Adversarial Regression with Multiple Learners

Authors: Liang Tong, Sixie Yu, Scott Alfeld, vorobeychik

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conducted experiments on three datasets: Wine Quality (redwine),PDF malware (PDF), and Boston Housing Market (boston).
Researcher Affiliation Academia 1Department of EECS, Vanderbilt University, Nashville, TN, USA 2Computer Science Department, Amherst College, Amherst, MA, USA.
Pseudocode No The paper describes computational methods and proofs, but it does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks, nor does it present structured steps in a code-like format.
Open Source Code No The paper refers to an open-sourced tool 'mimicus' (https://github.com/srndic/mimicus) used for data extraction, but it does not provide a link or explicit statement about the availability of the authors' own source code for the methodology described in the paper.
Open Datasets Yes We conducted experiments on three datasets: Wine Quality (redwine),PDF malware (PDF), and Boston Housing Market (boston). The Wine Quality dataset (Cortez et al., 2009)
Dataset Splits No The paper states: 'The dataset is equally divided into a training set (Xtrain, ytrain) and a testing set (Xtest, ytest).' It does not explicitly mention or detail a separate validation set split.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU/GPU models, memory, or specific computing environments with specifications) used for running the experiments.
Software Dependencies No The paper mentions tools like 'mimicus' and 'peepdf' but does not provide specific version numbers for these or any other software libraries or dependencies used in the experiments.
Experiment Setup Yes Remember that in Eq.(11) there are three hyper-parameters in the defender s loss function: λ, β, and z. λ is the regularization coefficient in the attacker s loss function shown in Eq.(4). It is negatively proportional to the attacker s strength. β is the probability of a test data being malicious. z is the predication targets of the attacker. [...] We denote by ˆλ = 0.5 and ˆβ = 0.8 the defender s estimates of the true λ and β. [...] We let = 5σr 1, where 1 is a vector with all elements equal to one. [...] The number of learners is set to 5. [...] The regularization parameters of Lasso and Ridge were selected by cross-validation.