In recent years, there has been a dramatic rise in interest in retrosynthesis prediction with artificial intelligence (AI) techniques. Unlike conventional retrosynthesis prediction performed by chemists and by rule-based expert systems, AI-driven retrosynthesis prediction automatically learns chemistry knowledge from off-the-shelf experimental datasets to predict reactions and retrosynthesis routes. This provides an opportunity to address many conventional challenges, including heavy reliance on extensive expertise, the sub-optimality of routes, and prohibitive computational cost. This review describes the current landscape of AI-driven retrosynthesis prediction. We first discuss formal definitions of the retrosynthesis problem and review the outstanding research challenges therein. We then review the related AI techniques and recent progress that enable retrosynthesis prediction. Moreover, we propose a novel landscape that provides a comprehensive categorization of different retrosynthesis prediction components and survey how AI reshapes each component. We conclude by discussing promising areas for future research.
Pre-trained language models have achieved striking success in natural language processing (NLP), leading to a paradigm shift from supervised learning to pre-training followed by fine-tuning. The NLP community has witnessed a surge of research interest in improving pre-trained models. This article presents a comprehensive review of representative work and recent progress in the NLP field and introduces the taxonomy of pre-trained models. We first give a brief introduction of pre-trained models, followed by characteristic methods and frameworks. We then introduce and analyze the impact and challenges of pre-trained models and their downstream applications. Finally, we briefly conclude and address future research directions in this field.
In recent years, data has become one of the most important resources in the digital economy. Unlike traditional resources, the digital nature of data makes it difficult to value and contract. Therefore, establishing an efficient and standard data-transaction market system would be beneficial for lowering cost and improving productivity among the parties in this industry. Although numerous studies have been dedicated to the issue of complying with data regulations and other data-transaction issues such as privacy and pricing, little work has been done to provide a comprehensive review of these studies in the fields of machine learning and data science. To provide a complete and up-to-date understanding of this topic, this review covers the three key issues of data transaction: data rights, data pricing, and privacy computing. By connecting these topics, this paper provides a big picture of a data ecosystem in which data is generated by data subjects such as individuals, research agencies, and governments, while data processors acquire data for innovational or operational purposes, and benefits are allocated according to the data's respective ownership via an appropriate price. With the long-term goal of making artificial intelligence (AI) beneficial to human society, AI algorithms will then be assessed by data protection regulations (i.e., privacy protection regulations) to help build trustworthy AI systems for daily life.
In this article, we propose a communicative learning (CL) formalism that unifies existing machine learning paradigms, such as passive learning, active learning, algorithmic teaching, and so forth, and facilitates the development of new learning methods. Arising from human cooperative communication, this formalism poses learning as a communicative process and combines pedagogy with the burgeoning field of machine learning. The pedagogical insight facilitates the adoption of alternative information sources in machine learning besides randomly sampled data, such as intentional messages given by a helpful teacher. More specifically, in CL, a teacher and a student exchange information with each other collaboratively to transmit and acquire certain knowledge. Each agent has a mind, which includes the agent's knowledge, utility, and mental dynamics. To establish effective communication, each agent also needs an estimation of its partner's mind. We define expressive mental representations and learning formulation sufficient for such recursive modeling, which endows CL with human-comparable learning efficiency. We demonstrate the application of CL to several prototypical collaboration tasks and illustrate that this formalism allows learning protocols to go beyond Shannon's communication limit. Finally, we present our contribution to the foundations of learning by putting forth hierarchies in learning and defining the halting problem of learning.
Sensing and understanding large-scale dynamic scenes require a high-performance imaging system. Conventional imaging systems pursue higher capability by simply increasing the pixel resolution via stitching cameras at the expense of a bulky system. Moreover, they strictly follow the feedforward pathway: that is, their pixel-level sensing is independent of semantic understanding. Differently, a human visual system owns superiority with both feedforward and feedback pathways: The feedforward pathway extracts object representation (referred to as memory engram) from visual inputs, while, in the feedback pathway, the associated engram is reactivated to generate hypotheses about an object. Inspired by this, we propose a dual-pathway imaging mechanism, called engram-driven videography. We start by abstracting the holistic representation of the scene, which is associated bidirectionally with local details, driven by an instance-level engram. Technically, the entire system works by alternating between the excitation–inhibition and association states. In the former state, pixel-level details become dynamically consolidated or inhibited to strengthen the instance-level engram. In the association state, the spatially and temporally consistent content becomes synthesized driven by its engram for outstanding videography quality of future scenes. The association state serves as the imaging of future scenes by synthesizing spatially and temporally consistent content driven by its engram. Results of extensive simulations and experiments demonstrate that the proposed system revolutionizes the conventional videography paradigm and shows great potential for videography of large-scale scenes with multi-objects.
In digital cameras, we find a major limitation: the image and video form inherited from a film camera obstructs it from capturing the rapidly changing photonic world. Here, we present vidar, a bit sequence array where each bit represents whether the accumulation of photons has reached a threshold, to record and reconstruct the scene radiance at any moment. By employing only consumer-level complementary metal-oxide-semiconductor (CMOS) sensors and integrated circuits, we have developed a vidar camera that is 1000× faster than conventional cameras. By treating vidar as spike trains in biological vision, we have further developed a spiking neural network (SNN)-based machine vision system that combines the speed of the machine and the mechanism of biological vision, achieving high-speed object detection and tracking 1000× faster than human vision. We demonstrate the utility of the vidar camera and the super vision system in an assistant referee and target pointing system. Our study is expected to fundamentally revolutionize the image and video concepts and related industries, including photography, movies, and visual media, and to unseal a new SNN-enabled speed-free machine vision era.
Wound healing has always been a focus of clinical study due to its universality, difficult treatment, large number of patients, and heavy medical burden. A great deal of effort has been devoted to generating various wound dressings with special features and functions to satisfy specific demands. Here, we present novel bio-printed textiles based on fish skin decellularized extracellular matrix (dECM) for wound healing. Thanks to the desirable biocompatibility of the fish-derived dECM, the bio-printed textiles exhibit excellent performance in terms of cell adherence and proliferation. Since the dECM-based hydrogels are generated using a bio-printing method, the bio-printed textiles exhibit an adjustable porous structure with good air permeability throughout the whole textile. Moreover, the high specific surface areas of the porous structure on the hydrogel skeleton make it possible to load a variety of active molecules to improve the wound healing effect. According to an in vivo study, we demonstrate that the prepared textiles loaded with the active drug molecules curcumin (Cur) and basic fibroblast growth factor (bFGF) can significantly speed up the chronic wound healing process. These remarkable properties indicate the potential value of fish-skin-dECM textiles in wound healing and biomedical engineering.
n-Type silicon (n-Si), with surface easily oxidized and passivated in an aqueous electrolyte, has suffered from sluggish oxygen evolution reaction (OER) kinetics for photoelectrochemical (PEC) water splitting. Herein, a trimetallic Ni0.9Fe0.05Co0.05 protective layer is successfully electrodeposited on a p+n-Si substrate by underpotential deposition. The prepared Ni0.9Fe0.05Co0.05/p+n-Si photoanode exhibits excellent stability and activity for PEC water oxidation, with a low onset potential of 0.938 V versus a reversible hydrogen electrode (RHE) and a remarkable photocurrent density of about 33.1 mA∙cm-2 at 1.23 V versus RHE, which significantly outperforms the Ni/p+n-Si photoanode as a reference. It is revealed that the incorporation of Fe into the Ni layer creates a large band bending at the Ni0.9Fe0.05Co0.05/p+n-Si interface, promoting interfacial charge separation. Moreover, the incorporation of Co produces abundant Ni3+ and oxygen vacancies (Ov) that act as active sites to accelerate the OER kinetics, synergistically contributing to a major enhancement of PEC water oxidation activity. Encouragingly, by connecting the Ni0.9Fe0.05Co0.05/p+n-Si photoanode to an inexpensive Si solar cell, an integrated photovoltaic/PEC (PV/PEC) device achieved a solar-to-hydrogen conversion efficiency of as high as 12.0% without bias. This work provides a facile approach to design efficient and stable n-Si-based photoanodes with a deep understanding of the structure–activity relationship, which exhibits great potential for the integration of low-cost PV/PEC devices for unassisted solar-driven water splitting.
It is well known that heat is generated when an electric current passes through an electrical conductor. While various applications in our daily lives and industries utilize the heating of electronic conductors, little attention has been paid to the potential of ionic conductors for heating purposes. This is because of the “inevitable” electrochemical reactions, which can result in unwanted electrolysis of conductors, corrosion of electrodes, and surface fouling. This paper reports the Joule heating of ionic conductors without electrochemical reactions. Electricity with a zero-phase frequency is employed to suppress the electrolysis of ionic conductors at high voltages. Demonstrations with various ionic conductors, both liquids and solids, show highly efficient energy conversion free of electrochemical reactions. This heating method is simple, direct, fast, clean, and uniform, and it has great potential in numerous industrial and household applications.
Since 1998, the Au–O–Ti4+ sites of Au/Ti-based catalysts have been widely accepted as the active sites for propene epoxidation with H2 and O2 at a relatively high temperature, although they are limited by poor H2 efficiency. Herein, we demonstrate a novel Au–O–Ti3+ active site aiming at low-temperature propene epoxidation. Notably, this active site results in a sharp shift in the optimum temperature, from 200 to 138 °C, and allows the catalyst to maintain an unprecedented H2 efficiency of 43.6%, a high propylene oxide (PO) selectivity of 90.7%, and a stability of over 100 h. The Au–O–coordinatively unsaturated Ti3+ active site is quantitively constructed by tuning the amount of Si–OH and Bu3NH+ in post-treated silicalite-1 seeds. Through operando ultraviolet–visible (UV–vis) spectroscopy, the dynamic evolution of the Ti–OOH intermediate was investigated. It was found that the Ti–OOH generation rate is higher on Au–O–Ti3+ than on conventional Au–O–Ti4+ sites. Moreover, ammonia temperature-programmed desorption (NH3-TPD) and X-ray photoelectron spectroscopy (XPS) characterizations, together with density-functional theory (DFT) calculations, demonstrated that the coordinatively unsaturated Ti3+ sites promote electron transfer between Au and Ti3+, thereby enhancing the O2 adsorption ability of the catalyst and promoting the in situ formation of H2O2 and the Ti–OOH intermediate, even at a low temperature. The insights and methodology reported here not only shed new light on maximizing H2 efficiency over a coordinatively unsaturated Ti3+ structure of titanium silicate-1 but also open up new opportunities for industrial direct gas-phase propene epoxidation in a low temperature range.
The development of microbial electrosynthesis (MES) for renewable electricity-driven bioutilization of CO2 has recently attracted considerable interest due to its ability to synthesize chemicals with the transition towards a circular carbon economy. However, the increase of acetate production and the decrease of energy consumption of MES using an advanced reactor design have received less attention. In this study, the total acetate production rate using novel flow-electrode MES reactors ((16 ± 1) g·m−2·d−1) was double that using reactors without powder activated carbon (PAC) amendment ((8 ± 3) g·m−2·d−1). The flow-electrode MES reactors had a Coulombic efficiency of 43.5% ± 3.1%, an energy consumption of (0.020 ± 0.005) kW·h·g−1, and an energy efficiency of 18.7% ± 1.3% during acetate production. The flow-electrode with PAC amendment could decrease the net water flux and charge transfer resistance, while had little impact on the cell voltage, rheological behavior, and acetate adsorption. In the flow-electrode MES reactors, the expression of genes involving in energy production and conversion were increased, and the increase of acetate production was found correlated with the increased abundance of Acetobacterium. The Wood–Ljungdahl pathway (WLP) and reductive citric acid cycle (rTCA) were found to be the complete pathways responsible for carbon fixation. The concentrations of acetate in the stacked flow-electrode MES reached 7.0 g·L−1. This study presents a new approach for the construction of scalable MES reactors with high-performance chemical generation and CO2 utilization.
Deep underground energy storage is the use of deep underground spaces for large-scale energy storage, which is an important way to provide a stable supply of clean energy, enable a strategic petroleum reserve, and promote the peak shaving of natural gas. Rock salt formations are ideal geological media for large-scale energy storage, and China is rich in salt rock resources and has a major shortage of energy storage space. Compared with the salt domes in other countries, the salt rock formations in China are typical lacustrine bedded salt rocks characterized by thin beds, high impurity content, and many interlayers. The development of large-scale energy storage in such salt formations presents scientific and technical challenges, including: ① developing a multiscale progressive failure and characterization method for the rock mass around an energy storage cavern, considering the effects of multifield and multiphase coupling; ② understanding the leakage evolution of large-scale deep underground energy storage caverns; ③ understanding the long-term performance evolution of large-scale deep underground energy storage caverns; ④ developing intelligent construction technologies for the deep underground salt caverns used for energy storage; and ⑤ ensuring the long-term function of deep underground energy storage spaces. The solution to these key scientific and technological problems lies in establishing a theoretical and technical foundation for the development of large-scale deep underground energy storage in China.
As valuable land in the ocean, coral islands are not only important bases for making use of marine resources and protecting marine rights and interests, but also important for breakthrough research in many fields of earth science. Hence, the economical and efficient determination of the underground structure of coral islands has become significant in coral island engineering geology, but remains challenging for traditional marine geophysical prospecting and drilling methods. While ambient noise tomography with dense arrays has been widely used in continental regions, its applicability to coral islands remains undetermined. In this study, based on the data recorded by a dense array on an isolated coral island in the South China Sea, we analyzed the ambient noise characteristics and obtained a 3D subsurface structure of the coral island using ambient noise tomography. We made the following findings: ① The ambient noise frequencies can be roughly categorized into three levels: < 1, 1–5, and > 5 Hz. The spectral characteristics of the noise below 5 Hz were consistent at different stations, but there were significant differences in the characteristics of the noise above 5 Hz. ② For ambient noise frequencies below 5 Hz, cross-correlation functions with high quality could be obtained with only 24 h of waveform data. However, it was difficult to extract meaningful cross-correlation functions for ambient noise frequencies above 5 Hz. ③ The S-wave velocity in the coral island was higher toward the sea and lower toward the lagoon, which was consistent with the high degree of cementation of the outer reef flat stratum on the seaward side. ④ There were two low-velocity horizons at 25–75 and 200–300 m, which were in good agreement with the high-porosity horizons that were revealed by drilling core samples, reflecting the weathering history of the reef. Our research demonstrates that ambient noise tomography is a potentially economical, efficient, and environmentally friendly method for the geological prospecting of coral reefs.
Intelligent sensing, mechanism understanding, and the deterioration forecasting based on spatio–temporal big data not only promote the safety of the infrastructure but also indicate the basic theory and key technology for the infrastructure construction to turn to intelligentization. The advancement of underground space utilization has led to the development of three characteristics (deep, big, and clustered) that help shape a tridimensional urban layout. However, compared to buildings and bridges overground, the diseases and degradation that occur underground are more insidious and difficult to identify. Numerous challenges during the construction and service periods remain. To address this gap, this paper summarizes the existing methods and evaluates their strong points and weak points based on real-world space safety management. The key scientific issues, as well as solutions, are discussed in a unified intelligent monitoring system.
Monovalent cation exchange membranes (M-CEMs) have been extensively applied in environmental remediation and energy harvesting such as the extraction of Na+ or Li+ from brine and seawater. However, owing to the limitations of membrane structures and materials, M-CEMs have a low perm-selectivity issue. Herein, we proposed a facile approach to construct a novel M-CEM with a Janus-charged structure, consisting of a positively-charged trimesic acid/polyethylenimine surface thin layer and a negatively charged commercial cation exchange membrane (CEM). Selectrodialysis results indicated that the Janus-charged M-CEMs could effectively suppress the migration of anions, which often occurred in porous CEMs, thereby enabling the novel Janus-charged M-CEMs to possess high perm-selectivity and high total cation fluxes. Compared with state-of-the-art M-CEMs, the Janus-charged M-CEM exhibited the highest perm-selectivity of 145.77 for Na+/Mg2+ and a Na+ flux of 14.3 × 10−8 mol·cm−2·s−1 beyond the contemporary “upper bound” plot as well as the excellent perm-selectivity of 14.11 for Li+/Mg2+, indicating its great potentials in ion separation. This study can provide novel insights into the design of Janus-charged M-CEMs for ion separation in diverse environmental and energy applications.
In-situ photocatalytic H2O2 production has been receiving increasing attention in recent years for sustainable H2O2 synthesis. Graphitic carbon nitride (g-C3N4) is regarded as one of the most promising semiconductor photocatalysts for H2O2 evolution. Introducing N defects in g-C3N4 has been proved to be an effective strategy to enhance photocatalytic activity. However, the photocatalytic mechanism of the N vacancies is ambiguous and different types of N vacancies in g-C3N4 may exhibit different effects on photocatalytic activity. Herein, we develop a facile sodium persulfate eutectic polymerization method to prepare the g-C3N4 with abundant three coordinate nitrogen (N3C) vacancies. This type of nitrogen vacancy has not been studied in g-C3N4 for photocatalytic H2O2 production. Our results showed that the introduction of N3C vacancies in the g-C3N4 successfully broadened the light absorption range, inhibited the photoexcited charge recombination with enhanced O2 adsorption to promote oxygen activation. The photocatalytic H2O2 evolution from the N3C-rich g-C3N4 is 4.5 times higher than that of the pristine g-C3N4. This study demonstrates a novel strategy to introduce N3C vacancies in g-C3N4, which offers a new method to develop active catalysts for photocatalytic H2O2 evolution.
Recent studies have revealed the potency of probiotics in alleviating metabolic diseases associated with intestinal barrier dysfunction. However, an efficient model for screening probiotic strains against specific metabolic diseases has not been well developed. In the present study, a Caco-2 cell monolayer membrane model treated with tumor necrosis factor (TNF-α) or alcohol was used to evaluate the effect of 139 Lactobacillus strains on intestinal barrier function in vitro. We then selected 11 Lactobacillus strains with different regulatory abilities on the gut barrier to determine their effect against ovariectomy-induced osteoporosis or chronic alcoholic liver injury in vivo. Our results showed that the Pearson coefficient between the data of cell and animal models were 0.82 and −0.97 for the protection of probiotics against osteoporosis and alcoholic liver disease, respectively, suggesting the reliability of the cell model to simulate the in vivo protective effects of probiotics. This study established a potential in vitro approach based on a Caco-2 cell monolayer membrane model for the efficient screening of potential probiotics against specific metabolic diseases such as osteoporosis and chronic alcoholic liver disease.