Rondo and BumbleBee: Scalable Randomness Beacon and Secure Two‑Party Inference for Large Transformers

In the upcoming top‑tier security conference NDSS 2025, Ant Group has five papers accepted, covering distributed systems, privacy computing and related fields. This article focuses on two of those papers, examining scalability in distributed systems and efficiency in privacy‑preserving inference.

Rondo: Scalable and Reconfiguration‑Friendly Randomness Beacon proposes a new distributed randomness beacon (DRB) protocol that generates unpredictable, unmanipulable random sequences. Rondo achieves high scalability and easy reconfiguration by employing a novel batch asynchronous verifiable secret sharing (bAVSS) variant with partial output (bAVSS‑PO), attaining optimal O(n) message cost in the sharing phase. To support dynamic reconfiguration, the Rondo‑BFT protocol periodically produces beacon outputs in a partially synchronous network, enhancing practicality and security. Experiments on a near‑hundred‑node inter‑continental deployment demonstrate high throughput and strong scalability, making Rondo suitable for blockchain and other distributed applications.

BumbleBee: Secure Two‑party Inference Framework for Large Transformers tackles private inference for massive Transformer models. By optimizing matrix multiplication and activation function protocols, BumbleBee dramatically reduces communication overhead—about a 90% reduction compared with prior work—and speeds up inference. In benchmark tests on five Transformer models, BumbleBee achieves superior performance, notably generating a token for the LLaMA‑7B model on CPU in roughly 14 minutes, offering a new solution for privacy‑preserving AI inference.

The two papers together address key data‑security challenges from the perspectives of randomness generation and privacy protection, illustrating a shift from isolated technical breakthroughs toward systematic innovation in secure distributed and AI systems.

Authors for further discussion: Xu Wenbo (Ant Chain Technical Expert) on Rondo and Huang Zhichong (Ant Research Institute Researcher) on BumbleBee.

Live Stream Guide Time: February 20, 2025, 18:00‑20:00 Platforms: WeChat Channels (Ant Technology Research Institute, Ant Technology AntTech), Bilibili (Ant Technology Research Institute) – please follow and reserve.