Learning Prototype Representations Across Few-Shot Tasks for Event Detection
Viet Dac Lai, Franck Dernoncourt, Thien Huu Nguyen
2021Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing20 citationsDOIOpen Access PDF
Abstract
We address the sampling bias and outlier issues in few-shot learning for event detection, a subtask of information extraction. We propose to model the relations between training tasks in episodic few-shot learning by introducing cross-task prototypes. We further propose to enforce prediction consistency among classifiers across tasks to make the model more robust to outliers. Our extensive experiment shows a consistent improvement on three fewshot learning datasets. The findings suggest that our model is more robust when labeled data of novel event types is limited.
Topics & Concepts
Computer scienceOutlierEvent (particle physics)Consistency (knowledge bases)Artificial intelligenceTask (project management)Machine learningShot (pellet)Anomaly detectionCode (set theory)Sampling (signal processing)One shotSource codeData miningSet (abstract data type)Computer visionProgramming languageManagementQuantum mechanicsFilter (signal processing)EconomicsEngineeringPhysicsOperating systemMechanical engineeringOrganic chemistryChemistryDomain Adaptation and Few-Shot LearningMultimodal Machine Learning ApplicationsAnomaly Detection Techniques and Applications