Adversarial Attacks on Online Learning to Rank with Click Feedback
Authors: Jinhang Zuo, Zhiyao Zhang, Zhiyong Wang, Shuai Li, Mohammad Hajiesmaili, Adam Wierman
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on synthetic and real data further validate the effectiveness of our proposed attack algorithms. We also conduct experiments on both synthetic and real-world data to evaluate our proposed attack strategies. Experimental results show that they can effectively attack the corresponding OLTR algorithms with less cost compared to other baselines. |
| Researcher Affiliation | Collaboration | 1University of Massachusetts Amherst 2California Institute of Technology 3Southeast University 4The Chinese University of Hong Kong 5Shanghai Jiao Tong University {jhzuo,hajiesmaili}@cs.umass.edu muirheadzhang@gmail.com zywang21@cse.cuhk.edu.hk shuaili8@sjtu.edu.cn adamw@caltech.edu |
| Pseudocode | Yes | Algorithm 1 Attacks against the UCB algorithm on stochastic bandits with binary feedback ... Algorithm 2 Attack against the PBM-UCB algorithm ... Algorithm 3 CAL_ALPHA ... Algorithm 4 Attack against the Cascade UCB algorithm ... Algorithm 5 Attack against arbitrary algorithm |
| Open Source Code | No | The paper does not provide an explicit statement or link indicating that the source code for their methodology is openly available. |
| Open Datasets | Yes | We conduct experiments using both synthetic and real data (Movie Lens 20M dataset [26]). |
| Dataset Splits | No | The paper mentions using synthetic and real data but does not specify training, validation, or test splits. It only states the total rounds T. |
| Hardware Specification | No | The paper does not specify any hardware used for the experiments (e.g., CPU, GPU models, or memory). |
| Software Dependencies | No | The paper does not mention specific software dependencies or their version numbers, such as programming languages, libraries, or frameworks. |
| Experiment Setup | Yes | We use ϵ = 0.1 for the PBM-UCB algorithm. For the synthetic data, we take L = 16, K = 8, T = 100,000; {µi}L i=1 are sampled from uniform distribution U(0, 1) for Figure 1a, and from U(0, x) for Figure 1b. For the real data, we take L = 100, K = 10, T = 100,000; {µi}L i=1 are extracted according to [2]. |