This repository, ‘AdvancedLearning_TextClassification_PaperCollection’, is a curated collection of papers focusing on advanced learning techniques such as meta-learning and few-shot learning, with a special emphasis on text classification. The papers included here provide valuable insights into the latest research and developments in these fields.
- Authors: Jincheng Xu, Qingfeng Du
- Affiliation: School of Software Engineering, Tongji University, Shanghai 200092, China
- Publication Details: Pattern Recognition Letters, Elsevier, Received 8 February 2019, Revised 4 October 2019, Accepted 4 May 2020, Available online 8 May 2020
- Description: The paper introduces FSML (Few-Shot text classification in a Meta-Learning framework), a novel approach for addressing few-shot learning challenges in NLP. By leveraging the Maximum Mean Discrepancy (MMD) metric, FSML minimizes the distance between support and query distributions within each learning task, leading to improved generalization and accuracy gains in few-shot text classification tasks compared to the established MAML framework.
- Authors: Huaxiu Yao, Yingxin Wu, Maruan Al-Shedivat, Eric P. Xing
- Publication Details: Submitted on 10 Sep 2021
- Authors: Pankaj Sharma, Minh Tran, and Imran Qureshi
- Description: This paper investigates the application of meta-learning and robustness techniques on a broad corpus of benchmark text and medical data. The authors developed new data pipelines, combined language models with meta-learning approaches, and extended existing meta-learning algorithms to minimize worst-case loss. They found that meta-learning on text is a suitable framework for text-based data, providing better data efficiency and comparable performance to few-shot language models. It can be successfully applied to medical note data, and when coupled with Distributionally Robust Optimization (DRO), it can improve worst-case loss across disease codes. The authors also explored the impact of predictions by meta-learning models once they are optimized for worst-case expected loss due to atypical groups of data. They concluded that DRO combined with MAML does improve prediction and accounts for distribution shift, and that DRO combined with ProtoNet gives mixed results.
- Authors: José Marcio Duarte, Lilian Berton
- Publication Details: Artificial Intelligence Review (2023) 56:9401–9469, Published online: 31 January 2023