Improving Anomaly Detection with a Self-Supervised Task Based on Generative Adversarial Network

Chai, Heyan; Su, Weijun; Tang, Siyu; Ding, Ye; Fang, Binxing; Liao, Qing

doi:10.1109/icassp43922.2022.9747461

Cited by 3 publications

(3 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although existing multi-task models take into account sentiment information, they still obtain a lower performance on SemEval-16. In terms of F M ac , the SMTL-HW model surpasses JOINT [10], MTIN [78], AT-JSS-Lex [11], and MT-LRM-BERT [51] models by 12.3%, 7.6%, 7.1%, and 5%, respectively. These results highlight the effectiveness of the main components incorporated in our SMTL-HW model, namely the sequential architecture, and task weighting.…”

Section: Comparisons With Previous Studiesmentioning

confidence: 92%

See 1 more Smart Citation

Enhancing Stance Detection through Sequential Weighted Multi-task Learning

Alturayeif

Luqman

Ahmed

2023

Preprint

View full text Add to dashboard Cite

The exponential growth of user-generated content on social media platforms, online news outlets, and digital communication has necessitated the development of automated tools for analyzing opinions and attitudes expressed in text. Stance detection, a critical task in Natural Language Processing (NLP), aims to identify the underlying perspective or viewpoint of an individual or group towards a specific topic or target. This paper explores the challenges of stance detection, particularly in the context of social media, where brevity, informality, and limited contextual information prevail. While sentiment analysis focuses on explicit sentiment polarity, stance detection classifies the stance or viewpoint of a text towards a target, often of an abstract nature. This study introduces two multi-task learning (MTL) models that integrate sentiment analysis and sarcasm detection tasks to enhance stance detection performance. Four task weighting techniques are proposed and evaluated, and their effectiveness in the MTL models is demonstrated. Extensive evaluations on English and Arabic benchmark datasets highlight the advantages of the proposed models. Among them, the multi-target sequential MTL model stands out with its hierarchical weighting approach, as it achieves state-of-the-art performance. The study underscores the potential of MTL in improving stance detection and provides insights into the interaction between sentiment and stance, while considering the impact of sarcasm.

show abstract

Section: Comparisons With Previous Studiesmentioning

confidence: 92%

“…MTIN [78] is a Multi-Task Interaction Network model that simultaneously learns stance and sentiment with a word-level task interaction and task-related graphs.…”

Section: Comparisons With Previous Studiesmentioning

confidence: 99%

Enhancing Stance Detection through Sequential Weighted Multi-task Learning

Alturayeif

Luqman

Ahmed

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Creating supervision signals from data itself is an essence insight in self-supervised learning. Inspired by self-supervised research in image anomaly detection [8], some self-supervised methods are also designed for time series data. They assign class labels to different augmentation operations (e.g., adding noise, reversing, scaling, and smoothing) [54], neural transformations [38], contiguous and separate time segments [9], or different time resolutions [18].…”

Section: Related Workmentioning

confidence: 99%

Calibrated One-class Classification for Unsupervised Time Series Anomaly Detection

Xu¹,

Wang²,

Jian³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Unsupervised time series anomaly detection is instrumental in monitoring and alarming potential faults of target systems in various domains. Current state-of-the-art time series anomaly detectors mainly focus on devising advanced neural network structures and new reconstruction/prediction learning objectives to learn data normality (normal patterns and behaviors) as accurately as possible. However, these one-class learning methods can be deceived by unknown anomalies in the training data (i.e., anomaly contamination). Further, their normality learning also lacks knowledge about the anomalies of interest. Consequently, they often learn a biased, inaccurate normality boundary. This paper proposes a novel one-class learning approach, named calibrated one-class classification, to tackle this problem. Our one-class classifier is calibrated in two ways: (1) by adaptively penalizing uncertain predictions, which helps eliminate the impact of anomaly contamination while accentuating the predictions that the one-class model is confident in, and (2) by discriminating the normal samples from native anomaly examples that are generated to simulate genuine time series abnormal behaviors on the basis of original data. These two calibrations result in contamination-tolerant, anomaly-informed one-class learning, yielding a significantly improved normality modeling. Extensive experiments on six real-world datasets show that our model substantially outperforms twelve state-of-the-art competitors and obtains 6% -31% 𝐹 1 score improvement. The source code is available at https://github.com/xuhongzuo/couta.

show abstract