Data Poisoning Attacks against Autoregressive Models

Authors: Scott Alfeld, Xiaojin Zhu, Paul Barford

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically demonstrate the effectiveness of the optimal attack, compared to random and greedy baselines on synthetic and realworld time series data. We conclude by discussing defensive strategies in the face of Alice-like adversaries.
Researcher Affiliation Collaboration Scott Alfeld, Xiaojin Zhu, and Paul Barford Department of Computer Sciences University of Wisconsin Madison Madison WI 53706, USA com Score, Inc. 11950 Democracy Drive, Suite 600 Reston, VA 20190, USA. {salfeld, jerryzhu, pb}@cs.wisc.edu
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any concrete access information for open-source code (e.g., a specific repository link or explicit statement of code release).
Open Datasets Yes We obtained5 historical US natural gas prices for 2014. After centering the series, we learned an order d = 5 AR model via Yule-Walker estimation (Box, Jenkins, and Reinsel 2011). ...5Data is publicly available from http://www.quandl.com, Quandl Code CME/NGF2015.
Dataset Splits No The paper does not specify dataset splits (e.g., training, validation, test percentages or counts) or cross-validation setup for reproducibility.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments.
Software Dependencies Yes Experiments were conducted using the optimization package cvxopt v1.1.7 (Dahl and Vandenberghe 2006), figures were made with Matplotlib v1.4.3 (Hunter 2007).
Experiment Setup No The paper describes some experimental settings (e.g., d=5, h=7 for synthetic data, β=1/3, W=I; and d=5, h=10, β=0.1 for real-world data), but it does not consolidate them in a dedicated 'Experimental Setup' section or explicitly list them as hyperparameters. It also lacks details like learning rates, optimizer settings, etc., which are common in deep learning papers but might not be applicable here given the problem type. The question asks for *specific* details which are given piecemeal, not as a comprehensive setup.