يمكن للبشر التمييز بين فئات جديدة بكفاءة للغاية مع عدد قليل من الأمثلة، إلى حد كبير بسبب حقيقة أن البشر يمكنهم الاستفادة من المعرفة التي تم الحصول عليها من المهام ذات الصلة.ومع ذلك، يميل نموذج تصنيف النص في التعلم العميق إلى الكفاح لتحقيق أداء مرض عندما تكون البيانات المسمى نادرة.مستوحاة من الذكاء البشري، نقترح تقديم المعرفة الخارجية إلى سلطة قليلة التعلم لتقليد المعرفة الإنسانية.يتم التحقيق في شبكة مولدات المعلمة الرواية بهذا الغاية، والتي تتمكن من استخدام المعرفة الخارجية لتوليد مقاييس مختلفة لمهام مختلفة.المسلحة مع هذه الشبكة، يمكن لمهام مماثلة استخدام مقاييس مماثلة في حين تستخدم المهام المختلفة مقاييس مختلفة.من خلال التجارب، نوضح أن أسلوبنا تتفوق على نماذج تصنيف النص القليلة لقلة سوتا.
Humans can distinguish new categories very efficiently with few examples, largely due to the fact that human beings can leverage knowledge obtained from relevant tasks. However, deep learning based text classification model tends to struggle to achieve satisfactory performance when labeled data are scarce. Inspired by human intelligence, we propose to introduce external knowledge into few-shot learning to imitate human knowledge. A novel parameter generator network is investigated to this end, which is able to use the external knowledge to generate different metrics for different tasks. Armed with this network, similar tasks can use similar metrics while different tasks use different metrics. Through experiments, we demonstrate that our method outperforms the SoTA few-shot text classification models.
References used
https://aclanthology.org/
The ability to continuously expand knowledge over time and utilize it to rapidly generalize to new tasks is a key feature of human linguistic intelligence. Existing models that pursue rapid generalization to new tasks (e.g., few-shot learning methods
Natural Language Processing (NLP) is increasingly relying on general end-to-end systems that need to handle many different linguistic phenomena and nuances. For example, a Natural Language Inference (NLI) system has to recognize sentiment, handle num
Meta-learning has achieved great success in leveraging the historical learned knowledge to facilitate the learning process of the new task. However, merely learning the knowledge from the historical tasks, adopted by current meta-learning algorithms,
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
Meta learning aims to optimize the model's capability to generalize to new tasks and domains. Lacking a data-efficient way to create meta training tasks has prevented the application of meta-learning to the real-world few shot learning scenarios. Rec