Institution profile

University of Arizona

Academic institutionnorthamerica · us
Official website
Research library517linked papers
Opportunities0open roles
Selected work

Representative Papers

Exploring Silicon-Based Societies: An Early Study of the Moltbook Agent Community

Feb 02, 2026

This study investigates how large-scale autonomous language model agents self-organize into complex social structures and collective behaviors in open environments. We introduce and implement a novel “data-driven silicon sociology” framework, leveraging non-intrusive observational data from over 150,000 agents and their subcommunities on the Moltbook platform. By applying procedural data collection, text preprocessing, contextual embedding, and unsupervised clustering, we directly uncover emergent social structures from machine-generated content without relying on pre-defined human sociological categories. Our analysis reveals three reproducible organizational patterns: anthropomorphic interest-based communities, silicon-native reflective collectives, and nascent economic coordination behaviors. These findings provide both empirical grounding and methodological innovation for understanding the evolutionary dynamics of autonomous agent ecosystems.

3 citationsRead paper

LLM-MC-Affect: LLM-Based Monte Carlo Modeling of Affective Trajectories and Latent Ambiguity for Interpersonal Dynamic Insight

Jan 07, 2026arXiv.org

This work proposes a Monte Carlo modeling framework grounded in large language models to address the limitations of traditional sentiment analysis, which oversimplifies emotions as deterministic labels and fails to capture the subjectivity, ambiguity, and sequential coupling inherent in interpersonal interactions. By modeling emotions as continuous latent probability distributions, the approach leverages stochastic decoding and Monte Carlo estimation to generate high-fidelity emotional dynamics trajectories. The framework further introduces interpretable cross-correlation and slope metrics to quantify emotional lead-lag relationships between interlocutors. Applied to teacher-student dialogues, the method successfully identifies high-level interaction patterns such as scaffolding instruction, demonstrating strong interpretability and generalization in uncovering the nuanced dynamics of interpersonal affect.

1 citationsRead paper

Sampling Decisions

Mar 17, 2025

This work addresses guided sampling from target distributions in discrete space-time settings. Methodologically, it introduces the Decision Flow (DF) framework—a unified generalization of path-integral diffusion and generative flow networks. It is the first to extend continuous-space-time path-integral diffusion to discrete domains; constructs a linearly solvable, neural-network-free DF formulation, where a novel Markov process is defined via convolution of the target distribution with the inverse-time Green’s function; and algorithmically adapts stochastic optimal control principles—specifically Markov decision processes—to enable explicit, closed-form guided sampling. Contributions include: empirical validation on the Ising model demonstrating DF’s efficacy for analytical, neural-network-free sampling; and establishment of a rigorous theoretical foundation for neural-network-augmented guided sampling, thereby bridging stochastic control theory and generative modeling.

1 citationsRead paper

Cyber-Physical Security Vulnerabilities Identification and Classification in Smart Manufacturing -- A Defense-in-Depth Driven Framework and Taxonomy

Dec 29, 2024

Traditional vulnerability identification methods in smart manufacturing overlook physical-layer weaknesses and cross-domain coordination flaws due to deep cyber-physical-human coupling. To address this, this paper proposes a纵深-defense-oriented vulnerability identification and classification framework. It formally defines the “vulnerability–defense” duality in manufacturing contexts and establishes the first cyber-physical-human co-vulnerability taxonomy and纵深-defense model tailored to intelligent manufacturing. The framework spans five dimensions—cyberspace, human behavior, process monitoring,出厂 inspection, and organizational policy—enabling cross-domain vulnerability mapping, threat modeling, and coordinated evaluation of multi-layered security mechanisms. Evaluated on a representative smart production line, the framework successfully identifies exploitable cross-domain gaps missed by conventional approaches, significantly improving domain-specific adaptability and actionable guidance for defense deployment.

1 citationsRead paper
Recent publications

Latest Papers