0 views collected around this technical thread.
The paper introduces Nimbus, a two‑party privacy‑preserving inference framework for Transformer models that leverages a client‑side outer‑product linear‑layer protocol and distribution‑aware polynomial approximations for non‑linear layers, achieving up to five‑fold speedups with negligible accuracy loss.
From August 26‑30, 2024, Ant Group showcased eight papers—including seven oral presentations—at the VLDB 2024 conference in Guangzhou, covering advances in knowledge‑graph warehouses, graph analytics, joint clustering and imputation, secure multi‑party queries, distributed logging, deep recommendation model training, cloud autoscaling, and LLM‑driven database interaction.
Alibaba Mama’s Secure Data Hub delivers a privacy‑enhanced clean‑room for advertising by combining multi‑party computation, federated learning and differential privacy with encrypted operators on a Flink engine, offering cloud‑agnostic, scalable deployment that enables cross‑domain analytics while protecting raw user data and boosting ROI.
This article presents a detailed case study of how Jiangsu Bank and Ningbo Bank leveraged Ant Group’s multi‑party secure computation platform and the “YinYu” privacy‑computing framework to build joint risk‑control models, enhancing data sharing, security, and approval rates for inclusive finance.
This article presents WeBank's exploration of privacy‑computing technologies combined with blockchain to enable secure, compliant data flow across enterprises, detailing the regulatory background, technical architectures, key use‑case scenarios such as anonymous query, privacy intersection, joint prediction and statistics, and real‑world deployments including the 2022 Big Data “Star River” benchmark cases.
Federated learning enables multiple organizations to jointly train high‑performing AI models without sharing raw data, using techniques such as secure multi‑party computation, differential privacy, and homomorphic encryption, thereby overcoming data‑island and regulatory constraints and supporting applications in mobile edge AI, finance, retail, and healthcare.
At the 32nd USENIX Security Symposium in Anaheim, Ant Group’s Research Institute sponsored the event and showcased two first‑author papers—one introducing the Squirrel framework for fast, secure two‑party computation of Gradient Boosting Decision Trees, and another proposing an efficient 3‑party protocol for binary circuits in maliciously‑secure DNN inference.
The SS‑LR open protocol describes a secret‑sharing based logistic regression algorithm split into four layers—machine learning, secure operators, cryptographic protocol, and network transmission—enabling interoperable, privacy‑preserving data flow and secure multi‑party model training across institutions.
The YinYu 1.0 release introduces an MVP deployment package for privacy‑computing beginners, expands the product, resource, and connectivity layers, adds the Kuscia orchestration framework, enhances algorithm, device, and cryptographic capabilities, and promotes full‑stack interoperability across platforms.
SCQL, an open‑source Secure Collaborative Query Language built on multi‑party computation, enables SQL‑style privacy‑preserving data analysis for small‑to‑medium organizations by offering easy integration, fine‑grained column‑level access control, broad data‑source support, and optimized performance for collaborative queries.
Alibaba’s Secure Data Hub (SDH) is a privacy‑preserving data clean‑room platform that uses secure multi‑party computation and privacy‑enhancing machine learning to let advertisers, ad platforms, and auditors jointly analyze marketing data via a simple SQL API while keeping raw data encrypted, column‑level protected, and confined to each party’s private domain.
The interview with Alibaba Cloud Datatrust architect Liang Aiping reveals that privacy computing is still in its early stage, facing technical challenges in data sources, algorithm theory‑engineering gaps, system management interoperability, and product trade‑offs, while outlining future trends toward cross‑platform interoperability and distributed computing.
Alibaba Mama’s Secure Data Hub, a privacy‑preserving data clean‑room platform that uses multi‑party computation and federated learning for secure advertising analytics, successfully passed the China Academy of Information and Communications Technology’s Trusted Privacy Computing evaluation, confirming its strong security, compliance and industry‑recognized capabilities.
This article explains Ant Group's privacy computing framework, covering its fundamental concepts, layered product architecture, and four concrete use‑cases in the medical sector—including insurance, hospitals, health commissions, and medical device manufacturers—demonstrating how secure multi‑party computation and federated learning enable data collaboration while preserving privacy.
This article examines the growing complexity of financial risk scenarios, outlines global privacy regulations, describes Ant Group's three‑layer multi‑party secure risk‑control architecture, showcases anti‑fraud and external client use cases, and discusses current challenges and future directions for secure, privacy‑preserving risk management.
Tencent’s Angel PowerFL team clinched the iDASH homomorphic encryption champion and secured top spots in MPC and SGX tracks, showcasing innovative privacy‑preserving machine‑learning models, CKKS‑based encrypted inference, and a scalable SGX clustering solution that push the boundaries of secure computation.
This article outlines Guangfa Group’s initiatives in privacy computing and federated learning, detailing the development of its federated learning platform, contributions to open‑source FATE, industry standards, various application scenarios such as joint statistics, precise marketing, risk control, cross‑domain verification, and introduces their newly published book on federated learning principles and applications.
The article introduces TECC, a privacy‑computing framework that balances security and performance by using trusted execution environments, data secret‑sharing, lightweight cryptographic protocols, and Rust‑based implementation to enable near‑plaintext speed for secure multi‑party machine learning and data analysis.
This article provides a comprehensive overview of privacy computing—covering its background, key techniques such as MPC, TEE, federated learning, homomorphic encryption, and differential privacy—and examines how these technologies are applied in open financial ecosystems, including use cases, challenges, and future directions.
This article reviews the rapid development of privacy-preserving computation, explains its classification, discusses differential privacy, secure multi‑party computation, federated and split learning, and demonstrates how these techniques can be combined for graph learning and recommendation systems, culminating in a description of the JinZhiTa privacy‑computing platform.