毕业生知识库

区块链和智能合约是构建数字社会的关键基础设施。它们有望解决传统中心化机构普遍存在的数据存储不安全、协作成本高昂和协作效率不足等问题，因此具有重要的研究与应用价值。然而，现阶段区块链受“不可能三角”问题的困扰尚存在性能瓶颈等计算限制，而且当前的智能合约智能化程度不足。这些问题严重制约了区块链的应用拓展，已成为区块链发展亟需解决的核心问题。为此，人工智能和区块链的的交叉研究领域——区块链智能应运而生。

区块链智能旨在实现人工智能和区块链的相互增益。区块链可以为分布式人工智能协作（多方之间数据、模型和计算资源的共享与协作计算）提供可信的基础设施，从而缓解中心化人工智能架构的资源（数据、模型和计算资源）限制；而人工智能可以为区块链应对“不可能三角”问题提取有价值的信息。更重要的是，人工智能有望提高智能合约的智能化程度，促进实现人工智能驱动型区块链智能。即，区块链和智能合约能够自主地完成用户请求的人工智能任务。目前，如何提高智能合约的智能化程度和如何实现人工智能驱动型区块链智能都是领域内亟待解决的问题。因此，本文围绕此问题展开研究，旨在提供一种有效的方法来提高智能合约的智能化程度，实现人工智能驱动型区块链智能。本文的主要工作和创新点归纳如下：

（1）提出了区块链智能驱动的智能合约构建与配置方法。智能合约是使区块链具备可编程性的核心技术，其发展可以直接推动区块链的应用拓展。因此，本文在区块链智能的背景下，结合人工智能技术，重新设计了现有的智能合约，提出了“区块链智能驱动的智能合约（Blockchain Intelligence-driven Smart Contracts，BISCs）”。BISCs是一类专为实现基于区块链的人工智能任务设计的智能合约。本文重点研究了BISCs的构建方法、特点和面向人工智能驱动型区块链智能的配置方法。具体而言，本文首先提出了两种构建BISCs的方法，即封装人工智能模型的模板化智能合约和调控人工智能协作的模板化智能合约。然后，本文从理论和实验的角度分析了它们的优势和挑战，为今后BISCs构建方法的选择提供了参考。最后，本文提出了动态配置BISCs以应对不同人工智能任务的方法，为结合智能合约与人工智能实现人工智能驱动型区块链智能提供了思路。本文基于以太坊智能合约验证和分析了本文提出的BISCs构建和配置方法，实验结果表明了提出方法的有效性。

（2）提出了基于区块链智能驱动智能合约的去中心化人工智能协作方法——学习市场。创新点（1）的研究表明，调控人工智能协作的BISCs具有广泛的应用潜力。因此，本文进一步聚焦于研究调控人工智能协作的BISCs。为了缓解现阶段中心化人工智能架构面临的资源限制，本文面向多方参与的去中心化人工智能协作场景，研究了BISCs的构建与应用方法，提出了BISCs调控的学习市场。该市场利用区块链为互不信任参与者间的协作计算和模型共享提供了可信环境，并且利用BISCs作为软件代理来调控可扩展的协作流程和市场机制。本文构建了权限管理、数据传输、模型验证、贡献评估、激励量化和模型交易六种BISCs。在这些BISCs的调控下，互不信任的参与者不仅能够在动态量化激励下实现分布式协作计算，而且能够在具有可追溯性和经济激励的市场中交易可信模型。本文基于以太坊和星际文件系统实现了学习市场并评估了其有效性。实验结果表明，学习市场可在不损失原有模型性能的前提下促进分布式计算资源协作，其协作成本与现有以太坊智能合约相当，并且在去中心化和可扩展性等方面具有优势。

（3）提出了基于区块链智能驱动智能合约的混合人工智能协作方法，实现了COVID-19联合预警。除了去中心化组织参与的人工智能协作外，联盟组织参与的人工智能协作是另一个重要的人工智能协作场景。本文首先研究面向联盟组织人工智能协作的BISCs构建方法，然后将其与学习市场BISCs结合，共同应用于混合组织（联盟组织和去中心化组织）参与的人工智能协作场景——COVID-19联合预警。为了缓解中心化传染病预警系统面临的监测资源（数据、模型和计算资源）限制，本文提出了一种BISCs调控的联合预警方法。该方法利用区块链为联合预警参与者提供了安全隐私的协作环境，利用BISCs作为软件代理调控可扩展的协作预警规则和激励机制，利用人工智能技术实现了多种传染病监测模型。本文提出的联合预警方法支持两种传染病监测模式：基于联邦学习的医疗端联盟组织监测和基于学习市场的社会端去中心化组织监测。本文构建了权限管理、医疗联邦监测、社会协作监测、医疗端与社会端融合预警和激励量化五种BISCs。在这些BISCs的调控下，医疗端联盟组织可在维护数据隐私的同时协作训练性能更优的联邦监测模型，而社会端去中心化组织可在学习市场中协作计算和共享经验证的监测方案。BISCs自动融合两端监测结果形成预警决策，最终提高预警性能和丰富预警依据。本文基于以太坊和星际文件系统实现和评估了提出的联合预警方法。实验表明，本文方法具有去中心化决策、量化激励、可追溯性和可扩展性等优点。

本文对BISCs的研究使得智能合约不再仅限于实现简单的控制逻辑，而可以应对更丰富的人工智能任务。基于BISCs，区块链的应用可以被有效地拓展到金融之外的其他社会领域。

Blockchain and smart contracts are key infrastructures for building the digital society. They are expected to address common issues found in traditional centralized institutions such as insecure data storage, high collaboration costs, and insufficient collaboration efficiency. Therefore, they have significant research and application value. However, at present, blockchain technology is plagued by the "blockchain trilemma" problem and still faces computational limitations such as performance bottlenecks. Additionally, current smart contracts are not intelligent enough. These problems have seriously hindered the expansion of blockchain applications, and have become the core problems that urgently need to be solved in the development of blockchain technology. For this reason, the cross-disciplinary research of Artificial Intelligence (AI) and blockchain -- blockchain intelligence, has emerged.

Blockchain intelligence aims to achieve mutual benefits between AI and blockchain. Blockchain can provide a trustworthy infrastructure for distributed AI collaboration (the sharing and collaborative computing of data, models, and computing resources among multiple parties), thereby alleviating the resource (data, models, and computing resources) limitations of centralized AI architectures. In turn, AI can extract valuable information for blockchain to tackle the "blockchain trilemma" problem. More importantly, AI is expected to improve the intelligence level of smart contracts, promoting the realization of AI-driven blockchain intelligence. That is, blockchain and smart contracts can autonomously complete AI tasks requested by users. Currently, how to improve the intelligence level of smart contracts and how to achieve AI-driven blockchain intelligence are urgent problems to be solved in the field. Therefore, this dissertation focuses on this problem, aiming to provide an effective method to improve the intelligence level of smart contracts and facilitate AI-driven blockchain intelligence. The main contributions and innovations of this dissertation are summarized as follows:

(1) A method for constructing and configuring Blockchain Intelligence-driven Smart Contracts (BISCs) is proposed. Smart contracts are the core technology that enables blockchain programmability, and their development can directly promote the expansion of blockchain applications. Therefore, this dissertation proposes BISCs by redesigning the existing smart contracts in the context of blockchain intelligence and combining AI technology. BISCs are a type of smart contracts designed for blockchain-based AI tasks. This dissertation focuses on the construction and characteristics of BISCs, as well as their configuration for AI-driven blockchain intelligence. Specifically, two methods for constructing BISCs are proposed in this dissertation, namely, templated smart contracts that encapsulate AI models and templated smart contracts that regulate AI collaboration. Then, this dissertation analyzes their advantages and challenges from theoretical and experimental perspectives, providing references for the selection of BISCs' construction methods in the future. Finally, a method for dynamically configuring BISCs to cope with different AI tasks is proposed, providing ideas for combining smart contracts and AI to achieve AI-driven blockchain intelligence. This dissertation validates and analyzes the proposed construction and configuration methods of BISCs based on Ethereum smart contracts, and the experimental results demonstrate their effectiveness.

(2) A decentralized AI collaboration method based on BISCs is proposed, namely Learning Market. The study in (1) shows that BISCs for regulating AI collaboration have broad application potential. Therefore, this dissertation further focuses on BISCs that regulate AI collaboration. In order to alleviate the resource limitations faced by the current centralized AI architecture, this dissertation investigates the construction and application of BISCs for decentralized AI collaboration involving multiple participants, and proposes the Learning Market regulated by BISCs. This market uses blockchain to provide a trustworthy environment for collaborative computing and model sharing among untrusted participants, and uses BISCs as software agents to regulate scalable collaboration processes and market mechanisms. This dissertation constructs six BISCs: templated authority management smart contract, templated data transmission smart contract, templated model validation smart contract, templated contribution evaluation smart contract, templated incentive quantification smart contract, and templated model trading smart contract. Under the regulation of these BISCs, untrusted participants can not only achieve distributed collaborative computing under dynamic quantified incentives, but also trade trustworthy models in a market with traceability and economic incentives. This dissertation implements and evaluates the Learning Market based on Ethereum and InterPlanetary File System. Experimental results show that the Learning Market can promote collaboration of distributed computing resources without sacrificing the performance of the original model. Its collaboration costs are comparable to that of existing Ethereum smart contracts, and it has advantages in terms of decentralization and scalability.

(3) A hybrid AI collaboration method based on BISCs is proposed to achieve collaborative COVID-19 early warning. In addition to AI collaboration involving decentralized organizations, AI collaboration involving federated organizations is another important AI collaboration scenario. This dissertation first studies the construction of BISCs for AI collaboration involving federated organizations, and then combines them with Learning Market's BISCs to apply in an AI collaboration scenario involving hybrid organizations (federated organizations and decentralized organizations), namely collaborative COVID-19 early warning. In order to alleviate the limitations of surveillance resource (data, models and computing resources) faced by centralized infectious disease warning systems, this dissertation proposes a collaborative early warning method regulated by BISCs. This method uses blockchain to provide a secure and private collaboration environment for participants, uses BISCs as software agents to regulate scalable early warning rules and incentive mechanisms, and uses AI technology to implement various surveillance models. The proposed method supports two surveillance modes: medical federation surveillance based on federated learning and social collaboration surveillance based on Learning Market. This dissertation constructs five BISCs: templated authority management smart contract, templated medical federated surveillance smart contract, templated social collaboration surveillance smart contract, templated medical & social fusion warning smart contract, and templated incentive quantification smart contract. Under the regulation of these BISCs, the medical federation can collaborate to train better-performing federated surveillance models while maintaining data privacy, and the social collaboration can collaborate in the Learning Market to train and share verified surveillance solutions. BISCs automatically fuse surveillance results from both ends to form warning decisions, ultimately improving early warning performance and enriching early warning basis. This dissertation implements and evaluates the proposed method based on Ethereum and InterPlanetary File System. The experiments show that the proposed method has advantages such as decentralized decision-making, quantitative incentives, traceability, and scalability.

The BISCs studied in this dissertation expand the scope of smart contracts beyond simple control logic, and enable them to cope with rich AI tasks. Based on BISCs, the applications of blockchain can be effectively expanded to other social domains beyond finance.