谛听Diting Lab

Crypto & Blockchain · Tue, Sep 29, 2026

Bitcoin, Ethereum, stablecoins, DeFi, smart contracts, scaling and consensus.

IACR ePrint Public-key cryptography

PQMZ: Formally Verified Falcon-PKR for Post-Quantum Cryptographic Migration in Zcash

In briefPQMZ: A Formally Verified Falcon-PKR Scheme for Post-Quantum Migration in Zcash

SummaryThis paper studies Falcon-PKR, a public-key-recovery variant of the Falcon signature scheme, for Zcash-like transparent transactions. It stores a compact digest instead of the full public key and reconstructs a candidate key from the signature during verification, reportedly reducing combined key and signature size by about 15%. The authors machine-check an EUF-CMA reduction in the Random Oracle Model using EasyCrypt and verify honest-signing correctness. A complementary case study covers an extended SECUER-based multi-party computation ceremony architecture.

Hannes Hacker, Shekoufeh Neisarian, Sara Zain et al.

★★★★☆ Notable
arXiv cs.GT · cs.CR · econ.TH

When One Leak Pays Forever: Context Binding and the Price of Deterring Collusion

In briefOne Leak, Lasting Gains: Binding Keys to Context to Deter Collusion

SummaryThe paper studies collusion in threshold-encrypted mempools, where a coalition of decryption committee members that sells its decryption capability to a front-runner can expose every later block the capability still decrypts. The authors show that deterrence holds if and only if each coalition's penalty covers the largest discounted value a single leak can reach. Binding each leak to its own round, such as with per-epoch keys, lowers the required stake substantially.

WhyModels collusive key leaks in threshold-encrypted mempools as a repeated game, proving penalties must cover the largest discounted single-leak value, with rigorous horizon-dependent bounds.

Tingyi Lin, Shawn Yu, Ruoran Lai et al.

★★★★☆ Notable
arXiv cs.AI Code

Reasoning with Continuous Latent Diffusion

In briefContinuous latent diffusion for math and code reasoning (CEDR)

SummaryThis paper introduces CEDR, a training and inference pipeline for continuous latent diffusion models aimed at mathematical reasoning and code generation. It learns compact answer representations from three hidden layers of the Qwen3-4B model and uses a staged curriculum with a compact learned prompt encoder, plus reinforcement learning with verifiable rewards. The authors report pass@1 and pass@8 accuracies on GSM8K, MATH500, HumanEval+, and MBPP+, and state that CEDR outperforms recent continuous-diffusion baselines at comparable scales.

WhyApplies continuous latent diffusion to math and code reasoning with a compact prompt encoder and RL; benchmark results and ablations compare favorably against continuous-diffusion baselines.

Xiang Cheng