\texttt{Range-Arithmetic}: Verifiable Deep Learning Inference on an Untrusted Party

The article introduces \texttt{Range-Arithmetic}, a framework designed for verifiable deep neural network (DNN) inference when computations are outsourced to untrusted parties in decentralized systems. It innovatively transforms non-arithmetic operations into verifiable arithmetic steps using sum-check protocols and concatenated range proofs, avoiding the complexities associated with Boolean encoding and high-degree polynomials. This approach not only matches the performance of existing methods but also significantly reduces verification costs, computational effort for the untrusted party, and communication overhead, making it a practical solution for practitioners in decentralized machine learning environments.