The use of a hydrate-based technology in seawater desalination is an interesting potential hydrate application since salt ions would be excluded from the hydrate crystal lattice. In order to better understand the hydrate-based desalination process, experiments have been conducted using cyclopentane (CyC5, sII) hydrates, which can be formed at atmospheric pressure and temperatures below 7.7 °C. The hydrate formation experiments were performed at various subcoolings for aqueous solutions with different salinities in a bubble column. The hydrate formation times decreased and the hydrate conversion increased with increasing subcooling and agitation. Various hydrate-former injection methods were studied, with the most effective method involving spraying finely dispersed CyC5 droplets (around 5 μm in diameter) into the water-filled bubble column. The latter method resulted in a 2-fold increase in seawater conversion to hydrate crystals compared with injecting millimeter-scale CyC5 droplets. A desalination efficiency of 81% (the salinity decreased from 3.5 to 0.67 wt %) was achieved by using a three-step separation method, including gravitational separation, filtration, and a washing step. Washing the hydrate sample using filtered water decreased the salinity from 1.5 wt % in the solid hydrates before washing to 1.05 wt % after washing.

Machine learning (ML) and deep learning (DL) are a subset of artificial intelligence (AI) that can automatically learn from data and can perform tasks such as predictions and decisionmaking. Interdisciplinary studies combining ML/DL with chemical health and safety have demonstrated their unparalleled advantages in identifying trend and prediction assistance, which can greatly save manpower, material resources, and financial resources. In this Review, commonly used ML/DL tools and concepts as well as popular ML/DL algorithms are introduced and discussed. More than 100 papers have been categorized and summarized to present the current development of ML/DL-based research in the area of chemical health and safety. In addition, the limitation of current studies and prospects of ML/DL-based study are also discussed. This Review can serve as useful guidance for researchers who are interested in implementing ML/DL into chemical health and safety research and for readers who try to learn more information about novel ML/DL techniques and applications.

The Transformer translation model employs residual connection and layer normalization to ease the optimization difficulties caused by its multi-layer encoder/decoder structure. Previous research shows that even with residual connection and layer normalization, deep Transformers still have difficulty in training, and particularly Transformer models with more than 12 encoder/decoder layers fail to converge. In this paper, we first empirically demonstrate that a simple modification made in the official implementation, which changes the computation order of residual connection and layer normalization, can significantly ease the optimization of deep Transformers. We then compare the subtle differences in computation order in considerable detail, and present a parameter initialization method that leverages the Lipschitz constraint on the initialization of Transformer parameters that effectively ensures training convergence. In contrast to findings in previous research we further demonstrate that with Lipschitz parameter initialization, deep Transformers with the original computation order can converge, and obtain significant BLEU improvements with up to 24 layers. In contrast to previous research which focuses on deep encoders, our approach additionally enables Transformers to also benefit from deep decoders.

The titled diblock copolymers are synthesized via cobalt-catalyzed living carbonylative polymerization of N-alkylaziridines under moderate pressures followed by a deprotection step. The poly(beta-alanine) block is solubilized by the poly(beta-alanoid) block in chloroform and remains fully hydrogen-bonded in the form of a sheet-like assembly.

Existing Neural Machine Translation (NMT) systems are generally trained on a large amount of sentence-level parallel data, and during prediction sentences are independently translated, ignoring cross-sentence contextual information. This leads to inconsistency between translated sentences. In order to address this issue, context-aware models have been proposed. However, document-level parallel data constitutes only a small part of the parallel data available, and many approaches build context-aware models based on a pre-trained frozen sentence-level translation model in a two-step training manner. The computational cost of these approaches is usually high. In this paper, we propose to make the most of layers pre-trained on sentence-level data in contextual representation learning, reusing representations from the sentence-level Transformer and significantly reducing the cost of incorporating contexts in translation. We find that representations from shallow layers of a pre-trained sentence-level encoder play a vital role in source context encoding, and propose to perform source context encoding upon weighted combinations of pre-trained encoder layers' outputs. Instead of separately performing source context and input encoding, we propose to iteratively and jointly encode the source input and its contexts and to generate input-aware context representations with a cross-attention layer and a gating mechanism, which resets irrelevant information in context encoding. Our context-aware Transformer model outperforms the recent CADec [Voita et al., 2019c] on the English-Russian subtitle data and is about twice as fast in training and decoding.

Controlling colloidal self-assemblies using external forces is essential to develop modern electrooptical and biomedical devices. Importantly, shape anisotropic colloids can provide optical properties such as birefringence. Here we demonstrate that external temperature gradients can be effective in controlling nematic liquid crystalline (LC) order in suspensions of plate-like colloids also known as nanoplates. Nanoplates, in an isotropic suspension, wherein their orientations are random, could be effectively moved using a temperature gradient environment causing a phase transition to LC nematic phase. Such controllably formed nematic phase featured large nematic monodomains and enabled topologically more stable structures that were evident from the absence of hedgehog-type defects which are typically found in nematics formed spontaneously via nucleation and growth mechanism in a sufficiently high concentration suspension of nanoplates. Due to their high surface area-to-volume ratio and excellent thermophoretic properties, nanoplates can prove to be ideal candidates for transport of biomolecules through temperature varying environments.

Catalytic ring-expanding carbonylation of 2-aryl-2-oxazolines is reported as a novel method for the synthesis of 4,5-dihydro-1,3-oxazin-6-ones. Various observations suggest the involvement of cobalt radicals as the catalytically active species. [reaction: see text]

No abstract