Your Agent Says Yes: Interpreting Adversarial Market Behavior Beyond Individual Transactions
研究使用十个角色条件语言模型代理在虚拟交易所中模拟市场行为，通过分析交易轨迹和策略事件，揭示了跨消息、代理、资产和时间的分布式市场行为模式。
0 citationsRead paper
Selected work
Representative Papers
Hallucination Mitigation with Agentic AI, Nested Learning, and AI Sustainability via Semantic Caching
This work addresses the susceptibility of large language models (LLMs) to generating and propagating hallucinations within multi-agent pipelines, which undermines system reliability. To mitigate this issue, the authors propose a three-stage asymmetric multi-agent architecture that integrates nested learning, continuous memory, and a novel semantic caching mechanism, orchestrated via the Open Floor protocol to enable extreme observability. Notably, this is the first application of semantic caching explicitly designed for hallucination mitigation, simultaneously enhancing factual accuracy, energy efficiency, and auditability. Experimental results demonstrate that the proposed approach reduces end-to-end hallucination scores by 31.3%–35.9%, achieves a semantic cache hit rate of 47.3%, and decreases LLM invocations by 52.7%, thereby substantially lowering computational energy consumption and carbon footprint.
0 citationsRead paper
MADP: A Multi-Agent Pipeline for Sustainable Document Processing with Human-in-the-Loop
This study addresses the long-standing reliance on manual labor in enterprise document processing, which suffers from low efficiency, high error rates, and excessive resource consumption. To overcome these limitations, the authors propose a multi-agent collaborative architecture that integrates a deep learning-based classifier, a document segmentation and parsing module, a large language model (LLM)-powered information extractor, and a validator. The system incorporates human-in-the-loop mechanisms and a novel prompt fine-tuning approach based on human feedback (PFTFI). Leveraging state-of-the-art LLM backends—including Granite-Docling, Mistral-Small, and DeepSeek-OCR—the framework achieves a 97.0% automation rate and 98.5% overall accuracy in processing 100,000 invoices annually. This reduces human labor requirements by 70% and significantly enhances sustainability, cutting carbon emissions and energy consumption by 69% each and water usage by 63%.
0 citationsRead paper
Prompt Injection Mitigation with Agentic AI, Nested Learning, and AI Sustainability via Semantic Caching
This work addresses the propagation and amplification of malicious instructions caused by prompt injection attacks in multi-agent systems by proposing a model-agnostic defense pipeline that integrates a nested learning architecture with a semantic caching mechanism to enable secure, efficient, and auditable mitigation. The study introduces the novel TIVS-O evaluation framework, which incorporates an observability scoring ratio to reveal a non-monotonic trade-off between security mitigation and audit transparency, and employs a five-dimensional metric suite to jointly optimize safety, performance, and sustainability. Experimental results demonstrate that the proposed approach eliminates high-risk vulnerabilities, reduces large language model invocations by 41.6%, and significantly lowers latency, energy consumption, and carbon emissions, thereby validating the feasibility of production-grade secure and green deployment.
0 citationsRead paper
Agentic AI Sustainability Assessment for Supply Chain Document Insights
This study addresses the challenge of balancing automation efficiency and environmental sustainability in supply chain document intelligence. We propose the first ESG-aligned sustainability evaluation framework for document processing. Methodologically, we design a multi-agent AI architecture incorporating “thinking-mode” reasoning, an adaptive document parser, a verifiable validation module, a human-in-the-loop benchmark, and a real-time energy consumption and carbon emission monitoring model—enabling joint quantitative assessment of performance and environmental metrics. Our key contribution lies in the deep integration of agentic AI with green AI evaluation, transcending conventional human-AI collaboration paradigms. Experimental results demonstrate that the fully autonomous agent-based workflow reduces energy consumption by 70–90%, carbon emissions by 90–97%, and water usage by 89–98% compared to manual processes—robustly validating the technical feasibility and governance value of sustainable intelligent document processing.
0 citationsRead paper
Recent publications
Latest Papers
Your Agent Says Yes: Interpreting Adversarial Market Behavior Beyond Individual Transactions
研究使用十个角色条件语言模型代理在虚拟交易所中模拟市场行为，通过分析交易轨迹和策略事件，揭示了跨消息、代理、资产和时间的分布式市场行为模式。
0 citationsRead paper
Hallucination Mitigation with Agentic AI, Nested Learning, and AI Sustainability via Semantic Caching
This work addresses the susceptibility of large language models (LLMs) to generating and propagating hallucinations within multi-agent pipelines, which undermines system reliability. To mitigate this issue, the authors propose a three-stage asymmetric multi-agent architecture that integrates nested learning, continuous memory, and a novel semantic caching mechanism, orchestrated via the Open Floor protocol to enable extreme observability. Notably, this is the first application of semantic caching explicitly designed for hallucination mitigation, simultaneously enhancing factual accuracy, energy efficiency, and auditability. Experimental results demonstrate that the proposed approach reduces end-to-end hallucination scores by 31.3%–35.9%, achieves a semantic cache hit rate of 47.3%, and decreases LLM invocations by 52.7%, thereby substantially lowering computational energy consumption and carbon footprint.
0 citationsRead paper
MADP: A Multi-Agent Pipeline for Sustainable Document Processing with Human-in-the-Loop
This study addresses the long-standing reliance on manual labor in enterprise document processing, which suffers from low efficiency, high error rates, and excessive resource consumption. To overcome these limitations, the authors propose a multi-agent collaborative architecture that integrates a deep learning-based classifier, a document segmentation and parsing module, a large language model (LLM)-powered information extractor, and a validator. The system incorporates human-in-the-loop mechanisms and a novel prompt fine-tuning approach based on human feedback (PFTFI). Leveraging state-of-the-art LLM backends—including Granite-Docling, Mistral-Small, and DeepSeek-OCR—the framework achieves a 97.0% automation rate and 98.5% overall accuracy in processing 100,000 invoices annually. This reduces human labor requirements by 70% and significantly enhances sustainability, cutting carbon emissions and energy consumption by 69% each and water usage by 63%.
0 citationsRead paper
Prompt Injection Mitigation with Agentic AI, Nested Learning, and AI Sustainability via Semantic Caching
This work addresses the propagation and amplification of malicious instructions caused by prompt injection attacks in multi-agent systems by proposing a model-agnostic defense pipeline that integrates a nested learning architecture with a semantic caching mechanism to enable secure, efficient, and auditable mitigation. The study introduces the novel TIVS-O evaluation framework, which incorporates an observability scoring ratio to reveal a non-monotonic trade-off between security mitigation and audit transparency, and employs a five-dimensional metric suite to jointly optimize safety, performance, and sustainability. Experimental results demonstrate that the proposed approach eliminates high-risk vulnerabilities, reduces large language model invocations by 41.6%, and significantly lowers latency, energy consumption, and carbon emissions, thereby validating the feasibility of production-grade secure and green deployment.
0 citationsRead paper
Agentic AI Sustainability Assessment for Supply Chain Document Insights
This study addresses the challenge of balancing automation efficiency and environmental sustainability in supply chain document intelligence. We propose the first ESG-aligned sustainability evaluation framework for document processing. Methodologically, we design a multi-agent AI architecture incorporating “thinking-mode” reasoning, an adaptive document parser, a verifiable validation module, a human-in-the-loop benchmark, and a real-time energy consumption and carbon emission monitoring model—enabling joint quantitative assessment of performance and environmental metrics. Our key contribution lies in the deep integration of agentic AI with green AI evaluation, transcending conventional human-AI collaboration paradigms. Experimental results demonstrate that the fully autonomous agent-based workflow reduces energy consumption by 70–90%, carbon emissions by 90–97%, and water usage by 89–98% compared to manual processes—robustly validating the technical feasibility and governance value of sustainable intelligent document processing.
0 citationsRead paper