Document-level event argument extraction (DEAE) is essential for knowledge acquisition, aiming to extract participants of events from documents. In the zero-shot setting, existing methods employ LLMs to generate synthetic data to address the challenge posed by the scarcity of annotated data. However, relying solely on Event-type-only prompts makes it difficult for the generated content to accurately capture the contextual and structural relationships of unseen events. Moreover, ensuring the reliability and usability of synthetic data remains a significant challenge due to the absence of quality evaluation mechanisms. To this end, we introduce a multi-agent collaboration framework for zero-shot document-level event argument extraction (ZS-DEAE), which simulates the human collaborative cognitive process of “Propose–Evaluate–Revise.” Specifically, the framework comprises a generation agent and an evaluation agent. The generation agent synthesizes data for unseen events by leveraging knowledge from seen events, while the evaluation agent extracts arguments from the synthetic data and assesses their semantic consistency with the context. The evaluation results are subsequently converted into reward signals, with event structure constraints incorporated into the reward design to enable iterative optimization of both agents via reinforcement learning. In three zero-shot scenarios constructed from the RAMS and WikiEvents datasets, our method achieves improvements both in data generation quality and argument extraction performance, while the generated data also effectively enhances the zero-shot performance of other DEAE models.

With the vigorous development and gradual maturity of machine learning (ML) technologies, the AI-assisted disease diagnosis and prediction ($\mathcal {AADP}$) system has been extensively studied and can be expected to be intensively deployed in the real world. However, as the scale of ML data increases exponentially, the training and application of ML models impose a great burden on resource-constrained terminals. Designing cloud/edge server-aided$\mathcal {AADP}$protocols is becoming a popular topic. Whereas, the sensitivity of ML data, the intellectual property of ML models, and the uncontrollability of servers bring great security challenges to this promising computing paradigm. In this article, we initialize a new four-party framework for the$\mathcal {AADP}$system which consists of users, third-party test institution, AI doctor, and cloud/edge server. With this framework, we design two high-efficiency and secure outsourcing$\mathcal {AADP}$protocols under two different security models. By comprehensively employing secure hash functions, Householder transformations, and random permutations, we realize the following design objectives: 1) user’s actual identification is invisible to the other parties; 2) user’s feature vector is blinded to the AI doctor and the server; 3) the ML model of the AI doctor is confidential to the server; 4) AI doctor can obtain decent computational savings compared with achieving the diagnosis task by itself; and 5) AI doctor can verify the server’s misbehaviors with a nonnegligible probability under the security model with a fully malicious server. We argue these claims with rigid theoretical proofs and corroborate them with extensive experimental analysis.