Update README

README.md

@@ -1,14 +1,9 @@
-# Effective End-to-end Unsupervised Outlier Detection via Inlier Priority of Discriminative Network
-By Siqi Wang, Yijie Zeng, Xinwang Liu, En Zhu, Jianping Yin, Chuanfu Xu, Marius Kloft.  In [NeurIPS 2019](https://papers.nips.cc/paper/8830-effective-end-to-end-unsupervised-outlier-detection-via-inlier-priority-of-discriminative-network).
-
-## Update
-- 2020/07/13: Added multiple score refinement strategies (ensemble, re-weighting) which improve performance of E3Outlier. Added several recent unsupervised outlier detection implementations, including DSEBM, RSRAE, RSRAE+, MO-GAAL.  Checkout the `extension` branch for more information.
+# Demo implementation of E3Outlier for Unsupervised Outlier Detection
 
 ## Introduction
-This is the official implementation of the E3Outlier framework presented by "Effective End-to-end Unsupervised Outlier Detection via Inlier Priority of Discriminative Network".
-The codes are used to reproduce experimental results of  E3Outlier and other unsupervised outlier detection (UOD) methods reported in the [paper](https://papers.nips.cc/paper/8830-effective-end-to-end-unsupervised-outlier-detection-via-inlier-priority-of-discriminative-network.pdf).
+This repository provides the implementation of discriminative E3Outlier, an effective and end-to-end method for the unsupervised outlier detection (UOD) task. UOD aims to directly detect outliers from a contaminated unlabeled dataset in a transductive manner, without using any labeled data (e.g. a labeled training set with pure normal data/inliers).
 
-## Requirements
+## Requirements 
 - Python 3.6
 - PyTorch 0.4.1 (GPU)
 - Keras 2.2.0 
@@ -18,39 +13,22 @@ The codes are used to reproduce experimental results of  E3Outlier and other uns
 
 ## Usage
 
-To obtain the results of E3Outlier and other UOD methods compared in the paper with default settings, simply run the following command:
+To run E3Outlier with default settings, simply run the following command:
 
 ```bash
 python outlier_experiments.py
 ```
 
-This will automatically run all UOD methods reported in the manuscript.  Please see ```outlier_experiments.py``` for more details.
+This will automatically run UOD methods on all datasets (`MNIST`, `Fashion-MNIST`, `SVHN`, `CIFAR10` and `CIFAR100`). On each dataset the experiment will be conducted with 5 outlier ratios: 0.05, 0.1, 0.15, 0.2 and 0.25.
 
-After training, to print UOD results for a specific algorithm in AUROC/AUPR, run:
+After learning, the prediction scores and ground truth labels are saved to an `npz` file. To obtain the UOD result for a specific algorithm, run ```evaluate_roc_auc.py``` for evaluation using *Area under the ROC curve (AUROC)*, or ```evaluate_pr_auc.py``` for evaluation using *Area under the PR curve (AUPR)*. Example usage:
 
 ```bash
 # AUROC of E3Outlier on CIFAR10 with outlier ratio 0.1
 python evaluate_roc_auc.py --dataset cifar10 --algo_name e3outlier-0.1
 
-# AUPR of CAE-IF on MNIST with outlier ratio 0.25 and inliers as the postive class
-python evaluate_pr_auc.py --dataset mnist --algo_name cae-iforest-0.25 --postive inliers
-```
-
-The algorithm names are defined in ```outlier_experiments.py```.
-
-## Citation
-
-```
-@incollection{NIPS2019_8830,
-title = {Effective End-to-end Unsupervised Outlier Detection via Inlier Priority of Discriminative Network},
-author = {Wang, Siqi and Zeng, Yijie and Liu, Xinwang and Zhu, En and Yin, Jianping and Xu, Chuanfu and Kloft, Marius},
-booktitle = {Advances in Neural Information Processing Systems 32},
-editor = {H. Wallach and H. Larochelle and A. Beygelzimer and F. d\textquotesingle Alch\'{e}-Buc and E. Fox and R. Garnett},
-pages = {5960--5973},
-year = {2019},
-publisher = {Curran Associates, Inc.},
-url = {http://papers.nips.cc/paper/8830-effective-end-to-end-unsupervised-outlier-detection-via-inlier-priority-of-discriminative-network.pdf}
-}
+# AUPR of E3Outlier on MNIST with outlier ratio 0.25 and inliers as the postive class
+python evaluate_pr_auc.py --dataset mnist --algo_name e3outlier-0.25 --postive inliers
 ```
 
 ## License