Active Learning on Pre-trained Language Model with Task-Independent Triplet Loss
Authors: Seungmin Seo, Donghyun Kim, Youbin Ahn, Kyong-Ho Lee11276-11284
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To assess the effectiveness of the proposed method, we compare the proposed method with state-of-the-art active learning methods on two tasks, relation extraction and sentence classification. Experimental results show that our method outperforms baselines on the benchmark datasets. |
| Researcher Affiliation | Academia | Seungmin Seo, Donghyun Kim, Youbin Ahn, and Kyong-Ho Lee Department of Computer Science, Yonsei University, Seoul, Republic of Korea |
| Pseudocode | Yes | Algorithm 1: Active learning with BATL |
| Open Source Code | No | The paper does not provide any explicit statement or link for open-source code availability. |
| Open Datasets | Yes | For relation extraction, we used two publicly accessible dataset, NYT-10 (Riedel, Yao, and Mc Callum 2010) and Wiki-KBP (Ellis et al. 2013). ... For sentence classification, we used two benchmark datasets, AG News (Zhang, Zhao, and Le Cun 2015) and Pub Med (Dernoncourt and Lee 2017). |
| Dataset Splits | No | Table 1 provides 'Train' and 'Test' splits with explicit numbers for each dataset (e.g., NYT-10: Train 522,611, Test 172,448), but it does not explicitly state a separate 'validation' dataset split. |
| Hardware Specification | Yes | The experiments are performed on Ge Force RTX 2080 Ti and AMD Ryzen 7 3700X CPUs. |
| Software Dependencies | No | The paper mentions models and frameworks like GPT, BERT, SCIBERT, but does not provide specific version numbers for any software dependencies or libraries. |
| Experiment Setup | Yes | We evaluated sampling strategies on the relation extraction with varying batch size K = {500, 2000} for NYT-10, and K = {50, 200} for Wiki KBP. We set the batch size K = 100 for sentence classification. The learning rate is 2e 5, and scaling parameter λ = 1. |