Search

Tool-augmented large language models (LLMs) have achieved remarkable progress in tackling a broad range of queries. However, existing work are still in the experimental stage and has limitations in extensibility and robustness, especially facing the real-world applications. In this paper, we consider a more realistic scenario, connecting LLMs with RESTful APIs, which use the commonly adopted REST software architectural style for web service development. To address the practical challenges of planning and API usage, we introduce RestGPT, which leverages LLMs to solve user requests by connecting with RESTful APIs. Specifically, we propose a coarse-to-fine online planning mechanism to enhance the ability of planning and API selection. For the complex scenario of calling RESTful APIs, we also specially designed an API executor to formulate parameters and parse API responses. Experiments show that RestGPT is able to achieve impressive results in complex tasks and has strong robustness, which paves a new way towards AGI., Comment: Work in progress

Continual learning (CL) aims to constantly learn new knowledge over time while avoiding catastrophic forgetting on old tasks. In this work, we focus on continual text classification under the class-incremental setting. Recent CL studies find that the representations learned in one task may not be effective for other tasks, namely representation bias problem. For the first time we formally analyze representation bias from an information bottleneck perspective and suggest that exploiting representations with more class-relevant information could alleviate the bias. To this end, we propose a novel replay-based continual text classification method, RepCL. Our approach utilizes contrastive and generative representation learning objectives to capture more class-relevant features. In addition, RepCL introduces an adversarial replay strategy to alleviate the overfitting problem of replay. Experiments demonstrate that RepCL effectively alleviates forgetting and achieves state-of-the-art performance on three text classification tasks.

Joint entity and relation extraction (JERE) is one of the most important tasks in information extraction. However, most existing works focus on sentence-level coarse-grained JERE, which have limitations in real-world scenarios. In this paper, we construct a large-scale document-level fine-grained JERE dataset DocRED-FE, which improves DocRED with Fine-Grained Entity Type. Specifically, we redesign a hierarchical entity type schema including 11 coarse-grained types and 119 fine-grained types, and then re-annotate DocRED manually according to this schema. Through comprehensive experiments we find that: (1) DocRED-FE is challenging to existing JERE models; (2) Our fine-grained entity types promote relation classification. We make DocRED-FE with instruction and the code for our baselines publicly available at https://github.com/PKU-TANGENT/DOCRED-FE., Comment: Accepted by IEEE ICASSP 2023. The first two authors contribute equally

Three-dimensional braiding composite material has the advantages of high strength, high modulus, high temperature resistance, non-layered structure and easy design. Traditional high-performance metal materials will be replaced by 3D braiding composite material, so it has been highly valued and applied in many fields such as aerospace, weaponry and equipment. At present, there are many researches on 3D braiding technology, such as 3D rotary braiding machine and Cartesian 3D braiding machine in terms of braiding equipment. The research on the structure of preforms and the performance of composite materials are also more in-depth. The development of 3D braiding technology from the aspects of 3D fabric, braiding technology, equipment are summarized in this article. The development of 3D braiding technology requires high-speed automation and commercialization, this will be a challenge for 3D braiding technology in the future.

Fine-tuning pretrained language models (PLMs) on downstream tasks has become common practice in natural language processing. However, most of the PLMs are vulnerable, e.g., they are brittle under adversarial attacks or imbalanced data, which hinders the application of the PLMs on some downstream tasks, especially in safe-critical scenarios. In this paper, we propose a simple yet effective fine-tuning method called Match-Tuning to force the PLMs to be more robust. For each instance in a batch, we involve other instances in the same batch to interact with it. To be specific, regarding the instances with other labels as a perturbation, Match-Tuning makes the model more robust to noise at the beginning of training. While nearing the end, Match-Tuning focuses more on performing an interpolation among the instances with the same label for better generalization. Extensive experiments on various tasks in GLUE benchmark show that Match-Tuning consistently outperforms the vanilla fine-tuning by $1.64$ scores. Moreover, Match-Tuning exhibits remarkable robustness to adversarial attacks and data imbalance., IJCAI-ECAI 2022 (the 31st International Joint Conference on Artificial Intelligence and the 25th European Conference on Artificial Intelligence)

In this study, 2-chloro-1,3-dimethoxy-5-methylbenzene (CDM), a natural product with anti

As a significant part of outdoor built-environment, public open spaces are closely associated with people's daily lives. Studies of outdoor behavior in these spaces can shed light on users' environmental perceptions and contribute to the promotion of physiological and psychological health. Many recent studies are case studies focused where observations, surveys and interviews have been conducted to understand the factors influencing people's behavior on one or few sites or city environments. There have been few reviews related to this topic, and none have been based on the systematic understanding of influencing factors. This paper presents a systematic review of interactions between behavior and the built environment in public open spaces, and highlights the impacts of diverse and objective influencing factors. Followed the rules of PRISMA method (Preferred Reporting Items for Systematic Reviews and Meta-Analyses), 109 papers published in 2000–2021 were selected and reviewed. The distribution of the studied interactions is analyzed, and the impacts of four distinct factors: personal background, location and context, environmental component, and climate stimuli, are extracted, categorized, and specified. Moreover, outdoor health benefits are discussed based on which, crucial factors that require emphasis after the outbreak of COVID-19 are identified. Throughout this paper, behavioral influencing processes, including objective influencing factors, subjective feedback, and the relationships involved, are considered to provide a comprehensive picture. With the robust classification of existing factors, architects, urban designers, policy makers and fellow researches could be easier to get a more comprehensive trend from the past. This paper also provides guidance for future research, especially given that COVID-19 has created huge changes to outdoor needs and customary behavior.Systematic Review Registration:http://www.prisma-statement.org/.

Continual relation extraction (CRE) requires the model to continually learn new relations from class-incremental data streams. In this paper, we propose a Frustratingly easy but Effective Approach (FEA) method with two learning stages for CRE: 1) Fast Adaption (FA) warms up the model with only new data. 2) Balanced Tuning (BT) finetunes the model on the balanced memory data. Despite its simplicity, FEA achieves comparable (on TACRED or superior (on FewRel) performance compared with the state-of-the-art baselines. With careful examinations, we find that the data imbalance between new and old relations leads to a skewed decision boundary in the head classifiers over the pretrained encoders, thus hurting the overall performance. In FEA, the FA stage unleashes the potential of memory data for the subsequent finetuning, while the BT stage helps establish a more balanced decision boundary. With a unified view, we find that two strong CRE baselines can be subsumed into the proposed training pipeline. The success of FEA also provides actionable insights and suggestions for future model designing in CRE.

Previous literature has proved that Pretrained Language Models (PLMs) can store factual knowledge. However, we find that facts stored in the PLMs are not always correct. It motivates us to explore a fundamental question: How do we calibrate factual knowledge in PLMs without re-training from scratch? In this work, we propose a simple and lightweight method CaliNet to achieve this goal. To be specific, we first detect whether PLMs can learn the right facts via a contrastive score between right and fake facts. If not, we then use a lightweight method to add and adapt new parameters to specific factual texts. Experiments on the knowledge probing task show the calibration effectiveness and efficiency. In addition, through closed-book question answering, we find that the calibrated PLM possesses knowledge generalization ability after fine-tuning. Beyond the calibration performance, we further investigate and visualize the knowledge calibration mechanism., Comment: Accepted by Findings of EMNLP 2022

Figures of speech, such as metaphor and irony, are ubiquitous in literature works and colloquial conversations. This poses great challenge for natural language understanding since figures of speech usually deviate from their ostensible meanings to express deeper semantic implications. Previous research lays emphasis on the literary aspect of figures and seldom provide a comprehensive exploration from a view of computational linguistics. In this paper, we first propose the concept of figurative unit, which is the carrier of a figure. Then we select 12 types of figures commonly used in Chinese, and build a Chinese corpus for Contextualized Figure Recognition (ConFiguRe). Different from previous token-level or sentence-level counterparts, ConFiguRe aims at extracting a figurative unit from discourse-level context, and classifying the figurative unit into the right figure type. On ConFiguRe, three tasks, i.e., figure extraction, figure type classification and figure recognition, are designed and the state-of-the-art techniques are utilized to implement the benchmarks. We conduct thorough experiments and show that all three tasks are challenging for existing models, thus requiring further research. Our dataset and code are publicly available at https://github.com/pku-tangent/ConFiguRe., Comment: Accepted to Coling 2022

Continual relation extraction (CRE) aims to continually learn new relations from a class-incremental data stream. CRE model usually suffers from catastrophic forgetting problem, i.e., the performance of old relations seriously degrades when the model learns new relations. Most previous work attributes catastrophic forgetting to the corruption of the learned representations as new relations come, with an implicit assumption that the CRE models have adequately learned the old relations. In this paper, through empirical studies we argue that this assumption may not hold, and an important reason for catastrophic forgetting is that the learned representations do not have good robustness against the appearance of analogous relations in the subsequent learning process. To address this issue, we encourage the model to learn more precise and robust representations through a simple yet effective adversarial class augmentation mechanism (ACA), which is easy to implement and model-agnostic. Experimental results show that ACA can consistently improve the performance of state-of-the-art CRE models on two popular benchmarks., Comment: Accepted by EMNLP 2022

Since the notion of Multiparty Computation (MPC) was proposed three decades ago, a lot of research and effort has been done to improve the efficiency of MPC protocols. However, the inefficiency of the current state-of-the-art is still the major barrier that prevents MPC from being used more broadly. In this thesis, we study the communication complexity of information-theoretic MPC protocols. Part I: Information Theoretic MPC with Honest Majority. First, we study the honest majority setting, i.e., the number of corrupted parties is t < n/2, where n is the number of parties. We consider both the semi-honest security and malicious security (with abort). In the setting of semi-honest security, we propose two improvements over the previously best-known result by Damgard and Nielsen (CRYPTO 2007). Our first ˚ protocol improves the communication complexity of the Damgard and Nielsen pro- ˚ tocol from 6 field elements per party per gate to 4 elements. Our second protocol reduces the round complexity by a factor of 2 and achieves 4.5 field elements per party per gate. As for malicious security, we show that it can be achieved with the same concrete efficiency as the semi-honest security. Previously best known results have a factor of 2 in the communication complexity when compiling a semi-honest protocol to a maliciously secure one. Our result closes the communication efficiency gap between semi-honest security and malicious security (with abort). Part II: Information-Theoretic MPC with Dishonest Majority in the Preprocessing Model. Then, we study the dishonest majority setting in the circuit-independent preprocessing model. We consider a sub-optimal corruption threshold where t = (1−?)·n for a constant ?. In this thesis, we construct the first MPC protocols in the preprocessing model for dishonest majority which achieves sub-linear amount of preprocessing data and communication complexity per gate in the number of parties n. To achieve our results, we extend the use of packed secret sharing to the dishonest majority setting. For a constant fraction of corrupted parties (i.e. if 99 percent of the parties are corrupt), we achieve O(1) field elements in both of the amount of preprocessing data and communication complexity per gate across all parties.

Reliable quantification of natural and anthropogenic gas release (e.g.\ CO2, methane) from the seafloor into the ocean, and ultimately, the atmosphere, is a challenging task. While ship-based echo sounders allow detection of free gas in the water even from a larger distance, exact quantification requires parameters such as rise speed and bubble size distribution not obtainable by such sensors. Optical methods are complementary in the sense that they can provide high temporal and spatial resolution of single bubbles or bubble streams from close distance. In this contribution we introduce a complete instrument and evaluation method for optical bubble stream characterization. The dedicated instrument employs a high-speed deep sea stereo camera system that can record terabytes of bubble imagery when deployed at a seep site for later automated analysis. Bubble characteristics can be obtained for short sequences of few minutes, then relocating the instrument to other locations, or in autonomous mode of intervals up to several days, in order to capture variations due to current and pressure changes and across tidal cycles. Beside reporting the steps to make bubble characterization robust and autonomous, we carefully evaluate the reachable accuracy and propose a novel calibration procedure that, due to the lack of point correspondences, uses only the silhouettes of bubbles. The system has been operated successfully in up to 1000m water depth in the Pacific Ocean to assess methane fluxes. Besides sample results we also report failure cases and lessons learnt during development.

With increasing environmental damage, passive architectural design strategies, particularly natural ventilation strategies, are essential techniques for architectural design to protect the environment. In order to improve human comfort and the health of the living environment, spaces for human activities are no longer bounded primarily by closed, indoor artificial spaces and have extended to open, semi-indoor naturally ventilated spaces. Singapore, located in the tropics, has a unique climate where the attention of the Singapore government to natural ventilation indicators in green building decision making, and application of natural ventilation tropical architecture, can be seen. This paper analyses the development of evaluation contents related to natural ventilation in the Singapore green building standard – Green Mark, studies typical naturally ventilated buildings and then proposes effective design strategies. The relationship between standards and cases are finally discussed to explore the differences from theory to practice, with a view to using suitable approaches and methods to encourage natural ventilation design, thus promoting the green building movement and achieving sustainability.

The Rosetta software for macromolecular modeling, docking and design is extensively used in laboratories worldwide. During two decades of development by a community of laboratories at more than 60 institutions, Rosetta has been continuously refactored and extended. Its advantages are its performance and interoperability between broad modeling capabilities. Here we review tools developed in the last 5 years, including over 80 methods. We discuss improvements to the score function, user interfaces and usability. Rosetta is available at http://www.rosettacommons.org.

Grassland ecosystems play an important role in maintaining biodiversity, global carbon balance. Aiming at the desert steppe with an extremely fragile ecological environment, the spatiotemporal relation between vegetation fractional coverage (VFC) and climate was explored in this study in a pixel-scale analysis method. There was an overall increasing trend in the VFC of the study area. The coefficient of correlation between average precipitation from 2000 to 2013 and VFC was 0.81, and during the growing season, the coefficient of correlation between average temperature and VFC was −0.76. The two dominant factors of VFC in the study area, precipitation and temperature, were used as indicators to establish an ecological suitability index (ESI) for desert steppe to reflect the hydrothermal conditions for vegetation growth. The ESI of the study area varied from 0.03 to 0.94 with its average being 0.39, and the coefficient of correlation between ESI and VFC was 0.84. On the basis of ESI, an optimized layout adjustment was made for existing farmland in the study area so as to take advantage of natural conditions to the greatest extent and reduce impacts of human activities. The optimized layout of farmlands would adapt better to the natural climatic characteristics of the desert steppe regions, meanwhile, bring about ecological and environment benefits to the original agro–pastoral ecotone of the desert grasslands areas. Results from this study further our understanding of the desert steppe, and provide some reference for ecological restoration and new cycle of returning farmland to grassland policy.

The nose deflecting mechanism is important in the development of deflectable nose control technology. In this paper,drawbacks of several existing deflection mechanisms are analyzed, and design requirements of the nose deflection mechanism are summarized. Then a Screw-Spike nose deflecting mechanism is established to meet the design requirements. After that, the relationships between nose deflecting angles and rotation of motors are deduced. At last,the ADAMS-MATLAB\Simulink co-simulations are employed to validate that the Screw-Spike mechanism which is of high precision, fast response, large achievable deflecting angles can meet the design requirements.

Chiral symmetry is always broken in cold, dense matter, by chiral condensation at low densities and by diquark condensation at high density. We construct here, within a schematic Nambu-Jona-Lasinio (NJL) model, the corresponding generalized Nambu-Goldstone pion, $\pi_G$. As we show, the $\pi_G$ mode naturally emerges as a linear combination of the $\langle \bar{q}q\rangle$ vacuum pion $\pi$ and the $\langle qq \rangle$ diquark condensate pion $\tilde{\pi}$, with $q$ the quark field, and continuously evolves with increasing density from being $\pi$-like in the vacuum to $\tilde{\pi}$-like in the high density diquark pairing phase. We calculate the density-dependent mass, decay constant, and coupling to quarks of the $\pi_G$, and derive a generalized Gell-Mann--Oakes--Renner (GMOR) relation in the presence of a finite bare quark mass $m_q$. We briefly discuss the implications of the results to possible Bose condensation of $\pi_G$ in more realistic models., Comment: 15 pages, 7 figures, 1 table

An InfraRed Free Electron Laser Light (IRFEL) is under commissioning at National Synchrotron Radiation Laboratory (NSRL). The control system of IRFEL is a distributed system based on Experimental Physics and Industrial Control System (EPICS). The alarm system is an essential part of the control system. It is developed based on the software Phoebus. The module named "Alarms" in Phoebus can store states and configuration information of the Process Variable (PV) in the Kafka topics. To meet our requirements, 3 kinds of alarm message distribution applications are developed, i.e. Web-Based GUI, WeChat and SMS. This paper will introduce the alarm system architecture and the implementations of the applications for alarm message distribution., Proceedings of the 10th Int. Particle Accelerator Conf., IPAC2019, Melbourne, Australia

Neither single nor discontinuous hydrological observation data can truly reflect periodic changes in soil moisture under natural conditions or interrelationships between various water sources. Therefore, in this study, precipitation pulse characteristics and variations in the soil water content (SWC) and actual evapotranspiration (ETa) in relation to pulses are explored through a field multi-water continuous observation system set in desert steppe in Inner Mongolia, China. A comparison between precipitation events in the growing seasons of 2016 and 2017 shows that precipitation events that are greater than 10 mm are the main cause of dramatic interannual precipitation variations in this region. A single small precipitation event has a limited impact on SWC and provides no obvious increase in the SWC within the top 10 cm soil layer. The precipitation interval ratio (P/I) is suitable for comparing water stresses of different drying-wetting cycles, and correlations between soil layers are found to be closer in humid years than in dry years. In this study, three modes of interpulse ETa in the desert steppe are discussed: a stable ETa mode under a water-sufficient condition, an attenuation ETa mode, and a stable ETa mode under extreme drought conditions.

The fleeting lifetimes of the transition states (TSs) of chemical reactions make determination of their three-dimensional structures by diffraction methods a challenge. Here, we used packing interactions within the core of a protein to stabilize the planar TS conformation for rotation around the central carbon-carbon bond of biphenyl so that it could be directly observed by x-ray crystallography. The computational protein design software Rosetta was used to design a pocket within threonyl-transfer RNA synthetase from the thermophile Pyrococcus abyssi that forms complementary van der Waals interactions with a planar biphenyl. This latter moiety was introduced biosynthetically as the side chain of the noncanonical amino acid p-biphenylalanine. Through iterative rounds of computational design and structural analysis, we identified a protein in which the side chain of p-biphenylalanine is trapped in the energetically disfavored, coplanar conformation of the TS of the bond rotation reaction.

In this paper, we combine the concept of reserve capacity with the continuous equilibrium network design problem. An integrated method is used to maximize the reserve capacity of a road network. On the one hand we try to find the maximum possible increase in traffic demand by setting traffic signals at individual intersections. On the other hand, we increase the road capacity in order to increase the whole capacity of a road network. A bilevel programming model and heuristic solution algorithm based on sensitivity analysis are proposed to model the reserve capacity problem of optimal signal control with user-equilibrium route choice. The applications of the model and its algorithm are illustrated with two numerical examples.

L-band Electron Accelerator is widely used for industrial irradiation. This paper describes a constant-impedance, disk-loaded structure operating on the 2Pi/3 mode. The design details of L-band travelling wave accelerating structures are presented. All RF parameters in metal disk-loaded waveguides and fields were calculated. The SUPERFISH code was used to design the bunching and accelerating cavities. At the same time, we also calculated the beam dynamics. Some model cavities have been fabricated and experimental studies were carried on. In this study, some valuable results were obtained, which can provide a beneficial datum for the design and manufacture of L-band travelling-wave accelerating structures of 50MeV LINAC., Proceedings of the 8th Int. Particle Accelerator Conf., IPAC2017, Copenhagen, Denmark

Based on the national precipitation dataset in the Huang-Huai-Hai River Basin of China for the 1961–2011 period, the China-Z growing season index was calculated to analyze the characteristics of the evolution of meteorological droughts. Data from statistical droughts (the droughts which are defined and classified by the actual statistic data) were compared with those of meteorological droughts during the 2000–2011 period. Nine indexes were selected to evaluate the drought coping ability of water conservancy projects based on the fuzzy comprehensive assessment model. The results showed that the China-Z growing season index was a downward trend with a rate of −0.063 per decade from 1961 to 2011, which indicates an increasing trend in drought intensity. Droughts were more frequent than average during the 1961–1979 period and returned to normal frequency during the 1980–2011 period. Both the ratio of drought affected area (RDAA) and the ratio of drought suffering area (RDSA) of statistical droughts decreased more quickly than those of meteorological droughts (2000–2011). The indexes of water conservancy projects were lower than the average all-China indexes. Half of the 59 three-level water resources districts exhibited a relatively poor drought coping ability, which means that enhancing the drought coping ability of the water conservancy project was quite important.