A5049450075- Topic Modeling
- Natural Language Processing Techniques
- Electrocatalysts for Energy Conversion
- Cooperative Communication and Network Coding
- Stochastic Gradient Optimization Techniques
- Advanced Photocatalysis Techniques
- Advanced Wireless Network Optimization
- Advancements in Battery Materials
- Advanced Cellulose Research Studies
- Multimodal Machine Learning Applications
- Infrastructure Resilience and Vulnerability Analysis
- MXene and MAX Phase Materials
- Advanced battery technologies research
- Polyoxometalates: Synthesis and Applications
- Text and Document Classification Technologies
- Network Security and Intrusion Detection
- Information and Cyber Security
- Caching and Content Delivery
- Supercapacitor Materials and Fabrication
- Mobile Ad Hoc Networks
- Catalysis and Hydrodesulfurization Studies
- Advanced Graph Neural Networks
- Speech Recognition and Synthesis
- Software Engineering Research
- Misinformation and Its Impacts
Fudan University
2013-2024
University of California, Los Angeles
2023
Nanchang Hangkong University
2022
People's Liberation Army 401 Hospital
2021
Meta (Israel)
2019-2021
Massachusetts Institute of Technology
2021
Seattle University
2020
The Ohio State University
2017-2020
Collaborative Innovation Center of Chemistry for Energy Materials
2013-2016
Shanghai Jiao Tong University
2014-2016
A highly active and stable electrochemical catalyst of nanoporous molybdenum carbide nanowires (np-Mo2C NWs) has been developed for hydrogen evolution reaction (HER). The np-Mo2C NWs were synthesized simply by pyrolysis a MoOx/amine hybrid precursor with sub-nanosized periodic structure under an inert atmosphere. enriched nanoporosity large reactive surface these dispersed uniform Mo2C nanocrystallites provide efficient electrocatalysis, leading to their superior HER activity lower onset...
Large transformer models have shown extraordinary success in achieving state-of-the-art results many natural language processing applications. However, training and deploying these can be prohibitively costly for long sequences, as the standard self-attention mechanism of Transformer uses $O(n^2)$ time space with respect to sequence length. In this paper, we demonstrate that approximated by a low-rank matrix. We further exploit finding propose new mechanism, which reduces overall complexity...
Hierarchical MoS2/polyaniline nanowires, integrating MoS2 nanosheets with conductive polyaniline, serve as prominent anode materials for Li-ion batteries, presenting high capacity and good cyclability. The polyaniline-hybrid structure hierarchical features significantly promote the Li-storage performance compared bare MoS2, indicating new opportunities developing electrode nanomaterials. As a service to our authors readers, this journal provides supporting information supplied by authors....
MoC–Mo<sub>2</sub>C heteronanowires accomplished <italic>via</italic> controlled carbonization are efficient in the hydrogen evolution reaction due to a synergistic enhancement.
Pre-trained language models have proven their unique powers in capturing implicit features. However, most pre-training approaches focus on the word-level training objective, while sentence-level objectives are rarely studied. In this paper, we propose Contrastive LEArning for sentence Representation (CLEAR), which employs multiple augmentation strategies order to learn a noise-invariant representation. These augmentations include word and span deletion, reordering, substitution. Furthermore,...
We present BlockBERT, a lightweight and efficient BERT model for better modeling long-distance dependencies. Our extends by introducing sparse block structures into the attention matrix to reduce both memory consumption training/inference time, which also enables heads capture either short- or long-range contextual information. conduct experiments on language pre-training several benchmark question answering datasets with various paragraph lengths. BlockBERT uses 18.7-36.1% less 12.0-25.1%...
The exposure of rich active sites is crucial for MoS2 nanocatalysts in efficient hydrogen evolution reaction (HER). However, the (010) and (100) planes tend to vanish during preparation because their high surface energy. Employing protection by thiourea (TU) reactant, a microwave-assisted reactant-protecting strategy successfully introduced fabricate active-site-rich (AS-rich MoS2). bifunctionality TU, as both reactant capping agent, ensures interactions effective easy MoS2, avoiding...
Large pre-trained language models (LMs) have demonstrated remarkable ability as few-shot learners. However, their success hinges largely on scaling model parameters to a degree that makes it challenging train and serve. In this paper, we propose new approach, named EFL, can turn small LMs into better The key idea of approach is reformulate potential NLP task an entailment one, then fine-tune the with little 8 examples. We further demonstrate our proposed method be: (i) naturally combined...
In a large-scale and distributed matrix multiplication problem $C=A^{\intercal}B$, where $C\in\mathbb{R}^{r\times t}$, the coded computation plays an important role to effectively deal with "stragglers" (distributed computations that may get delayed due few slow or faulty processors). However, existing schemes could destroy significant sparsity exists in machine learning problems, result much higher overhead, i.e., $O(rt)$ decoding time. this paper, we develop new strategy, call \emph{sparse...
The diverse structures of molybdate anions significantly provide new opportunities to design various nanostructures MoOx-based organic–inorganic hybrids with prominent catalytic, electrochemical and photo/electrochromic properties. In this paper, the one-dimensional (1D) growth originating from anisotropic is successfully introduced prepare a series hybrid nanowires Mo3O10(C6H8N)2·2H2O (anilinium trimolybdate), Mo3O10(C2H10N2) (ethylenediamine trimolybdate) Mo3O10(C5H6N)2·H2O (pyridium...
Mo2C nanowires and composites of nanoparticles formed on multiwalled carbon nanotubes (Mo2C/CNT) were developed as advanced catalysts for hydrogen evolution at a polarised water–1,2-dichloroethane interface. Each catalyst acts catalytic nano-raft suspended the interface to markedly enhance rates biphasic proton reduction in presence an organic solubilised electron donor, decamethylferrocene. grown situ conductive CNT support, achieving high dispersion intimate contact, thereby facilitating...
The quadratic computational and memory complexities of the Transformer's attention mechanism have limited its scalability for modeling long sequences. In this paper, we propose Luna, a linear unified nested that approximates softmax with two functions, yielding only (as opposed to quadratic) time space complexity. Specifically, first function, Luna packs input sequence into fixed length. Then, packed is unpacked using second function. As compared more traditional mechanism, introduces an...
Remarkable progress has been made in developing efficient catalysts of metal carbides, nitrides, sulfides and selenides based on organic–inorganic nanohybrids.
Qifan Wang, Jingang Xiaojun Quan, Fuli Feng, Zenglin Xu, Shaoliang Nie, Sinong Madian Khabsa, Hamed Firooz, Dongfang Liu. Proceedings of the 61st Annual Meeting Association for Computational Linguistics (Volume 1: Long Papers). 2023.
Abstract A novel chemical oxidative polymerization approach has been proposed for the controllable preparation of organic–inorganic hybrid MoO x /polyaniline (PANI) nanocomposites based on nanowire precursor Mo 3 O 10 (C 6 H 8 N) 2 ⋅2 with sub‐nanometer periodic structures. The nanotubes, nanowires, and rambutan‐like nanoparticles /PANI were successfully obtained through simply modulating pH values to 2.5–3.5, ≈2.0 ≈1.0, respectively. Through systematic physicochemical characterization, such...
The rapid growth of data volume and the accompanying congestion problems over wireless networks have been critical issues to content providers. A novel technique, termed as coded cache, is proposed relieve burden. Through creating coded-multicasting opportunities, coded-cache scheme can provide extra performance gain conventional push technique that simply pre-stores contents at local caches during network idle period. But existing works on caching assumed availability an error-free shared...
In large scale distributed linear transform problems, coded computation plays an important role to effectively deal with "stragglers" (distributed computations that may get delayed due few slow or faulty processors). We propose a strategy, referred as diagonal code, achieves the optimum recovery threshold and load. This is first code simultaneously two-fold optimality in transforms. Furthermore, by leveraging idea from random proposal graph theory, we design two codes can guarantee high...
Recent work has suggested that language models (LMs) store both common-sense and factual knowledge learned from pre-training data. In this paper, we leverage implicit to create an effective end-to-end fact checker using a solely model, without any external or explicit retrieval components. While previous on extracting LMs have focused the task of open-domain question answering, best our knowledge, is first examine use as checkers. closed-book setting, show zero-shot LM approach outperforms...
In the distributed graident coding problem, it has been established that, to exactly recover gradient under s slow machines, mmimum computation load (number of stored data partitions) each worker is at least linear ($s+1$), which incurs a large overhead when large~\citetandon2017gradient. this paper, we focus on approximate that aims with bounded error ε. Theoretically, our main contributions are three-fold: (i) analyze structure optimal codes, and derive information-theoretical lower bound...