Logic Rule Guided Attribution with Dynamic Ablation

Authors: Jianqiao An, Yuandu Lai, Yahong Han77-85

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Both qualitative and quantitative experiments are conducted to evaluate the proposed DA.
Researcher Affiliation Academia Jianqiao An1,2, Yuandu Lai1,2,3, , Yahong Han1,2, 1College of Intelligence and Computing, Tianjin University, Tianjin, China 2Tianjin Key Lab of Machine Learning, Tianjin University, Tianjin, China 3Peng Cheng Laboratory, Shenzhen, China {anjianqiao, yuandulai, yahong}@tju.edu.cn
Pseudocode Yes Algorithm 1: Dynamic Ablation
Open Source Code No The paper does not provide an explicit statement about releasing its source code for the Dynamic Ablation (DA) method, nor does it include a link to a code repository.
Open Datasets Yes Publicly available object classification datasets, namely, ILSVRC2012 (Russakovsky et al. 2015) val and CIFAR100 (Krizhevsky, Nair, and Hinton 2009) are used as input images.
Dataset Splits No The paper mentions using "ILSVRC2012 val" as input images, indicating a standard validation set was used for a dataset. However, it does not provide details on specific train/validation/test splits, percentages, or methodology for its overall experimental setup or how it handled validation within its own pipeline for all experiments.
Hardware Specification Yes We use a Ge Force GTX TITAN X GPU with 12GB of memory and the Res Net-50 as the base model in this experiment.
Software Dependencies No The paper mentions using "Py Torch model zoo" for base models and the "SLIC" algorithm, but it does not specify version numbers for any of these software components (e.g., PyTorch version, SLIC version).
Experiment Setup Yes We set the iteration number as 500, superpixel number as 500. The hyperparameters αstep, fc, and k in our method are set to 0.2, 0.95, and 4, respectively.