Google将联邦学习迁移至TEE,Gboard实现可验证差分隐私
Google Research Moves Federated Learning Into TEEs: Gboard Now Trains With Externally Verifiable Differential Privacy
联邦学习落地生产环境的重大突破,首次实现外部可验证的差分隐私,对隐私计算从业者极具参考价值。
Google Research has announced a next-generation Federated Learning (FL) system built on Trusted Execution Environments (TEEs). The research team claims externally verifiable central differential privacy (DP) guarantees for FL for the first time.
Google Research 宣布推出一个基于可信执行环境(TEE)的下一代联邦学习(FL)系统。研究团队首次声称 FL 具备可外部验证的中心差分隐私(DP)保证。
What Problem Does TEE-Based Federated Learning Solve?
基于 TEE 的联邦学习解决了什么问题?
Google introduced Federated Learning FL in 2017. It powers next-word prediction and Smart Compose on Gboard, reply suggestions in Google Messages, and Smart Text Selection in Android.
Google 于 2017 年推出了联邦学习 FL。它支撑了 Gboard 上的下一个词预测和智能撰写、Google Messages 中的回复建议以及 Android 上的智能文本选择功能。
Earlier systems had a trust gap. Devices uploaded data for immediate aggregation, but outsiders could not verify that data was never logged or inspected. Secure Aggregation added cryptographic protection. However, it was not compatible with state-of-the-art central DP algorithms like matrix factorization DP-FTRL. Google also had to be trusted to add DP noise correctly.
早期的系统存在信任缺口。设备上传数据用于即时聚合,但外部人员无法验证数据是否从未被记录或检查。安全聚合增加了密码学保护。然而,它与最先进的中心 DP 算法(如矩阵分解 DP-FTRL)不兼容。Google 还必须被信任以正确添加 DP 噪声。
The new design moves client gradient computation to the server. It then makes that server logic attestable, so the operator no longer needs to be trusted.
新设计将客户端梯度计算移至服务器。随后使该服务器逻辑可证明,因此不再需要信任运营商。
How Does the System Work?
该系统如何工作?
The system builds on Google’s earlier confidential federated analytics work. It coordinates 4 core components:
该系统建立在 Google 早期的机密联邦分析工作之上。它协调四个核心组件:
- Data upload: Devices encrypt training examples locally and pre-authorize an access policy. The policy lists which TEE computations may process the data. Policies must appear in a public transparency log.
- KMS and policy verification: A Key Management System, built from TEEs running the RAFT consensus protocol, releases keys only to workloads matching the policy.
- Workload execution: A root TEE runs a Python training loop and delegates subtasks to worker TEEs. Orchestration uses Federated Language, derived from TensorFlow Federated. Only DP model weights are released.
- Fault-tolerant recovery: Each round saves a KMS-encrypted recovery state for handling root or worker failures.
- 数据上传:设备在本地加密训练样本并预先授权访问策略。该策略列出了哪些 TEE 计算可以处理数据。策略必须出现在公共透明度日志中。
- KMS 和策略验证:一个由运行 RAFT 共识协议的 TEE 构建的密钥管理系统,仅向符合策略的工作负载发布密钥。
- 工作负载执行:根 TEE 运行 Python 训练循环并将子任务委派给工作 TEE。编排使用源自 TensorFlow Federated 的联邦语言。仅发布 DP 模型权重。
- 容错恢复:每轮保存一个经 KMS 加密的恢复状态,以处理根节点或工作节点故障。
Why Is the Privacy Guarantee Verifiable?
为什么隐私保证是可验证的?
Access policies are published to Rekor, Sigstore’s public transparency log. External auditors can track every server workload a device could feed. The KMS and data processing binaries are reproducibly buildable from open source code.
访问策略发布到 Rekor,即 Sigstore 的公共透明度日志。外部审计人员可以追踪设备可能馈送的每个服务器工作负载。KMS 和数据处理的二进制文件可以从开源代码中可复现地构建。
The policies directly describe the Python training program. To protect proprietary model architectures, TEEs support sideloading serialized logic at runtime. All privacy-relevant logic must stay hardcoded in the attested program. Workload operators see only metrics and DP model weights. Encrypted data can be decrypted only for a limited time after upload.
策略直接描述 Python 训练程序。为了保护专有模型架构,TEE 支持在运行时侧载序列化逻辑。所有与隐私相关的逻辑必须硬编码在已证明的程序中。工作负载操作员只能看到指标和 DP 模型权重。加密的数据仅在上传后有限时间内可解密。
What Did Gboard Gain?
Gboard 获得了什么好处?
Gboard used the system to launch English and Japanese next-word prediction models with stronger privacy guarantees and improved accuracy. Two design choices drive this:
Gboard 利用该系统推出了具有更强隐私保障和更高准确率的英文及日文下词预测模型。这一成果由两项设计选择驱动:
- First, all uploads are collected before server-side training runs. Diurnal swings in device availability no longer slow training. The program can compute an optimal participation schedule and tune DP parameters. Google’s privacy-utility curves come from training an English model for 5000 rounds with cohorts of 6500 devices on both systems.
- Second, the bottleneck moved to the server. Previous FL models took 1 to 2 months each to train. Training now parallelizes across machines, limited only by TEE resource availability. Google reports substantially faster compute times but does not publish a single speedup figure.
- 首先,所有上传数据均在服务器端训练运行前收集。设备可用性的昼夜波动不再拖慢训练进度。该程序能够计算最优的参与时间表并调整差分隐私(DP)参数。Google 的隐私-效用曲线来源于在两种系统上分别使用 6500 个设备的群体进行 5000 轮英文模型训练的结果。
- 其次,瓶颈转移到了服务器端。之前的联邦学习(FL)模型每个需要 1 到 2 个月才能完成训练。现在的训练可以在多台机器上并行进行,仅受限于可信执行环境(TEE)资源的可用性。Google 报告称计算时间大幅缩短,但未公布具体的加速倍数。
Interactive Explainer: Inside the TEE-Based FL Pipeline
交互式解释器:基于 TEE 的 FL 管道内部
How Google’s TEE-based Federated Learning works
Google 基于 TEE 的联邦学习工作原理
Interactive explainer based on Google Research’s Oct 2, 2026 post and paper (arXiv:2609.31494).
基于 Google Research 2026 年 10 月 2 日的帖子和论文(arXiv:2609.31494)构建的交互式解释器。
1. Data pipeline 2. Who can see what 3. Tamper test
1. 数据管道 2. 谁能看到什么 3. 篡改测试
Phonesencrypt locally+ access policy KMS (TEEs)RAFT clusterchecks policy Root TEEPython training+ worker TEEs AnalystDP modelweights only KMS-encrypted recovery state
手机本地加密 + 访问策略 KMS(TEE)RAFT 集群检查策略 Root TEE Python 训练 + 工作节点 TEEs 分析师 DP 模型权重仅 KMS 加密恢复状态
Step 1Data upload Step 2KMS + policy Step 3Workload execution Step 4Fault recovery
步骤 1 数据上传 步骤 2 KMS + 策略 步骤 3 工作负载执行 步骤 4 故障恢复
▶ Auto-playNext step
▶ 自动播放下一步
Earlier FLTEE-based FL
早期的 FL TEE 支持的 FL
Pick the server workload that asks the KMS for decryption keys. Only code listed in the published access policy gets them.
选择向 KMS 请求解密密钥的工作负载。只有列入已发布访问策略中的代码才能获得密钥。
Approved training programhash = matches policy in Rekor log
已批准的训练程序 hash = 与 Rekor 日志中的策略匹配
Modified program (logs raw data)hash = not in access policy
修改后的程序(记录原始数据)hash = 不在访问策略中
Request decryption key
请求解密密钥
🔒
Waiting for a request. The KMS verifies the TEE’s remote attestation against the access policy.
等待请求。KMS 根据访问策略验证 TEE 的远程证明。
Sources: research.google blog, arXiv:2609.31494Built by Marktechpost
来源:research.google 博客,arXiv:2609.31494 由 Marktechpost 构建
How Does It Compare With Other FL Frameworks?
它与其他 FL 框架相比如何?
| Feature | Google TEE-based FL | NVIDIA FLARE | Flower | Apple pfl-research |
|---|---|---|---|---|
| Primary use | Production cross-device training (live in Gboard) | Production FL SDK with Docker, Kubernetes and cloud tooling | Framework for building federated AI systems | Simulation only; not intended for third-party deployments |
| Where client updates are computed | Server-side TEEs | At each participating site | On clients | Simulated |
| Hardware TEE support | Yes, built on Project Oak | Yes: AMD SEV-SNP, Intel TDX, NVIDIA GPU confidential computing | Not part of the core framework | No |
| Differential privacy | Central DP, externally verifiable | DP filters, DP-SGD via Opacus | Central and local DP | Local and central DP mechanisms |
| Other privacy tech | KMS-gated decryption; Willow secure aggregation container | Homomorphic encryption, private set intersection | SecAgg and SecAgg+ | Not applicable (simulation) |
| Public transparency log for server code | Yes, Sigstore Rekor | Not documented | Not documented | Not applicable |
| Reproducible TEE builds | Yes (KMS and data processing binaries) | Not documented | Not applicable | Not applicable |
| License | Apache 2.0 | Apache 2.0 | Apache 2.0 | Apache 2.0 |
| 特性 | 基于 Google TEE 的联邦学习 (FL) | NVIDIA FLARE | Flower | Apple pfl-research |
|---|---|---|---|---|
| 主要用途 | 生产级跨设备训练(Gboard 已上线) | 配备 Docker、Kubernetes 和云工具的生产级 FL SDK | 构建联邦 AI 系统的框架 | 仅用于模拟;不打算用于第三方部署 |
| 客户端更新计算位置 | 服务器端 TEE | 在每个参与站点 | 在客户端 | 模拟 |
| 硬件 TEE 支持 | 是,基于 Project Oak 构建 | 是:AMD SEV-SNP、Intel TDX、NVIDIA GPU 机密计算 | 不是核心框架的一部分 | 无 |
| 差分隐私 | 中心式 DP,可外部验证 | 通过 Opacus 实现 DP 过滤器、DP-SGD | 中心式和本地 DP | 本地和中心式 DP 机制 |
| 其他隐私技术 | KMS 门控解密;Willow 安全聚合容器 | 同态加密、私有集合求交 | SecAgg 和 SecAgg+ | 不适用(模拟) |
| 服务器代码的公开透明日志 | 是,Sigstore Rekor | 未记录 | 未记录 | 不适用 |
| 可复现的 TEE 构建 | 是(KMS 和数据处理的二进制文件) | 未记录 | 不适用 | 不适用 |
| 许可证 | Apache 2.0 | Apache 2.0 | Apache 2.0 | Apache 2.0 |
Sources: Google Research blog, Confidential Federated Compute repo, NVIDIA FLARE docs, FLARE attestation guide, Flower 1.8 release notes, pfl-research repo. Verified October 4, 2026.
来源:Google Research 博客、Confidential Federated Compute 仓库、NVIDIA FLARE 文档、FLARE 证明指南、Flower 1.8 发行说明、pfl-research 仓库。于 2026 年 10 月 4 日核实。
Key Takeaways
关键要点
- Google moved FL gradient computation from phones into attested server-side TEEs.
- Central DP guarantees are now externally verifiable via Rekor and reproducible builds.
- Gboard ships English and Japanese next-word models on the new system.
- Training once took 1 to 2 months per model; TEE capacity is now the limit.
- Core TEE binaries and Federated Language are open source under Apache 2.0.
- Google 将 FL 梯度计算从手机移至经过证明的服务器端 TEE 中。
- 中心式 DP 保证现在可通过 Rekor 和可复现构建进行外部验证。
- Gboard 在新系统上推出了英语和日语的下一个词预测模型。
- 以前每个模型的训练需要 1 到 2 个月;现在 TEE 容量成为限制因素。
- 核心 TEE 二进制文件和 Federated Language 以 Apache 2.0 许可开源。
Check out the Paper, Technical details and GitHub Repo. All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.
查看论文、技术细节和 GitHub 仓库。所有功劳归于该项目的研究人员。此外,欢迎在 Twitter 上关注我们,别忘了加入我们有 15 万+成员的 ML SubReddit 并订阅我们的新闻通讯。等等!你在 Telegram 上吗?现在你也可以在 Telegram 上加入我们。
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力