Data Poisoning Attacks on Multi-Task Relationship Learning
Authors: Mengchen Zhao, Bo An, Yaodong Yu, Sulin Liu, Sinno Pan
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on realworld datasets show that MTRL models are very sensitive to poisoning attacks and the attacker can significantly degrade the performance of target tasks, by either directly poisoning the target tasks or indirectly poisoning the related tasks exploiting the task relatedness. |
| Researcher Affiliation | Academia | Mengchen Zhao, Bo An, Yaodong Yu, Sulin Liu, Sinno Jialin Pan School of Computer Science and Engineering, Nanyang Technological University, Singapore 639798 zhao0204@e.ntu.edu.sg, {boan,ydyu,liusl,sinnopan}@ntu.edu.sg |
| Pseudocode | Yes | Algorithm 1: computing Poisoning ATtacks On Multi-task relationship learning (PATOM) |
| Open Source Code | No | No concrete access to source code for the methodology was provided. The paper does not state that the code is publicly available, nor does it provide a link to a repository or supplementary materials containing the code. |
| Open Datasets | Yes | We use three real-world datasets to validate our proposed methods. The Landmine and the MNIST datasets are used for classification tasks and Sarcos dataset is used for regression tasks... Sarcos1 relates to an inverse dynamics problem... http://www.gaussianprocess.org/gpml/data/. Landmine2 consists of 29 tasks... http://people.ee.duke.edu/ lcarin/Landmine Data.zip. MNIST3 is a hand-written digit dataset with 10 classes... http://yann.lecun.com/exdb/mnist/. |
| Dataset Splits | No | Sarcos dataset... The dataset contains 44,484 training examples and 4,449 test examples. MNIST... The dataset contains 60,000 training examples and 10,000 test examples. We use principal component analysis (PCA) to reduce the feature space to a 128-dimensional space. The paper mentions training and testing sets but does not explicitly specify a validation set or detailed split percentages beyond stating total training and test examples for some datasets. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, processor types, memory amounts) were provided for running the experiments. |
| Software Dependencies | No | No specific ancillary software details (e.g., library or solver names with version numbers) were provided. |
| Experiment Setup | Yes | We set the step size η = 100 and the lower level problem parameters λ1 = λ2 = 0.1. The batch size is set to be three times of the clean data. The number of injected data points in each task is set to be 20% of the clean data. |