分类: 计算机科学 >> 计算机科学的集成理论 提交时间: 2022-11-28 合作期刊: 《数据智能(英文)》
摘要: Few-shot learning has been proposed and rapidly emerging as a viable means for completing various tasks. Many few-shot models have been widely used for relation learning tasks. However, each of these models has a shortage of capturing a certain aspect of semantic features, for example, CNN on long-range dependencies part, Transformer on local features. It is difficult for a single model to adapt to various relation learning, which results in a high variance problem. Ensemble strategy could be competitive in improving the accuracy of few-shot relation extraction and mitigating high variance risks. This paper explores an ensemble approach to reduce the variance and introduces fine-tuning and feature attention strategies to calibrate relation-level features. Results on several few-shot relation learning tasks show that our model significantly outperforms the previous state-of-the-art models.
分类: 计算机科学 >> 计算机科学技术其他学科 提交时间: 2022-11-15
摘要: Few-shot learning has been proposed and rapidly emerging as a viable means for completing various tasks. Many few-shot models have been widely used for relation learning tasks. However, each of these models has a shortage of capturing a certain aspect of semantic features, for example, CNN on long-range dependencies part, Transformer on local features. It is difficult for a single model to adapt to various relation learning, which results in a high variance problem. Ensemble strategy could be competitive in improving the accuracy of few-shot relation extraction and mitigating high variance risks. This paper explores an ensemble approach to reduce the variance and introduces fine-tuning and feature attention strategies to calibrate relation-level features. Results on several few-shot relation learning tasks show that our model significantly outperforms the previous state-of-the-art models.
分类: 计算机科学 >> 自然语言理解与机器翻译 提交时间: 2019-05-12
摘要: Abstract. Computational chemistry develops fast in recent years due to the rapid growth and breakthroughs in AI. Thanks for the progress in natural language processing, researchers can extract more fine-grained knowledge in publications to stimulate the development in computational chemistry. While the works and corpora in chemical entity extraction have been restricted in the biomedicine or life science field instead of the chemistry field, we build a new corpus in chemical bond field anno- tated for 7 types of entities: compound, solvent, method, bond, reaction, pKa and pKa value. This paper presents a novel BERT-CRF model to build scientific chemical data chains by extracting 7 chemical entities and relations from publications. And we propose a joint model to ex- tract the entities and relations simultaneously. Experimental results on our Chemical Special Corpus demonstrate that we achieve state-of-art and competitive NER performance.
分类: 物理学 >> 基本粒子与场物理学 提交时间: 2016-05-09
摘要: Motivated by the recent progress of direct search for the productions of stop pair and sbottom pair at the LHC, we examine the constraints of the search results on the stop ( (t) over tilde (1)) mass in natural SUSY. We first scan the parameter space of natural SUSY in the framework of MSSM, considering the constraints from the Higgs mass, B-physics and electroweak precision measurements. Then in the allowed parameter space we perform a Monte Carlo simulation for stop pair production followed by (t) over tilde (1). t (chi) over tilde (0)(1) or (t) over tilde (1). b (chi) over tilde (+)(1) and sbottom pair production followed by (b) over tilde (1) -> b (chi) over tilde (0)(1) or (b) over tilde (1) -> t (chi) over tilde (-)(1). Using the combined results of ATLAS with 20.1 fb(-1) from the search of l + jets + (sic)(T), hadronic t (t) over bar + (sic)(T) and 2b + (sic)(T), we find that a stop lighter than 600 GeV can be excluded at 95% CL in this scenario.
分类: 物理学 >> 基本粒子与场物理学 提交时间: 2016-05-14
摘要: Confronted with the LHC data of a Higgs boson around 125 GeV, different models of low energy SUSY show different behaviors: some are favored, some are marginally survived and some are strongly disfavored or excluded. In this note we update our previous scan over the parameter space of various low energy SUSY models by considering the latest experimental limits like the LHCb data for B-s -> mu(+)mu(-) and the XENON 100 (2012) data for dark matter-nucleon scattering. Then we confront the predicted properties of the SM-like Higgs boson in each model with the combined 7 TeV and 8 TeV Higgs search data of the LHC. For a SM-like Higgs boson around 125 GeV, we have the following observations: (i) The most favored model is the NMSSM, whose predictions about the Higgs boson can naturally (without any fine tuning) agree with the experimental data at 1 sigma level, better than the SM; (ii) The MSSM can fit the LHC data quite well but suffer from some extent of fine tuning; (iii) The nMSSM is excluded at 3 sigma level after considering all the available Higgs data; (iv) The CMSSM is quite disfavored since it is hard to give a 125 GeV Higgs boson mass and at the same time cannot enhance the di-photon signal rate.