing structured, committee-level assessments across ten ethical dimensions. Empirical evaluations demonstrate that Mirror significantly improves the quality, consistency, and professionalism of ethics assessments compared with strong generalist LLMs.

cs.AI↗

arXiv2025

AMP-based Joint Activity Detection and Channel Estimation for Massive Grant-Free Access in OFDM-based Wideband Systems

To realize orthogonal frequency division multiplexing (OFDM)-based grant-free access for wideband systems under frequency-selective fading, existing device activity detection and channel estimation methods need substantial accuracy improvement or computation time reduction. In this paper, we aim to resolve this issue. First, we present an exact time-domain signal model for OFDM-based grant-free access under frequency-selective fading. Then, we present a maximum a posteriori (MAP)-based device activity detection problem and two minimum mean square error (MMSE)-based channel estimation problems. The MAP-based device activity detection problem and one of the MMSE-based channel estimation problems are formulated for the first time. Next, we build a new factor graph that captures the exact statistics of time-domain channels and device activities. Based on it, we propose two approximate message passing (AMP)-based algorithms, AMP-A-EC and AMP-A-AC, to approximately solve the MAP-based device activity detection problem and two MMSE-based channel estimation problems. Both proposed algorithms alleviate the AMP's inherent convergence problem when the pilot length is smaller or comparable to the number of active devices. Then, we analyze AMP-A-EC's error probability of activity detection and mean square error (MSE) of channel estimation via state evolution and show that AMP-A-AC has the lower computational complexity (in dominant term). Finally, numerical results show the two proposed AMP-based algorithms' superior performance and respective preferable regions, revealing their significant values for OFDM-based grant-free access.

cs.IT↗

arXiv2025

Federated Linear Dueling Bandits

Contextual linear dueling bandits have recently garnered significant attention due to their widespread applications in important domains such as recommender systems and large language models. Classical dueling bandit algorithms are typically only applicable to a single agent. However, many applications of dueling bandits involve multiple agents who wish to collaborate for improved performance yet are unwilling to share their data. This motivates us to draw inspirations from federated learning, which involves multiple agents aiming to collaboratively train their neural networks via gradient descent (GD) without sharing their raw data. Previous works have developed federated linear bandit algorithms which rely on closed-form updates of the bandit parameters (e.g., the linear function parameters) to achieve collaboration. However, in linear dueling bandits, the linear function parameters lack a closed-form expression and their estimation requires minimizing a loss function. This renders these previous methods inapplicable. In this work, we overcome this challenge through an innovative and principled combination of online gradient descent (OGD, for minimizing the loss function to estimate the linear function parameters) and federated learning, hence introducing our federated linear dueling bandit with OGD (FLDB-OGD) algorithm. Through rigorous theoretical analysis, we prove that FLDB-OGD enjoys a sub-linear upper bound on its cumulative regret and demonstrate a theoretical trade-off between regret and communication complexity. We conduct empirical experiments to demonstrate the effectiveness of FLDB-OGD and reveal valuable insights, such as the benefit of a larger number of agents, the regret-communication trade-off, among others.

cs.LG↗

arXiv2016

A Novel Human Computer Interaction Platform based College Mathematical Education Methodology

This article proposes the analysis on novel human computer interaction (HCI) platform based college mathematical education methodology. Above for the application of virtual reality technology in teaching the problems in the study, only through the organization focus on the professional and technical personnel, and constantly improve researchers in development process of professional knowledge, close to the actual needs of the teaching can we achieve the satisfactory result. To obtain better education output, we combine the Kinect to form the HCI based teaching environment. We firstly review the latest HCI technique and principles of college math courses, then we introduce basic components of the Kinect including the gesture segmentation, systematic implementation and the primary characteristics of the platform. As the further step, we implement the system with the re-write of script code to build up the personalized HCI assisted education scenario. The verification and simulation proves the feasibility of our method.

cs.HC↗

arXiv2016

Research on Information Security Enhancement Approaches and the Applications on HCI Systems

With rapid development of computer techniques, the human computer interaction scenarios are becoming more and more frequent. The development history of the human-computer interaction is from a person to adapt to the computer to the computer and continually adapt to the rapid development. Facing the process of human-computer interaction, information system daily operation to produce huge amounts of data, how to ensure human-computer interaction interface clear, generated data safe and reliable, has become a problem to be solved in the world of information. To deal with the challenging, we propose the information security enhancement approaches and the core applications on HCI systems. Through reviewing the other state-of-the-art methods, we propose the data encryption system to deal with the issues that uses mixed encryption system to make full use of the symmetric cipher algorithm encryption speed and encryption intensity is high while the encryption of large amounts of data efficiently. Our method could enhance the general safety of the HCI system, the experimental result verities the feasibility and general robustness of our approach.

cs.HC↗