Selected work

Representative Papers

Generating Attacks for LLMs with GFlowNets

Aug 10, 2026

Current red-teaming approaches for large language models (LLMs) rely heavily on manual efforts or static datasets, resulting in low efficiency and limited capacity to uncover deep-seated security vulnerabilities. This work proposes the first automatic and adaptive red-teaming framework based on Generative Flow Networks (GFlowNets), which leverages an attacker LLM to dynamically generate highly creative adversarial inputs. The framework autonomously identifies vulnerabilities in target models and quantifies their robustness without human intervention. By introducing GFlowNets into LLM red-teaming for the first time, the method outperforms existing benchmarks in English attack generation and pioneers support for automatic adversarial input generation in low-resource languages such as Turkish, substantially enhancing test coverage and evaluation efficiency.

0 citationsRead paper

A Comparative and Measurement-Based Study on Real-Time Network KPI Extraction Methods for 5G and Beyond Applications

Apr 22, 2025

This paper addresses the challenge of real-time acquisition and exchange of Key Performance Indicators (KPIs) across multi-vendor equipment in 5G and beyond networks. We propose a KPI extraction and exchange framework compatible with both standardized/commercial components and proprietary tools. Leveraging 3GPP-standard interfaces (e.g., N4, N6, N11), we conduct systematic empirical comparisons of three KPI collection techniques—active probing, passive traffic mirroring, and API polling—across latency, sampling granularity (down to millisecond-level), signaling load sensitivity, and deployment overhead. To our knowledge, this is the first cross-vendor, multi-dimensional empirical evaluation that quantifies performance boundaries and identifies precise applicability conditions for each method. The proposed framework enables on-demand KPI acquisition and protocol-level interoperability, providing telecom operators with reusable, evidence-based guidelines for intelligent network operations and closed-loop optimization.

0 citationsRead paper

Recent publications

Latest Papers

Generating Attacks for LLMs with GFlowNets

Aug 10, 2026

Current red-teaming approaches for large language models (LLMs) rely heavily on manual efforts or static datasets, resulting in low efficiency and limited capacity to uncover deep-seated security vulnerabilities. This work proposes the first automatic and adaptive red-teaming framework based on Generative Flow Networks (GFlowNets), which leverages an attacker LLM to dynamically generate highly creative adversarial inputs. The framework autonomously identifies vulnerabilities in target models and quantifies their robustness without human intervention. By introducing GFlowNets into LLM red-teaming for the first time, the method outperforms existing benchmarks in English attack generation and pioneers support for automatic adversarial input generation in low-resource languages such as Turkish, substantially enhancing test coverage and evaluation efficiency.

0 citationsRead paper

A Comparative and Measurement-Based Study on Real-Time Network KPI Extraction Methods for 5G and Beyond Applications

Apr 22, 2025

This paper addresses the challenge of real-time acquisition and exchange of Key Performance Indicators (KPIs) across multi-vendor equipment in 5G and beyond networks. We propose a KPI extraction and exchange framework compatible with both standardized/commercial components and proprietary tools. Leveraging 3GPP-standard interfaces (e.g., N4, N6, N11), we conduct systematic empirical comparisons of three KPI collection techniques—active probing, passive traffic mirroring, and API polling—across latency, sampling granularity (down to millisecond-level), signaling load sensitivity, and deployment overhead. To our knowledge, this is the first cross-vendor, multi-dimensional empirical evaluation that quantifies performance boundaries and identifies precise applicability conditions for each method. The proposed framework enables on-demand KPI acquisition and protocol-level interoperability, providing telecom operators with reusable, evidence-based guidelines for intelligent network operations and closed-loop optimization.

0 citationsRead paper