Optimal Attack against Autoregressive Models by Manipulating the Environment
Authors: Yiding Chen, Xiaojin Zhu3545-3552
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments demonstrate the effectiveness of our attacks. We now demonstrate the effectiveness of control-based whitebox attacks on time series forecast problems. We compare the optimal attacks computed by LQR (for linear f), MPC+i LQR (for nonlinear f), black-box attack, greedy attacks, and the no-attack baseline. |
| Researcher Affiliation | Academia | Yiding Chen, Xiaojin Zhu Department of Computer Sciences, University of Wisconsin-Madison {yiding, jerryzhu}@cs.wisc.edu |
| Pseudocode | Yes | Algorithm 1: LQR(F, C, {y t |t}, {Qt |t}, R, T), Algorithm 2: MPC, Algorithm 3: ILQR, Algorithm 4: System identification attack |
| Open Source Code | No | The paper does not provide an explicit statement or link for open-source code availability for the described methodology. |
| Open Datasets | Yes | This real world data (Tiao and Tsay 1994) models the growth rate of quarterly US real GNP from the first quarter of 1947 to the first quarter of 1991. ... We use the same dynamic in (Fan and Yao 2008) but we change the noise to be wt N(0, 0.12). |
| Dataset Splits | No | The paper does not explicitly provide training/validation/test dataset splits using specific percentages, sample counts, or references to predefined splits for its own experiments. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions using 'lsqnonlin solver in Matlab' but does not provide specific version numbers for Matlab or any other software dependencies. |
| Experiment Setup | Yes | We let T = 10, x0 = 0.0065 (according to (Tiao and Tsay 1994)), x 1 = 0. ... We let λ = 0.001. The time step of MPC is set to be l = 5. Inside the MPC loop, the stopping condition of i LQR is tol = 10 4. The maximum iteration of i LQR is set to be 1000. ... For system identification, we let b = 15, l = 5, p = 3. |