arXiv subjects

Huiming Wang

Publications and source records attributed to Huiming Wang.

23 records · Page 2Linked to original sources

arXiv2024

AdaMergeX: Cross-Lingual Transfer with Large Language Models via Adaptive Adapter Merging

As an effective alternative to the direct fine-tuning on target tasks in specific languages, cross-lingual transfer addresses the challenges of limited training data by decoupling ''task ability'' and ''language ability'' by fine-tuning on the target task in the source language and another selected task in the target language, respectively. However, they fail to fully separate the task ability from the source language or the language ability from the chosen task. In this paper, we acknowledge the mutual reliance between task ability and language ability and direct our attention toward the gap between the target language and the source language on tasks. As the gap removes the impact of tasks, we assume that it remains consistent across tasks. Based on this assumption, we propose a new cross-lingual transfer method called $\texttt{AdaMergeX}$ that utilizes adaptive adapter merging. By introducing a reference task, we can determine that the divergence of adapters fine-tuned on the reference task in both languages follows the same distribution as the divergence of adapters fine-tuned on the target task in both languages. Hence, we can obtain target adapters by combining the other three adapters. Furthermore, we propose a structure-adaptive adapter merging method. Our empirical results demonstrate that our approach yields new and effective cross-lingual transfer, outperforming existing methods across all settings.

cs.CL↗

arXiv2023

Janus-graphene: a two-dimensional half-auxetic carbon allotropes with non-chemical Janus configuration

The asymmetric properties of Janus two-dimensional materials commonly depend on chemical effects, such as different atoms, elements, material types, etc. Herein, based on carbon gene recombination strategy, we identify an intrinsic non-chemical Janus configuration in a novel purely sp$^2$ hybridized carbon monolayer, named as Janus-graphene. With the carbon gene of tetragonal, hexagonal, and octagonal rings, the spontaneous unilateral growth of carbon atoms drives the non-chemical Janus configuration in Janus-graphene, which is totally different from the chemical effect in common Janus materials such as MoSSe. A structure-independent half-auxetic behavior is mapped in Janus-graphene that the structure maintains expansion whether stretched or compressed, which lies in the key role of $p_z$ orbital. The unprecedented half-auxeticity in Janus-graphene extends intrinsic auxeticity into pure sp$^2$ hybrid carbon configurations. With the unique half-auxeticity emerged in the non-chemical Janus configuration, Janus-graphene enriches the functional carbon family as a promising candidate for micro/nanoelectronic device applications.

cond-mat.mtrl-sci↗

arXiv2023

Enhancing Few-shot NER with Prompt Ordering based Data Augmentation

Recently, data augmentation (DA) methods have been proven to be effective for pre-trained language models (PLMs) in low-resource settings, including few-shot named entity recognition (NER). However, conventional NER DA methods are mostly aimed at sequence labeling models, i.e., token-level classification, and few are compatible with unified autoregressive generation frameworks, which can handle a wider range of NER tasks, such as nested NER. Furthermore, these generation frameworks have a strong assumption that the entities will appear in the target sequence with the same left-to-right order as the source sequence. In this paper, we claim that there is no need to keep this strict order, and more diversified but reasonable target entity sequences can be provided during the training stage as a novel DA method. Nevertheless, a naive mixture of augmented data can confuse the model since one source sequence will then be paired with different target sequences. Therefore, we propose a simple but effective Prompt Ordering based Data Augmentation (PODA) method to improve the training of unified autoregressive generation frameworks under few-shot NER scenarios. Experimental results on three public NER datasets and further analyses demonstrate the effectiveness of our approach.

cs.CL↗

arXiv2022

Emerging negative Poisson's ratio driven by strong intralayer interaction response in rectangular transition metal chalcogenides

Auxetic behavior quantified by the negative Poisson's ratio (NPR) is commonly attributed to geometry evolution with re-entrant mechanism or other mechanical factors, which is thought to be independent of electronic structures. Thus, searching for electronic effect dominated auxetic behavior is challenging. Herein, from state-of-the-art first-principles calculations, by studying a class of two-dimensional (2D) transition metal chalcogenides (TMCs), namely X2Y2-type (X=Cu, Ag, Au; Y=O, S, Se) rectangular TMCs (R-TMCs), we identify that the monolayer R-Cu2Se2 unconventionally demonstrates a structure-independent anisotropic NPR. In contrast, the NPR is absent in other R-TMCs. The emerging NPR is attributed to the strong strain response of intralayer interaction in R-Cu2Se2, which can be traced to the lone pair electrons and weak electronegativity of Se atoms under multi-orbital hybridization. The emerging NPR would make R-Cu2Se2 a promising candidate in electronics protection, and our study would provide valuable clues and useful guidance for designing advanced auxetic materials.

cond-mat.mtrl-sci↗

arXiv2022

The synergistic modulation of electronic and geometry structures leads to ultra-low thermal conductivity of graphene-like borides (g-B3X5, X=N, P, As)

The design of novel devices with specific technical interests through modulating structural properties and bonding characteristics promotes the vigorous development of materials informatics. Herein, we propose a synergy strategy of component reconstruction by combining geometric configuration and bonding characteristics. With the synergy strategy, we designed a novel two-dimensional (2D) graphene-like borides, e.g. g-B3N5, which possesses counter-intuitive ultra-low thermal conductivity of 21.08 W/mK despite the small atomic mass. The ultra-low thermal conductivity is attributed to the synergy effect of electronics and geometry on thermal transport due to the combining reconstruction of g-BN and nitrogene. With the synergy effect, the dominant acoustic branches are strongly softened, and the scattering absorption and Umklapp process are simultaneously suppressed. Thus, the thermal conductivity is significantly lowered. To verify the component reconstruction strategy, we further constructed g-B3P5 and g-B3As5, and uncovered the ultra-low thermal conductivity of 2.50 and 1.85 W/mK, respectively. The synergy effect and the designed ultra-low thermal conductivity materials with lightweight atomic mass cater to the demand for light development of momentum machinery and heat protection, such as aerospace vehicles, high-speed rail, automobiles.

cond-mat.mtrl-sci↗