AR-Pro: Counterfactual Explanations for Anomaly Repair with Formal Properties

Authors: Xiayan Ji, Anton Xue, Eric Wong, Oleg Sokolsky, Insup Lee

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of our anomaly explainability framework, AR-Pro, on vision (MVTec, Vis A) and time-series (SWa T, WADI, HAI) anomaly datasets.
Researcher Affiliation Academia Xiayan Ji Anton Xue Eric Wong Oleg Sokolsky Insup Lee Department of Computer and Information Science University of Pennsylvania Philadelphia, PA 19104 {xjiae,antonxue,exwong,sokolsky,lee}@seas.upenn.edu
Pseudocode No The paper describes methods and processes but does not include a clearly labeled 'Pseudocode' or 'Algorithm' block.
Open Source Code Yes The code used for the experiments is accessible at: https://github.com/xjiae/arpro.
Open Datasets Yes We demonstrate the effectiveness of our anomaly explainability framework, AR-Pro, on vision (MVTec, Vis A) and time-series (SWa T, WADI, HAI) anomaly datasets.
Dataset Splits Yes Each experiment employs a representative anomaly detector and dataset with predefined train-test splits.
Hardware Specification Yes All experiments were done on a server with three NVIDIA Ge Force RTX 4090 GPUs.
Software Dependencies No The paper mentions using GPT-2, Llama2, DDPM, and Diffusion-TS models, and frameworks like anomalib and Hugging Face implementations, but does not provide specific version numbers for underlying software dependencies such as Python, PyTorch, or CUDA.
Experiment Setup Yes Both Fast Flow and Efficient-AD were trained with Adam W and a learning rate of 10 4 until convergence. Both our versions of GPT-2 and Llama-2 were trained with Adam W and a learning rate of 10 5 until convergence. We randomly sampled 100 instances to compute the mean of each metric in order to evaluate the effect of hyper-parameters λ1, λ2, λ3, λ4 associated with each property-based loss.