Classification-Based Anomaly Detection for General Data
Authors: Liron Bergman, Yedid Hoshen
ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our method is shown to obtain state-of-the-art accuracy and is applicable to broad data types. The strong performance of our method is extensively validated on multiple datasets from different domains. We perform experiments to validate the effectiveness of our distance-based approach and the performance of the general class of transformations we introduced for non-image data. |
| Researcher Affiliation | Academia | School of Computer Science and Engineering The Hebrew University of Jerusalem, Israel |
| Pseudocode | Yes | Algorithm 1 GOAD: Training Algorithm. Algorithm 2 GOAD: Evaluation Algorithm. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described in this paper. |
| Open Datasets | Yes | Cifar10: To evaluate the performance of our method, we perform experiments on the Cifar10 dataset. Fasion MNIST: In Tab. 2, we present a comparison between our method (GOAD) and the strongest baseline methods (Deep SVDD and GEOM) on the Fashion MNIST dataset. Datasets: We evaluate on small-scale medical datasets Arrhythmia, Thyroid as well as large-scale cyber intrusion detection datasets KDD and KDDRev. Our configuration follows that of Zong et al. (2018). Categorical attributes are encoded as one-hot vectors. For completeness the datasets are described in the appendix A.2. The Arrhythmia dataset was the smallest examined. A cardiology dataset from the UCI repository (Asuncion & Newman, 2007) containing attributes related to the diagnosis of cardiac arrhythmia in patients. Thyroid: A medical dataset from the UCI repository (Asuncion & Newman, 2007). KDD: The KDD Intrusion Detection dataset was created by an extensive simulation of a US Air Force LAN network. We use the UCI KDD 10% dataset. |
| Dataset Splits | Yes | We train all compared methods on 50% of the normal data. The methods are evaluated on 50% of the normal data as well as all the anomalies. We use the standard protocol of training on all training images of a single digit and testing on all test images. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions using ADAM for optimization but does not provide specific software names with version numbers (e.g., library or solver names with version numbers like Python 3.8, PyTorch 1.9) needed to replicate the experiment. |
| Experiment Setup | Yes | In our method, we used a margin of s = 0.1 (we also run GOAD with s = 1, shown in the appendix). Similarly to He et al. (2018), to stabilize training, we added a softmax + cross entropy loss, as well as L2 norm regularization for the extracted features f(x). Each matrix has dimensionality L r, where L is the data dimension and r is a reduced dimension. For Arryhthmia and Thyroid we used r = 32, for KDD and KDDrev we used r = 128 and r = 64 respectively, the latter due to high memory requirements. We used 256 tasks for all datasets apart from KDD (64) due to high memory requirements. We set the bias term to 0. For C we used fully-connected hidden layers and leaky-Re LU activations (8 hidden nodes for the small datasets, 128 and 32 for KDDRev and KDD). We optimized using ADAM with a learning rate of 0.001. Similarly to He et al. (2018), to stabilize the triplet center loss training, we added a softmax + cross entropy loss. |