Recursive Zero-Knowledge Secures Private Verifiable AI Model Inference
The new recursive ZK framework allows constant-size proofs for massive AI models, solving the critical trade-off between model privacy and verifiability.
Verifiable Fine-Tuning Secures Large Language Models with Zero-Knowledge Proofs
zkLoRA is a new framework that cryptographically verifies LLM fine-tuning correctness without revealing model weights, unlocking private, auditable AI.
Unified Framework Achieves Private Scalable Verifiable Machine Learning
The new proof-composition framework casts verifiable machine learning as succinct matrix computations, delivering linear prover time and architecture privacy for decentralized AI.
Zero-Knowledge Proofs Verifiably Secure Large Language Model Inference
A novel ZKP system, zkLLM, enables the efficient, private verification of 13-billion-parameter LLM outputs, securing AI integrity and intellectual property.
Distributed zkSNARKs Achieve Linear Prover Scalability with Constant Communication
A new distributed zkSNARK protocol, Pianist, achieves linear prover scalability by parallelizing proof generation with constant communication overhead, resolving the ZKP bottleneck for zkRollups.
Zero-Knowledge Mechanisms Enable Private Rules with Public Verifiability
This framework introduces a new cryptographic primitive that allows mechanism rules to remain secret while using ZKPs to publicly verify incentive compatibility and outcomes, removing the need for a trusted mediator.
Scaling zkSNARKs through Application and Proof System Co-Design
This research introduces "silently verifiable proofs" and a co-design approach to drastically reduce communication costs for scalable, privacy-preserving analytics.
Scaling Zero-Knowledge Proofs for Private Aggregation and Delegation
This research introduces novel zero-knowledge proof systems that dramatically reduce server communication costs for private analytics and enhance distributed proof generation scalability, fundamentally improving the efficiency of privacy-preserving computations.
