CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances (2020-07-16T00:00:00.000000Z)

TL;DR

A simple, yet effective method named contrasting shifted instances (CSI), inspired by the recent success on contrastive learning of visual representations, in addition to contrasting a given sample with other instances as in conventional Contrastive learning methods, contrasts the sample with distributionally-shifted augmentations of itself.

Abstract

Novelty detection, i.e., identifying whether a given sample is drawn from outside the training distribution, is essential for reliable machine learning. To this end, there have been many attempts at learning a representation well-suited for novelty detection and designing a score based on such representation. In this paper, we propose a simple, yet effective method named contrasting shifted instances (CSI), inspired by the recent success on contrastive learning of visual representations. Specifically, in addition to contrasting a given sample with other instances as in conventional contrastive learning methods, our training scheme contrasts the sample with distributionally-shifted augmentations of itself. Based on this, we propose a new detection score that is specific to the proposed training scheme. Our experiments demonstrate the superiority of our method under various novelty detection scenarios, including unlabeled one-class, unlabeled multi-class and labeled multi-class settings, with various image benchmark datasets.

Authors

Jihoon Tack

3 papers

Sangwoo Mo

1 papers

Jongheon Jeong

4 papers

TL;DR

Abstract

Authors

References87 items

A Framework For Contrastive Self-Supervised Learning And Designing A New Approach

What Should Not Be Contrastive in Contrastive Learning

Hybrid Discriminative-Generative Training via Contrastive Learning

Contrastive Training for Improved Out-of-Distribution Detection

Adversarial Self-Supervised Contrastive Learning

What makes for good views for contrastive learning

Classification-Based Anomaly Detection for General Data

Supervised Contrastive Learning

CURL: Contrastive Unsupervised Representations for Reinforcement Learning

Regularizing Class-Wise Predictions via Self-Knowledge Distillation

Improved Baselines with Momentum Contrastive Learning

A Simple Framework for Contrastive Learning of Visual Representations

Your Classifier is Secretly an Energy Based Model and You Should Treat it Like One

Novelty Detection Via Blurring

Momentum Contrast for Unsupervised Visual Representation Learning

Self-supervised Label Augmentation via Input Transformations

Input complexity and out-of-distribution detection with likelihood-based generative models

Rethinking Data Augmentation: Self-Supervision and Self-Distillation

Detecting Out-of-Distribution Inputs to Deep Generative Models Using Typicality

Using Self-Supervised Learning Can Improve Model Robustness and Uncertainty

Contrastive Multiview Coding

Likelihood Ratios for Out-of-Distribution Detection

Detecting Out-of-Distribution Inputs to Deep Generative Models Using a Test for Typicality

Deep Semi-Supervised Anomaly Detection

Data-Efficient Image Recognition with Contrastive Predictive Coding

OCGAN: One-Class Novelty Detection Using GANs With Constrained Latent Representations

Using Pre-Training Can Improve Model Robustness and Uncertainty

Revisiting Self-Supervised Visual Representation Learning

WAIC, but Why? Generative Ensembles for Robust Anomaly Detection

Deep Anomaly Detection with Outlier Exposure

Do Deep Generative Models Know What They Don't Know?

Image Anomaly Detection with Generative Adversarial Networks

A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks

Representation Learning with Contrastive Predictive Coding

Generative Probabilistic Novelty Detection with Adversarial Autoencoders

Deep One-Class Classification

Robust Physical-World Attacks on Deep Learning Visual Classification

Places: A 10 Million Image Database for Scene Recognition

Unsupervised Feature Learning via Non-parametric Instance Discrimination

Deep Anomaly Detection Using Geometric Transformations

Deep Autoencoding Gaussian Mixture Model for Unsupervised Anomaly Detection

Unsupervised Representation Learning by Predicting Image Rotations

DOTA: A Large-Scale Dataset for Object Detection in Aerial Images

Training Confidence-calibrated Classifiers for Detecting Out-of-Distribution Samples

Improved Regularization of Convolutional Neural Networks with Cutout

Large Batch Training of Convolutional Networks

Scaling SGD Batch Size to 32K for ImageNet Training

On Calibration of Modern Neural Networks

Confident Multiple Choice Learning

Enhancing The Reliability of Out-of-distribution Image Detection in Neural Networks

Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour

Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery

A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks

SGDR: Stochastic Gradient Descent with Warm Restarts

Concrete Problems in AI Safety

Deep Structured Energy Based Models for Anomaly Detection

Deep Residual Learning for Image Recognition

Intelligible Models for HealthCare: Predicting Pneumonia Risk and Hospital 30-day Readmission

Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

Obtaining Well Calibrated Probabilities Using Bayesian Binning

Going deeper with convolutions

Food-101 - Mining Discriminative Components with Random Forests

A review of novelty detection

Describing Textures in the Wild

The Caltech-UCSD Birds-200-2011 Dataset

Identifying the signs of fraudulent accounts using data mining techniques

Anomaly detection: A survey

Caltech-256 Object Category Dataset

A Visual Vocabulary for Flower Classification

Anomaly detection of web-based attacks

Support Vector Method for Novelty Detection

Design of an image edge detection filter using the Sobel operator

Care of aged doctors

Implicit Generation and Modeling with Energy Based Models

Novel Dataset for Fine-Grained Image Categorization : Stanford Dogs

Reading Digits in Natural Images with Unsupervised Feature Learning