零知识机器学习(zkML)是链上AI的未来吗?(附优质项目介绍)

Odaily星球日报Published on 2024-05-16Last updated on 2024-05-16

Abstract

尽管zkML是最理想的链上AI/ML方法,但目前要比常规计算慢1000倍。

原文作者 | @DistilledCrypto

编译 | Golem

零知识机器学习(zkML)是链上AI的未来吗?(附优质项目介绍)

自 ChatGPT 等大语言模型火爆后,在去中心化网络上运行类似的机器学习模型也成为了区块链+AI 的主要叙事之一。但我们无法像信任 OpenAI 这样有信誉支撑的公司一样信任去中心化网络会使用特定的 ML 模型进行推理,因此我们需要进行验证。考虑到数据的隐私性,零知识机器学习(zkML)被普遍看好,那么它会是链上 AI 的未来吗?

Odaily 星球日报将在本文中简单介绍关于 zkML 的基础知识、值得关注的 zkML 项目,最后在简单说明 zkML 的局限性及代替方案。

关于 zkML 的基础知识

零知识机器学习 (zkML) 类似于一种计算中的保密方法。它主要涉及两个部分:

  • 使用机器学习(ML)执行任务;

  • 证明任务正确完成,但不透露所有细节。

简单来说,它的工作原理如下:

a. 运行任务

有人使用 ML 模型来处理一些数据并得到结果,这就像厨师按照食谱烤蛋糕但却不告诉任何人原料一样。

b. 证明任务

任务完成后,他们可以展示一个证明。例如,“我在这个特定的模型中使用了特定的输入,并得到了这个结果。”他们实际上在证明他们正确遵循了食谱上的步骤。

c. 保守秘密

zkML 的妙处在于,当他们证明任务正确完成时,他们可以保留一些细节,例如将输入的数据、模型的运作方式或结果保密。简而言之,zkML 可以让证明者说“相信我,我做对了”,同时仍然保持他们的方法和数据的私密性。

零知识机器学习(zkML)是链上AI的未来吗?(附优质项目介绍)

值得关注的 zkML 项目介绍

zkML 概念自提出到现在已有将近一年时间,目前已经有许多相关项目正在建设,其中少部分还在市场上发行了代币。Messari 列出了一些知名 VC 投资的 zkML 项目,下面将对它们进行介绍。

零知识机器学习(zkML)是链上AI的未来吗?(附优质项目介绍)

来源:Messari

Spectral

Spectral 正在为 Web3 构建链上代理经济。他们的旗舰产品 SYNTAX 是一种专有的 LLM(Large Language Model),可以生成 Solidity 代码。Spectral 能够用户使创建链上自主代理,同时利用去中心化的 ML 推断来改进智能合约。此外,利用 zkML,Spectral 能够提供证据表明特定的预测是由特定的 ML 模型生成的,确保了流程中的信任和真实性。

Spectral 已发币,代币为 SPEC,市值 1.19 亿美元。

Worldcoin

Worldcoin 正在开发一个开源系统,旨在让每个人都能参与全球经济。在 Worldcoin 中,zkML 的一个潜在用途是提高虹膜识别技术的安全性和隐私性。代币 WLD 市值目前为 10.7 亿美元。

它的工作原理如下:

a. 生物识别自托管

World ID 的用户可以将自己的生物特征数据(如虹膜扫描)安全加密地存储在他们的移动设备上。

b. 本地处理

然后,用户可以将 ML 模型下载到他们的设备中,以从虹膜扫描中生成唯一的代码。

c. 隐私保护证明

使用 zkML,他们可以直接在自己的设备上创建证明。这一证明证实了他们的虹膜代码是使用正确的模型通过扫描准确生成的。所有这些操作都是在不暴露用户实际数据的情况下进行的。

Risc Zero

RISC Zero 旨在增强互联网的信任和效率,这将通过提供无需各方互相信任的计算服务来实现。

以下是 RISC Zero 关注的重点: 

a. 扩展区块链

它使用 Bonsai 证明服务来执行复杂操作,从而增强区块链的安全性。Bonsai 在链下管理复杂的计算和隐私数据,从而提高效率。

b. 与 Spice AI 的合作

Spice AI提供可组合、即用型数据和 AI 基础设施,包括托管的云级 Spice.ai OSS。此次合作旨在为开发人员提供全面的 zkML 工具包。

c. 机器学习服务

开发人员可以使用 RISC Zero 来安全地访问和查询数据、私密训练 ML 模型及提供数据被正确处理的证明。

本质上,RISC Zero 为开发人员提供 MLaaS(ML as a service)服务,同时确保数据和执行过程保持私密和安全。

Giza

Giza 是一个在 Starknet 网络上运行的机器学习平台。

a. 主要目标

Giza 旨在直接在区块链上扩展 ML 操作。

b. 技术基础

其使用支持零知识(ZK)证明的 Starknet 来验证 ML 操作,确保计算的准确性和安全性,并且不会泄露基础数据。 

c. 应用

在 Starknet 上,Giza 启用“Giza Agents”来自动执行各种财务策略,包括跨协议收益聚合、资产配置、无风险做市。本质上,利用 zkML 的优势,Giza 允许在区块链上安全、自动地执行金融策略。

Vanna

Vanna 是模块化 AI 推理网络,不仅与 EVM 链兼容,而且提供灵活的安全性,用户可选择 zkML、optimistic ZK、opML,、teeML 等多种验证方式。结合 Vanna 未来的使用场景为使用 LLM 生成链上 GameFi 游戏对话;链上智能合约漏洞检测;针对 DeFi 协议的风险预警引擎;用于标记空投中的女巫账户信誉系统。

除了以上介绍的几个项目外,zkML 生态中还有如下图中的项目,因为篇幅原因就不再介绍了,供读者自行参考。

零知识机器学习(zkML)是链上AI的未来吗?(附优质项目介绍)

来源:SevenX Ventures

zkML 的局限性及代替方案

尽管在理论上能够吸引人,但 zkML 目前并不太实用。AI 计算本身就属于资源密集型,添加类似 zkML 中使用的加密方法会使它们变得更慢,Modulus Labs 报告称 zkML 可能比常规计算慢 1000 倍。实际上对于大多数用户来说,多等待几分钟在日常体验上都难以接受。

因此,由于这些限制,zkML 现在可能仅适用于非常小的 ML 模型。在这种情况下许多 AI 项目不得不考虑其他的验证方法。目前主要有两种替代方案:

  • opML(Optimistic ML)

  • teeML(Trusted Execution Environment ML)

下图简单说明了三者之间的区别:

零知识机器学习(zkML)是链上AI的未来吗?(附优质项目介绍)

来源: Marlin Protocol

Trending Cryptos

Related Reads

Interview with Strategy CEO: Can STRC Recover After Selling Bitcoin?

Interview with Strategy CEO Phong Le on the recent sale of 32 Bitcoin and its impact. He clarifies the move was a small, strategic action to demonstrate liquidity to debt holders, test internal processes, and prove operational discipline—not a response to fears of a "death spiral" from DeFi protocols leveraging STRC (Strategy's preferred stock product), which he notes holds less than 10% of STRC. Le emphasizes Strategy’s long-term focus as the largest corporate Bitcoin holder, using the adage that markets are a "voting machine" short-term but a "weighing machine" long-term. Decision-making is data-driven, involving the board, complex modeling, and multiple stakeholder considerations, moving beyond a founder-centric model. He outlines various capital options but stresses the strategic importance of "doing nothing" as a valid choice, citing resilience built during the 2022 bear market. Le expresses unwavering belief in Bitcoin's foundational value for global sovereignty and its future role in an AI-driven economy with trillions of autonomous agents. Addressing STRC's current price below its $100 face value, Le explains recent pressure was due to using dollar reserves for bond buybacks. He expects STRC to return to par as reserves are replenished and its semi-monthly dividend payments begin, noting the product is heavily over-collateralized. Finally, Le confirms the company sold Bitcoin the week prior to May 31st, as disclosed in an 8-K filing, leaving prediction market interpretations to others. The overarching philosophy remains "Spread Bitcoin with love," embracing all methods of gaining Bitcoin exposure.

marsbit10m ago

Interview with Strategy CEO: Can STRC Recover After Selling Bitcoin?

marsbit10m ago

IOSG Founder: Ethereum Doesn't Need Another Leap of Technical Faith, It Needs a Musk-style Compromise

Jocy, founder of IOSG Ventures, argues that Ethereum does not need renewed technological faith but a "Musk-like compromise." The recent formation of ETHLabs—funded by major ETH holders like BitMine and Lubin—highlights a market-driven move to fill a gap left by the Ethereum Foundation (EF), signaling a loss of confidence in its decentralized, hands-off approach. The core critique contrasts Vitalik Buterin's (V) idealistic, technology-first vision with Elon Musk's pragmatic, business-driven execution. The author asserts Ethereum's current shortage is not another technical roadmap but a clear, real-world application narrative and a leader willing to engage directly with commercial realities—like Musk. Internal issues are emphasized, citing EF's management problems and talent drain. While the new decentralized model with independent nodes like ETHLabs addresses the single foundation's limitations, it risks fragmentation without cohesive direction. True cohesion, the author suggests, must come from a shared, compelling narrative around ETH's value, not just from aligned financial interests. Independence claims for new entities are seen as aspirational, needing years of transparency to build trust. The ultimate threat is not competitors like Solana, but the broader shift of attention and talent toward AI. Ethereum has a limited window—12 to 18 months—to recapture focus by delivering tangible, real-world applications. The conclusion urges V to shift from abstract ideals to grounded, pragmatic leadership. The time for this crucial pivot is running out.

marsbit1h ago

IOSG Founder: Ethereum Doesn't Need Another Leap of Technical Faith, It Needs a Musk-style Compromise

marsbit1h ago

Google Starts Selling TPUs, Big Tech Aims to Produce "Low-Cost Tokens" with AI Chips

Google has begun selling its proprietary TPU chips and AI computing hardware directly to third-party data centers and clients, marking a strategic shift. Previously only accessible via cloud rentals, TPUs are specialized processors designed for the matrix and tensor operations central to AI models. By combining thousands into supercomputing clusters managed by CPUs, Google achieves high-efficiency AI processing. This move enables Google’s Gemini AI to offer competitive token pricing, challenging rivals like OpenAI. It also signals a broader industry trend where AI compute is becoming a commoditized resource like electricity. While NVIDIA remains dominant with its CUDA ecosystem and high-performance GPUs, the focus is shifting from raw power to cost efficiency and system integration. Google’s approach mirrors NVIDIA’s by selling an entire ecosystem—hardware, software, and data center expertise—rather than just chips. This threatens NVIDIA’s grip on the mid-range inference market, where lower-cost, efficient solutions are increasingly demanded. Similarly, cloud providers like Huawei Cloud and Alibaba Cloud in China are developing their own AI chip ecosystems (e.g., Ascend, Zhenwu), packaging chips, clusters, and tools into full-stack solutions. They aim to reduce token costs and capture market share through integrated systems. In summary, the AI infrastructure race is evolving from a competition for the strongest chips to a contest for the most efficient and cost-effective systems. Google’s TPU sales highlight this transition, emphasizing that future success lies in delivering affordable, scalable AI compute as a foundational service.

marsbit1h ago

Google Starts Selling TPUs, Big Tech Aims to Produce "Low-Cost Tokens" with AI Chips

marsbit1h ago

Trading

Spot
Futures

Hot Articles

Discussions

Welcome to the HTX Community. Here, you can stay informed about the latest platform developments and gain access to professional market insights. Users' opinions on the price of AI (AI) are presented below.

活动图片