zkML zero-knowledge machine learning faces a key challenge in its application: input data often leads to a significant expansion of proof size, which directly affects the efficiency and scalability of the system. Some projects have found solutions by optimizing the witness generation process—performing intelligent preprocessing before proof generation effectively reduces redundant data, significantly compressing the final proof size. This approach is crucial for enhancing the performance of zero-knowledge proofs in practical applications, especially in scenarios sensitive to on-chain costs.