Selected work

Representative Papers

Multi-Agentic System Leveraging Open-Source LLMs to Mitigate Disinformation Threats

Jun 29, 2026

This study addresses the growing challenge of misinformation amplified by social media and AI technologies, where traditional manual fact-checking proves inadequate. The authors propose a novel multi-agent system that uniquely integrates consensus mechanisms from human annotations, cognitive and knowledge diversity, and hierarchical collaborative structures. Built upon open-source large language models—including LLaMA, Qwen, Kimi, Deepseek, and LLaMA-Nemotron—the framework enables automated detection and verification of false claims. Evaluated on English, Polish, Slovak, and Bulgarian datasets, the approach significantly outperforms monolithic models such as GPT-4 and GPT-3.5 across three key tasks: direct misinformation identification, filtering of claims requiring verification, and detection of verifiable factual statements. The system also demonstrates high transparency and reproducibility.

0 citationsRead paper

MultiCW: A Large-Scale Balanced Benchmark Dataset for Training Robust Check-Worthiness Detection Models

Feb 18, 2026

This study addresses the challenges of data imbalance and limited generalization in automated verifiable claim detection across multilingual, multidomain, and multi-style settings. To this end, the authors introduce MultiCW, a large-scale, strictly class-balanced benchmark dataset spanning 16 languages, 7 thematic domains, and 2 writing styles, comprising 123,722 training instances and 27,761 out-of-distribution test samples. The work presents the first systematic comparison between fine-tuned multilingual Transformer models and 15 zero-shot large language models. Results demonstrate that fine-tuned models significantly outperform zero-shot approaches on verifiability classification and exhibit strong cross-lingual, cross-domain, and cross-style generalization capabilities.

0 citationsRead paper

Fast algorithms for complex-valued discrete Fourier transform with separate real and imaginary inputs/outputs

Apr 09, 2025

This work addresses the need for independent post-processing of spectral coefficients across two channels in signal processing applications. We propose a novel real-imaginary separated complex discrete Fourier transform (DFT) algorithm, departing from conventional approaches that treat complex inputs as atomic entities. Our method establishes the first fully decoupled computational framework for real and imaginary components, formulated via vector-matrix representation and integrated with a divide-and-conquer strategy leveraging real-domain-optimized FFT structures. By eliminating redundant complex arithmetic, the algorithm retains the asymptotic O(N log N) complexity while significantly reducing memory access overhead and computational latency on hardware platforms—thereby enhancing throughput for dual-channel spectral processing. The core contribution lies in the physical separation of real and imaginary components at both input and output stages, coupled with complete decoupling of their computational paths. This enables an efficient new paradigm for resource-constrained or channel-isolated signal processing systems.

0 citationsRead paper

Recent publications

Latest Papers

Multi-Agentic System Leveraging Open-Source LLMs to Mitigate Disinformation Threats

Jun 29, 2026

This study addresses the growing challenge of misinformation amplified by social media and AI technologies, where traditional manual fact-checking proves inadequate. The authors propose a novel multi-agent system that uniquely integrates consensus mechanisms from human annotations, cognitive and knowledge diversity, and hierarchical collaborative structures. Built upon open-source large language models—including LLaMA, Qwen, Kimi, Deepseek, and LLaMA-Nemotron—the framework enables automated detection and verification of false claims. Evaluated on English, Polish, Slovak, and Bulgarian datasets, the approach significantly outperforms monolithic models such as GPT-4 and GPT-3.5 across three key tasks: direct misinformation identification, filtering of claims requiring verification, and detection of verifiable factual statements. The system also demonstrates high transparency and reproducibility.

0 citationsRead paper

MultiCW: A Large-Scale Balanced Benchmark Dataset for Training Robust Check-Worthiness Detection Models

Feb 18, 2026

This study addresses the challenges of data imbalance and limited generalization in automated verifiable claim detection across multilingual, multidomain, and multi-style settings. To this end, the authors introduce MultiCW, a large-scale, strictly class-balanced benchmark dataset spanning 16 languages, 7 thematic domains, and 2 writing styles, comprising 123,722 training instances and 27,761 out-of-distribution test samples. The work presents the first systematic comparison between fine-tuned multilingual Transformer models and 15 zero-shot large language models. Results demonstrate that fine-tuned models significantly outperform zero-shot approaches on verifiability classification and exhibit strong cross-lingual, cross-domain, and cross-style generalization capabilities.

0 citationsRead paper

Fast algorithms for complex-valued discrete Fourier transform with separate real and imaginary inputs/outputs

Apr 09, 2025

This work addresses the need for independent post-processing of spectral coefficients across two channels in signal processing applications. We propose a novel real-imaginary separated complex discrete Fourier transform (DFT) algorithm, departing from conventional approaches that treat complex inputs as atomic entities. Our method establishes the first fully decoupled computational framework for real and imaginary components, formulated via vector-matrix representation and integrated with a divide-and-conquer strategy leveraging real-domain-optimized FFT structures. By eliminating redundant complex arithmetic, the algorithm retains the asymptotic O(N log N) complexity while significantly reducing memory access overhead and computational latency on hardware platforms—thereby enhancing throughput for dual-channel spectral processing. The core contribution lies in the physical separation of real and imaginary components at both input and output stages, coupled with complete decoupling of their computational paths. This enables an efficient new paradigm for resource-constrained or channel-isolated signal processing systems.

0 citationsRead paper