Privacy-Preserving UCB Decision Process Verification via zk-SNARKs
Authors: Xikun Jiang, He Lyu, Chenhao Ying, Yibin Xu, Boris Düdder, Yuan Luo
IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4 Experimental Evaluation and Analysis In this section, we conduct a comprehensive assessment of the effectiveness of zk UCB. Our evaluation focuses on several key aspects: 1. Reward Comparison: We analyze the performance disparities between the standard UCB algorithm and zk UCB, with a particular emphasis on the average reward generated during each step across 100 iterations over 200 steps. 2. Time Analysis: We delve into various time-related metrics for zk UCB, including setup time, compile time (the duration required for converting high-level programs to R1CS), compute witness time, generate proof time, and verify time. These measurements are conducted over 200 steps, with results recorded at intervals of 10 steps. 3. Components Size: We examine the sizes of essential components for zk UCB, including the proving key size, proof size, verifying key size, and witness size. These metrics are assessed over 200 steps, with measurements recorded at 10-step intervals. |
| Researcher Affiliation | Academia | Xikun Jiang1,3 , He Lyu3 , Chenhao Ying1,2 , Yibin Xu3 , Boris D udder3 , Yuan Luo1,2 1Department of Computer Science, Shanghai Jiao Tong University, China 2Shanghai Jiao Tong University (Wuxi) Blockchain Advanced Research Center, China 3Department of Computer Science, University of Copenhagen, Denmark |
| Pseudocode | Yes | Algorithm 1 UCB1 ... Algorithm 2 zk UCB |
| Open Source Code | No | The paper does not provide a statement about releasing its source code or a link to a code repository for the methodology described. |
| Open Datasets | No | The paper describes a simulated experimental setup (a three-armed MAB model) and how data is generated from uniform distributions, rather than utilizing or providing access information for an established publicly available dataset with a citation or link. |
| Dataset Splits | No | The paper describes a reinforcement learning experimental setup involving steps and iterations with a simulated MAB environment, which does not involve explicit training/validation/test dataset splits in the conventional sense of static datasets. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., CPU, GPU models, or memory) used to run the experiments, only discussing software and cryptographic backend choices. |
| Software Dependencies | No | The paper mentions 'Zokrates' as a toolkit and specific components like 'ALT BN128 elliptic curve', 'Groth-16 protocol', and 'Ark backend', but it does not specify version numbers for these software dependencies, which is required for reproducibility. |
| Experiment Setup | Yes | 4.2 Experimental Setup Our experiment leverages Zo Krates, as it enables the selection of a proof system, elliptic curves for pairing, and zk SNARKs backend at our discretion. In this context, we have opted for the ALT BN128 elliptic curve. The Groth-16 protocol is chosen as the proof scheme. Additionally, the Ark backend is selected. ... Setting I This experiment is under a three-armed MAB model and runs a UCB algorithm with 200 steps, where each arm has initially expected rewards set at 0.9, 1.0, and 1.1, respectively. We set three different levels of quantization (2^4, 2^8, 2^16) and calculate the average reward across 100 iterations... Setting II Maintaining the same foundational parameters as in Setting I, this experiment executes a UCB algorithm with various steps (ranging from 20 to 200, with increments of 10 steps per interval) for 100 iterations to average the results. |