ZKP Frontier Tracker 🎯
零知识证明 前沿热点追踪 🎯
Thu 星期四
01.22
2026
zkRNN: Zero-Knowledge Proofs for Recurrent Neural Network Inference https://eprint.iacr.org/2026/073
Zarinjouei et al. proposed zkRNN in their paper, a ZKP system for RNN inference based on GKR protocol and recursive sum-check, enabling sequence-length-independent verification. Zarinjouei等人在论文中提出zkRNN系统，基于GKR协议和递归sum-check框架，为RNN推理提供ZK证明，实现序列长度无关的验证。
Notes
- zkRNN addresses ZKP challenges for RNNs due to temporal dependencies and weight sharing
- Built on GKR protocol and recursive sum-check framework, adapted for RNN structure
- Circuit representation encodes hidden-state transitions, unrolls computation across time steps
- Polylogarithmic verifier time and proof size in final iteration circuit, independent of sequence length
- Evaluation shows second-scale proving and millisecond-scale verification
- Enables verifiable inference while preserving model parameter privacy