The complex submarine geography, variable ocean current dynamics, and challenging underwater communication conditions present significant technical challenges to the implementation of data collection systems for autonomous underwater vehicles (AUVs), which are crucial for the deployment of Internet of Underwater Things (IoUT) applications. Multi agent reinforcement learning (MARL), in which agents interact with their environment to obtain rewards by joint trial-and-error, has been thoroughly researched for the cooperation of intelligent vehicles. Nevertheless, the conventional MARL approaches for cooperative AUVs are challenging to apply due to the difficulties in information exchange and the time-varying submarine environment. Accordingly, this study proposes a novel deep reinforcement learning (DRL) approach named staged learning with distributed negotiation (SLDN). Compare to prevalent MARL approaches like centralized training with distribute execution (CTDE), the proposed method relies on staged and independent DRL and accomplishes multi-AUV cooperation tasks through staged negotiation and training, aiming to facilitate collaborative operations in dynamic underwater environments while significantly reducing communication costs. This method addresses the IoUT scenario of a cluster-structured hybrid underwater network that integrates acoustic and MI communications, and its superiority in terms of data collection capabilities is confirmed by simulations and comparative analysis.

Formation path planning of autonomous underwater vehicles (AUVs) entails establishing optimal collision-free routes over challenging underwater terrain while maintaining state coherence to preserve an intended formation, and path planning techniques have been the subject of significant study over the last decade, with swarm intelligence algorithms such as the sparrow search algorithm (SSA) being among the most commonly employed. However, the algorithms typically are constrained by the imbalanced adjustment between local development and global exploration, which reduces the optimization capability, and they are relatively understudied for the formation movement issues. Accordingly, this paper proposes a consensus-SSA based formation path planning (CSFPP) method, which applies an improved SSA for planning an optimal path, and then incorporates the path into a consensus algorithm that introduces an artificial potential field (APF) to enable collaborative formation movement. In the path planning phrase, the CSFPP employs an improved SSA which applies the golden search optimization (GSO) and an adaptive iteration approach to adjust the local development and global exploration in order to improve the overall optimization performance. Then in the formation control phrase, the CSFPP introduces a virtual point scheme for APF-based obstacle avoidance in order to navigate an AUV formation in an obstacle environment while maintaining the formation shape controlled by a consensus algorithm. The superiority of the proposed path planning capability is demonstrated by comparing the convergence performance of the improved SSA with the recent contributions; and simulations of formation movement in underwater space verify the feasibility of the proposed formation control method in the obstacle environment.

Intelligent control of autonomous marine vehicles (AMVs) is one of the essential technologies for exploring marine resources. In the deep sea with a complicated exploration environment, collaboration between heterogeneous AMVs can maximize exploration efficiency by utilizing various functional benefits. Accordingly, this article proposes a method for collaborative path planning for heterogeneous AMVs that employs a fused metaheuristic algorithm for the underwater path planning of an autonomous underwater glider (AUG), and an adaptive surface path planning of an autonomous surface vehicle (ASV), respectively. The fused metaheuristic method balances global and local path explorations for underwater path planning by integrating the gray wolf optimizer (GWO) and equilibrium optimizer (EO), and it reduces the local optimum problem by using a conditional convergence factor; and the adaptive surface path planning approach considers the influence of ocean currents at various locations to guide the ASV collaboratively to track the AUG underwater in the horizontal plane. The fused metaheuristic algorithm has demonstrated superior convergence performance in simulations, which indicates that the proposed method has advantage in terms of underwater path planning for complex marine exploration.

This paper focuses on using natural language descriptions to enhance predictive models in the chemistry field.Conventionally, chemoinformatics models are trained with extensive structured data manually extracted from the literature.In this paper, we introduce TextReact, a novel method that directly augments predictive chemistry with texts retrieved from the literature.TextReact retrieves text descriptions relevant for a given chemical reaction, and then aligns them with the molecular representation of the reaction.This alignment is enhanced via an auxiliary masked LM objective incorporated in the predictor training.We empirically validate the framework on two chemistry tasks: reaction condition recommendation and onestep retrosynthesis.By leveraging text retrieval, TextReact significantly outperforms state-ofthe-art chemoinformatics models trained solely on molecular data.

Underwater acoustic communication networks (UACNs) have been widely utilized in recent years because of the growing interest in interactive information in the deep ocean. Compared with traditional radio wireless networks, UACNs are characterized by complex and dynamic 3-D network topology and longer signal propagation delay. Therefore, recent studies on UACNs usually apply dynamic routes in data communication to adapt to complex UACN structure. However, the approaches are hardly adequate for UACN scenarios with high-traffic requirements. This is because dynamic routing methods, such as opportunistic routing, usually require external contention costs to control the routing paths, which leads to decreased network throughput. Accordingly, this article focuses on enabling high-speed acoustic communications for underwater application scenarios with high-traffic requirements, and proposes an underwater data transmission method using the multichannel full-duplex (FD) communication technique. The proposed method applies the underwater orthogonal frequency division multiple access (OFDM) technique, that uses collision-free channels to relay data in multihop routes for simultaneous transmission and reception. Unlike the traditional communication methods of UACNs with dynamic routing, the proposed one uses static routes when transporting data over hop-by-hop paths to achieve stable and uninterrupted FD communication. The approach also includes a directional forwarding-based route request method and an elimination-based channel evaluation proposal, which purpose to reduce the overheads from exploring routes and enable collision-free FD communications. The performance analysis of the proposed method is under simulated conditions of high-traffic UACN scenarios, and the results show that it has superior performance compare to the classical underwater data transmission methods.

Zhengxiao Du, Yujie Qian, Xiao Liu, Ming Ding, Jiezhong Qiu, Zhilin Yang, Jie Tang. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Yanan Zheng, Jing Zhou, Yujie Qian, Ming Ding, Chonghua Liao, Li Jian, Ruslan Salakhutdinov, Jie Tang, Sebastian Ruder, Zhilin Yang. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

The coming 6G communication technology introduces the possibility of practice underwater Internet of Things (UIoT) applications with high-speed and reliable underwater communications. Among them, the cooperative coverage path planning (CPP) with autonomous underwater vehicles (AUVs) is a promising approach for enabling deep ocean exploration. The cooperative CPP in underwater is challenged by several marine factors, typically the harsh underwater communication environment, which brings difficulties in sharing the coverage progresses of AUVs and the environmental information such as the seafloor bathymetry, obstacles, etc. Accordingly, this paper proposes a novel CPP method based on 6G enabled cooperative AUVs, named the Pac-AUV. As the name suggests, the method is based on the mechanism of the Ms. Pac-Man game, where AUV-assisted subsea exploration is considered as cooperative Pac-Men sharing the Pac-Dots distributed on the seafloor. The Pac-AUV involves two steps in cooperative CPP. One is a dot-spreading-based mission assignment (DMA), which is discretely performed by each AUV and requires support from reliable underwater communication in sharing the Pac-Dots. The other step is virtual attraction-based coverage path planning (V-CPP), which adopts the virtual attraction force from the Pac-Dots, results in low computational complexities in generating the coverage paths and avoids obstacles. Simulations are performed to demonstrate the performance of the Pac-AUV, and the results prove the advantages of cooperative and balanced CPP executions.

Underwater acoustic sensor networks (UASNs) have been proposed as a promising way in supporting the Underwater-Internet-of-Things (UIoT) applications. However, to guarantee and improve Quality of Service (QoS), they are still facing great challenges especially when it comes to enabling reliable data communication for the UIoT applications and meanwhile protecting the marine ecosystems for sustainable underwater monitoring and exploration; this is because the acoustic signals used by UASNs can do harmful interference to vocalizing marine mammals. Therefore, the article proposed an ecologically friendly data transmission scheme for UASNs, which adopts an interference-aware opportunistic route discovery method and a frequency-division multiplexing (FDM)-based full-duplex communication scheme. First, in the route discovery phase, a virtual-void zone scheme, an FDM-based channel allocation approach and a Bayesian network-based mammal avoidance strategy is introduced to establish interference-free routing paths for enabling reliable and environmentally friendly data transmissions. Then during the data transmission phase, an FDM-based full-duplex communication technique is adopted to enable high-speed data flow from the seabed to the surface. Extensive simulations indicate that the proposed scheme has excellent advantages in terms of QoS while also taking marine mammals into account since the signal interference to both underwater nodes and nearby vocalizing mammals is significantly mitigated.

Escorting a moving object in a subsea environment with cooperative autonomous underwater vehicles (AUVs) is a typical subject in Underwater Internet of Things (UIoT) applications. It involves two issues that should be studied. First, a mobile task assignment method is required to lead the AUVs to the escorting positions; then, a formation control scheme should be utilized to safely escort the moving object to the destination. Accordingly, in this article, a comprehensive target escorting mechanism called the cooperative-control-based underwater target estimating mechanism (CUTE) is proposed, which includes a belief-function-method-based self-organizing map algorithm for task assignment and an artificial potential field-based formation control method. The task assignment method aims to establish smooth routes from the AUVs to the escort position while flexibly avoiding obstacles, and the formation control method aims to improve the monitoring coverage on the escort route by rotating the formation structure while following the moving object. Simulations show that the proposed CUTE method may be very practical in underwater target escorting scenarios.

This paper introduces how to infer trust relationships from billion-scale networked data to benefit Alibaba E-Commerce business. To effectively leverage the network correlations between labeled and unlabeled relationships to predict trust relationships, we formalize trust into multiple types and propose a graphical model to incorporate type-based dyadic and triadic correlations, namely eTrust. We also present a fast learning algorithm in order to handle billion-scale networks. Systematically, we evaluate the proposed methods on four different genres of datasets with labeled trust relationships: Alibaba, Epinions, Ciao, and Advogato. Experimental results show that the proposed methods achieve significantly better performance than several comparison methods (+1.7-32.3% by accuracy; p <; <; 0:01, with t-test). Most importantly, when handling the real large networked data with over 1,200,000,000 edges (Ali-large), our method achieves 2,000× speedup to infer trust relationships, comparing with the traditional graph learning algorithms. Finally, we have applied the inferred trust relationships to Alibaba E-commerce platform: Taobao, and achieved 2.75 percent improvement on gross merchandise volume (GMV).

In online question-and-answer (QA) websites like Quora, one central issue is to find (invite) users who are able to provide answers to a given question and at the same time would be unlikely to say "no" to the invitation. The challenge is how to trade off the matching degree between users’ expertise and the question topic, and the likelihood of positive response from the invited users. In this paper, we formally formulate the problem and develop a weakly supervised factor graph (WeakFG) model to address the problem. The model explicitly captures expertise matching degree between questions and users. To model the likelihood that an invited user is willing to answer a specific question, we incorporate a set of correlations based on social identity theory into the WeakFG model. We use two different genres of datasets: QA-Expert and Paper-Reviewer, to validate the proposed model. Our experimental results show that the proposed model can significantly outperform (+1.5-10.7% by MAP) the state-of-the-art algorithms for matching users (experts) with community questions. We have also developed an online system to further demonstrate the advantages of the proposed method.

This paper presents PLoRa, an ambient backscatter design that enables long-range wireless connectivity for batteryless IoT devices. PLoRa takes ambient LoRa transmissions as the excitation signals, conveys data by modulating an excitation signal into a new standard LoRa "chirp" signal, and shifts this new signal to a different LoRa channel to be received at a gateway faraway. PLoRa achieves this by a holistic RF front-end hardware and software design, including a low-power packet detection circuit, a blind chirp modulation algorithm and a low-power energy management circuit. To form a complete ambient LoRa backscatter network, we integrate a light-weight backscatter signal decoding algorithm with a MAC-layer protocol that work together to make coexistence of PLoRa tags and active LoRa nodes possible in the network. We prototype PLoRa on a four-layer printed circuit board, and test it in various outdoor and indoor environments. Our experimental results demonstrate that our prototype PCB PLoRa tag can backscatter an ambient LoRa transmission sent from a nearby LoRa node (20 cm away) to a gateway up to 1.1 km away, and deliver 284 bytes data every 24 minutes indoors, or every 17 minutes outdoors. We also simulate a 28-nm low-power FPGA based prototype whose digital baseband processor achieves 220 μW power consumption.