Search

Pre-trained large language model (LLM) is under exploration to perform NLP tasks that may require logical reasoning. Logic-driven data augmentation for representation learning has been shown to improve the performance of tasks requiring logical reasoning, but most of these data rely on designed templates and therefore lack generalization. In this regard, we propose an AMR-based logical equivalence-driven data augmentation method (AMR-LE) for generating logically equivalent data. Specifically, we first parse a text into the form of an AMR graph, next apply four logical equivalence laws (contraposition, double negation, commutative and implication laws) on the AMR graph to construct a logically equivalent/inequivalent AMR graph, and then convert it into a logically equivalent/inequivalent sentence. To help the model to better learn these logical equivalence laws, we propose a logical equivalence-driven contrastive learning training paradigm, which aims to distinguish the difference between logical equivalence and inequivalence. Our AMR-LE (Ensemble) achieves #2 on the ReClor leaderboard https://eval.ai/web/challenges/challenge-page/503/leaderboard/1347 . Our model shows better performance on seven downstream tasks, including ReClor, LogiQA, MNLI, MRPC, RTE, QNLI, and QQP. The source code and dataset are public at https://github.com/Strong-AI-Lab/Logical-Equivalence-driven-AMR-Data-Augmentation-for-Representation-Learning .

Multi-hop question answering (QA) requires reasoning over multiple documents to answer a complex question and provide interpretable supporting evidence. However, providing supporting evidence is not enough to demonstrate that a model has performed the desired reasoning to reach the correct answer. Most existing multi-hop QA methods fail to answer a large fraction of sub-questions, even if their parent questions are answered correctly. In this paper, we propose the Prompt-based Conservation Learning (PCL) framework for multi-hop QA, which acquires new knowledge from multi-hop QA tasks while conserving old knowledge learned on single-hop QA tasks, mitigating forgetting. Specifically, we first train a model on existing single-hop QA tasks, and then freeze this model and expand it by allocating additional sub-networks for the multi-hop QA task. Moreover, to condition pre-trained language models to stimulate the kind of reasoning required for specific multi-hop questions, we learn soft prompts for the novel sub-networks to perform type-specific reasoning. Experimental results on the HotpotQA benchmark show that PCL is competitive for multi-hop QA and retains good performance on the corresponding single-hop sub-questions, demonstrating the efficacy of PCL in mitigating knowledge loss by forgetting., Accepted to COLING 2022

The classification service over a stream of data is becoming an important offering for cloud providers, but users may encounter obstacles in providing sensitive data due to privacy concerns. While Trusted Execution Environments (TEEs) are promising solutions for protecting private data, they remain vulnerable to side-channel attacks induced by data-dependent access patterns. We propose a Privacy-preserving Data Stream Training and Inference scheme, called EnclaveTree, that provides confidentiality for user's data and the target models against a compromised cloud service provider. We design a matrix-based training and inference procedure to train the Hoeffding Tree (HT) model and perform inference with the trained model inside the trusted area of TEEs, which provably prevent the exploitation of access-pattern-based attacks. The performance evaluation shows that EnclaveTree is practical for processing the data streams with small or medium number of features. When there are less than 63 binary features, EnclaveTree is up to ${\thicksim}10{\times}$ and ${\thicksim}9{\times}$ faster than na\"ive oblivious solution on training and inference, respectively., Comment: 15 pages, 12 figures

Effective multi-hop question answering (QA) requires reasoning over multiple scattered paragraphs and providing explanations for answers. Most existing approaches cannot provide an interpretable reasoning process to illustrate how these models arrive at an answer. In this paper, we propose a Question Decomposition method based on Abstract Meaning Representation (QDAMR) for multi-hop QA, which achieves interpretable reasoning by decomposing a multi-hop question into simpler sub-questions and answering them in order. Since annotating the decomposition is expensive, we first delegate the complexity of understanding the multi-hop question to an AMR parser. We then achieve the decomposition of a multi-hop question via segmentation of the corresponding AMR graph based on the required reasoning type. Finally, we generate sub-questions using an AMR-to-Text generation model and answer them with an off-the-shelf QA model. Experimental results on HotpotQA demonstrate that our approach is competitive for interpretable reasoning and that the sub-questions generated by QDAMR are well-formed, outperforming existing question-decomposition-based multi-hop QA approaches., Comment: Accepted by IJCAI 2022

In order to extract the information from the massive unstructured recipe text, a named entity recognition method based on BERT pre-trained language model is proposed by this paper. Firstly, after pre-processing the public recipe texts obtained from the web, the entities were classified into three categories of main ingredients, auxiliary ingredients and cooking methods by combining the domain expert knowledge and BIO annotation method to build a large-scale Chinese cuisine domain named entity recognition dataset, and then the BERT-Attention-BiLSTM-CRF(BABC) model proposed in this paper is used to train on this dataset. Finally, we compared the results of other named entity recognition models on this dataset, and the experimental results showed that BABC could identify the entities in the recipe text more accurately.

Mung bean nuclease is a single stranded specific DNA and RNA endonuclease purified from mung bean sprouts. It yields 5'-phosphate terminated mono- and oligonucleotides. The activity level of this nuclease can act as a marker to monitor the developmental process of mung bean sprouts. In order to facilitate the activity and physiological analysis of this nuclease, we have developed a biosensing assay system based on the mung bean nuclease-induced single-stranded DNA scission and the affinity difference of graphene oxide for single-stranded DNA containing different numbers of bases. This end-point measurement method can detect mung bean nuclease in a range of 2×10(-4) to 4×10(-2) with a detection limit of 1×10(-4) unit/mL. In addition, we demonstrate the utility of the assay for screening chemical antibiotics and metal ions, resulting in the identification of several inhibitors of this enzyme in vitro. Furthermore, we firstly report that inhibiting mung bean nuclease by gentamycin sulfate and kanamycin in vivo can suppress mung bean sprouts growth. In summary, this method provides an alternative tool for the biochemical analysis for mung bean nuclease and indicates the feasibility of high-throughput screening specific inhibitors of this nuclease in vitro and in vivo.

Social tags providing abundant information can stimulate a better recommender system equipped with stronger sense of description and analysis on user’s interest. In this paper, graph-based personalized recommendation techniques have been studied. Complete tripartite graph model was proposed and the user’s interest migration was researched comprehensively, Focusing on the dilemma of accuracy and diversity in recommender system, the mass diffusion algorithm and heat spreading algorithm on complete tripartite graph model were carried out. Then, from the perspective of improving confidence in recommender system, the item-tag joint recommendation mechanism was studied. Experimental results show the effectiveness of the algorithm in this paper.

With the development of deep learning, it has achieved impressive results in feature extraction field. This paper drives research in feature extraction based on deep learning. First, this paper gives a brief introduction on the world's research status on deep learning and principle of Restricted Boltzmann machine (RBM). Then this paper conducts reducing experiment based on RBM for handwritten digits. According to the analysis based on the results of the experiments, this paper tries to get a proper dimension which handwritten digits reduced to achieve better performance. Finally, this paper finds that it reach the goal when handwritten digits is reduced to half dimensional raw digits. This is an important foundation of deep learning layering and offers help to researchers in feature extraction based on deep learning.

Integrated optimal control strategies can reduce the overall building HVAC system energy consumption as well as improved air quality resulting in improved health and cognitive function for the occupants. However, it is time consuming to quantitatively evaluate the design-intended building HVAC automation system performance before on-site deployment, because: 1) the building and HVAC system design specs are in 2D or 3D drawings that require significant efforts to develop the system steady state or dynamic models based on them; 2) the building HVAC control strategies are designed and implemented in building automation (BA) system that could not smoothly connect with the building HVAC system steady state or dynamic models for performance evaluation through close-loop simulation. This paper presents the tool chain of an integrated simulation platform for building HVAC system automation and simulation as well as its implementation in a real case. First, building information from a Revit BIM model is automatically parsed to an EnergyPlus building energy model. Second, the HVAC system model is quickly populated with a scalable HVAC system library in Dymola. Third, the HVAC controls are developed in WebCTRL, a building HVAC automation system by Automated Logic Corporation (ALC). Finally, both the building energy model and HVAC system model are wrapped up as Functional Mock-up Units (FMU) and connected with embedded simulator in WebCTRL to perform close-loop building automation system performance simulation. A real case study, a chiller plant system in a hotel building, is conducted to verify the scalability and benefit of the developed tool chain. The case study demonstrates the values in identifying both HVAC automation system design-intended control issues and improvement areas for integrated optimal controls. This platform enables testing of building HVAC control strategies before on-site deployment, which reduces the labor and time required for building HVAC control development-to-market process and ensure the delivering quality. Furthermore, this platform can be calibrated with metered real-time data from the specific building HVAC system and serve as its “digital twin” that empowers the system fault detection, diagnostics and predictive maintenance.

UNLABELLED This study focused on an endophytic bacterial strain, Pantoea sp. Sd-1, which can be used to degrade lignin and rice straw. This strain was isolated from rice seeds by an optimized surface sterilization method. Pantoea sp. Sd-1 showed exceptional ability to degrade rice straw and lignin. In rice straw or kraft lignin-containing medium supplemented with 1% glucose and 0.5% peptone, Pantoea sp. Sd-1 effectively reduced the rice straw mass weight by 54.5% after 6 days of treatment. The strain was also capable of reducing the lignin colour (52.4%) and content (69.1%) after 4 days of incubation. The findings suggested that the rice endophytic bacterium Pantoea sp. Sd-1 could be applied for the degradation of lignocellulose biomass, such as rice straw. SIGNIFICANCE AND IMPACT OF THE STUDY Rice straw, an abundant agricultural by-product in China, is very difficult to degrade because of its high lignin content. Due to the immense environmental adaptability and biochemical versatility of bacteria, endophytic bacteria are useful resources for biodegradation. In this study, we screened for endophytic bacteria capable of biodegrading rice straw and lignin and obtained one strain, Pantoea sp. Sd-1, with suitable characteristics. Sd-1 could be used for degradation of rice straw and lignin, and may play an important role in biodegradation of this agricultural by-product.

In an attempt to improve the enzymatic efficiency in ball mill, a novel immobilized cellulase on polyvinyl alcohol/Fe2O3 magnetic nanoparticle with high activity was synthesized and characterized by transmission electron microscopy, Zetasizer, Fourier transform infrared (FTIR) spectroscopy and vibrating sample magnetometry. It was found that the immobilized cellulase was a kind of spherical complex with approximate 270 nm, 4.87 A m2 kg−1, and mainly constructed by the accumulation of 10 nm Fe2O3 nanoparticles with the diffused polymer. When the immobilized cellulase was applied to degrade microcrystalline cellulose by combining with wet ball milling, the yielded glucose was 1.89 mg mL−1, at least three times than the sum of individual yield. The immobilized cellulose maintained 40% activity even after four cycles of reuse. From these results, it can be concluded that the immobilization of cellulase with wet ball milling is a novel method to significantly improve the efficiency of cellulose conversion.

An energy flow model of an airport terminal’s HVAC system is established in this paper. Based on energy flow model, the exergy efficiency, exergy loss and exergy cost distribution ratio of each component are calculated and analyzed by the energy flow model. Optimization method and exergy balance equations are used to calculate the least exergy loss of HVAC system under certain operation condition, which is defined as the near-optimum operation level. DEA method is then applied to obtain the benchmarking frontier of near-optimum operation levels, and the frontier will illustrate the ideal operation level under overall operation conditions. In order to measure the gap between actual operation level and ideal operation level, a new index—control perfect index (CPI) is defined by the ratio between the actual exergy loss and exergy loss of ideal operation level of HVAC system, and it can reflect control influences on operation performance of HVAC system. Thus, a new evaluation method is presented which regards ideal operation level as the benchmark and uses CPI to evaluate actual operation performance of HVAC system. Two kinds of control strategies, optimal supply air temperature reset strategy (SAT Strategy) and optimal load allocation control strategy are implemented to validate this evaluation method. Test’s results show that optimal load allocation control strategy can improve the operation performance of the system more greatly than the SAT Strategy. This evaluation method not only can evaluate the operation performance of HVAC system, but also can indicate the direction of optimal control of HVAC system.

Although variable refrigerant flow (VRF) systems have become attractive due to good energy performances in part load conditions, the shortcoming of no outdoor air intake has not been solved thoroughly. A VRF and VAV combined air conditioning system is proposed to solve this problem. VAV part of the combined system consists of an outdoor air processing (OAP) unit and VAV boxes. Generally the VRF unit operates to maintain indoor temperature and the OAP unit operates to process the outdoor air. A control strategy for the combined system aiming at reducing energy consumption is presented in this paper. When both VRF unit and OAP unit are operating, a load allocation optimization module is executed to find the best load allocation between them to minimize the energy consumption of the combined system. When the allocated load of the OAP unit is very small, the proposed control strategy stops the OAP unit, leaving only the VRF unit to operate to improve the overall energy efficiency of the combined system. When load requirements are met, the OAP unit is restarted and the load allocation optimization module is executed again. The proposed control strategy is evaluated based on the developed simulation platform. Results show that the proposed control strategy can effectively decrease the energy consumption of the combined system.

A lignin and cellulose-degrading fungal strain Bio-1 was isolated from the soil of Yuelu Mountain in China. It was identified as a member of the genus Cladosporium by 18s rDNA, ITS sequences analysis and morphological characters. Bio-1 could hydrolyze at least 35% of the alkaline lignin in 4 days when incubated in screening liquid medium. Both laccase and cellulase activities could be detected in the fermentation liquid and cell homogenate. High extracellular and intracellular laccase activities of 3800 and 5352 U/L were obtained with copper induction, respectively. The extracellular cellulase production was reached (51 U/Ml). Based on these characteristics, Bio-1 could be well suited as a commercial fungus applied in degradation of lignocellulose biomass for environment protection. Key words: Cladosporium, identification, laccase, cellulase, CuSO4.

This paper examined eight additional parameters relating to photosynthetic rate of Lonicera japonica Thunb. Correlation analysis, regression analysis and factor analysis were used to analyze the relationship between net photosynthetic rate (Pn) and environment factors, which included stomatal conductance (Sc), intercellular CO2 concentration (Ci), transpiration rate (Tr), air temperature (Ta), leaf temperature (Tl), air CO2 concentration (Ca), relative air humidity (RH), photo flux density (PFD), etc.. The results showed that the diurnal Pn of L. japonica Thunb. It may vary with double peak curves; there was no significant midday depression from June to August. The stepwise regression analysis which introduced characters into a multiple regression equation in the order in which they contribute to Pn, together with factor analysis was used to analyze the same dependence structure. The 8 causes were grouped into the basic factors which generate the dependence structure. The result suggested that high photo flux density (PFD), high Tl and low RH may result in the high Pn. Increasing Tr and PFD, accelerating the hot air flow, would be the most effective way of increasing Pn. The analysis result would assist the medicinal herb grower by adjusting environment factors level to regulate Pn of L. japonica Thunb. and improve output and quality of flower. Key words: Lonicera japonica Thunb., environment factors, net photosynthetic rate, interrelationships.

Ant Colony Optimization ( ACO ) is a novel bionic evolutionary algorithm for solving complex combinatorial optimization problems. This research approach lies at initial stage at present, and a new adaptive ant algorithm is proposed for the traditional ant algorithm easily appears precocious and stagnation behavior phenomenon in this paper. And the traditional parameter of pheromone of ant colony algorithm is self-adaptive adjusted. Selecting a number of typical TSP problems to experiment, the results are indicated that the new adaptive ant colony algorithm has a better ability to search the global optimal solution and have better stability and astringency.

Aimed at the controlled object’s character of long delay-time, this paper presents a strategy that the long delay-time object is remodeled into short delay-time object through decreasing time-delay control, after the adoption of BP neural network PID controller to control the transformed system. In this paper, the three-layer BP neural network is designed and BP neural network PID controller algorithm is analyzed. The simulation results showed that the PID controller based on BP neural network has good control effect in the performance of following, anti-interference, robust.

Wireless Sensor Networks (WSN) are composed of a large number of sensor nodes which are connected by wireless channels. Some of these sensor nodes are regarded as critical nodes which are responsible for the management of the other nodes in their sphere. Therefore, update of node information must be implemented through these critical nodes’ multicast operation. In this paper, we present a novel Node Information Update Multicast (NIUM) algorithm to optimize the transmission performance among critical nodes. After applying with our algorithm, each critical node could explore a reasonable transportation way according to its distance apart from other critical nodes. Despite of the ideal assumption, our algorithm is also tested under different circumstances and the results prove that it behaves well for prompt node information update in wireless sensor networks. Both velocity and efficiency of the network can be improved tremendously.

With the swift development of hardware simulation technology, virtual laboratory environment has appealed more and more attention of educators. In this paper, we develop an interactive virtual laboratory environment for hardware chips application experiment. It provides the students with real expression, convenient operation and powerful simulation capability. Furthermore, interactivity and expansibility make this system distinctive under different circumstances. Till now, our virtual laboratory environment has been implemented and successfully applied in a Computer Interface Technology course in Shanghai University and has gained positive appraisal from teachers and students.

The purpose of an intellectual answering system is that the system can answer questions automatically. To construct this kind of system, it must collect and store questions and answers organically. The intelligent Q&A system will contain a Q&A library, which is built through the semantic understanding of natural language technology and it will be growing automatically during students question and getting answer with the system.

Select the close-to-optimal collective algorithm based on the parameters of the collective call at the run time is an important step in achieving good performance of collective operation in base station wireless network. In this paper, we present our Reversed Multicast Tag Collective (RMTC) scheme to optimize the collective routing in base station wireless network. We verify our mechanism in a simulation platform which is implemented according to the message passing mechanism of base station wireless network and analyze the result and expect run time performance penalty. The results indicate that our Reversed Multicast Tag Collective scheme is flexible and feasible in base station wireless network with its high efficiency and wide applicability.

Recently, service oriented architecture (SOA) has become a trend of implementing software systems including grids. SOA provides more flexibility for grid users at the service level. In this paper, a scalable service-oriented grid computing architecture utilizing Web-service technology is presented. With the service discovery scheme and resource assignment strategy proposed, the whole architecture can manage the service provider dynamically. Fault tolerant computing scheme is also introduced. The performance of the algorithm proposed is evaluated by experiments. Numerical results show that our work provide great efficiency and flexibility for the intensive applications in grid computing or utility computing.

Base station multicast is an important enabling service for the current and future Internet. With the explosive growth of the Internet, a challenging issue facing Base Station multicast is scalability, in particular, the problem of multicast forwarding state and control explosion. In this paper, we propose a new methodology to address the multicast scalability problem for backbone domains - distributed message copying multicast (DMCM). This multicast scheme is designed for the wireless network, and is independent of any underlying intra-domain multicast protocols. The feasibility and performance of our algorithm is demonstrated through analysis and simulations.

A mathematical model for sideways overturns of wheel-tractor and trailer combinations, by the conventional method of disassembly and assembly of system components supplemented with a method of energy symmetry, devised for the present purpose to take care of the singularity range of the roll angle, was developed with a corresponding digital computer program and verified with tests on a 1/4.5 scaled, self-powered remote-controlled test model with agreeable results.

This paper proposes a one-step spectral clustering method by learning an intrinsic affinity matrix (i.e., the clustering result) from the low-dimensional space (i.e., intrinsic subspace) of original data. Specifically, the intrinsic affinitymatrix is learnt by: 1) the alignment of the initial affinity matrix learnt from original data; 2) the adjustment of the transformation matrix, which transfers the original feature space into its intrinsic subspace by simultaneously conducting feature selection and subspace learning; and 3) the clustering result constraint, i.e., the graph constructed by the intrinsic affinity matrix has exact c connected components where c is the number of clusters. In this way, two affinity matrices and a transformation matrix are iteratively updated until achieving their individual optimum, so that these two affinity matrices are consistent and the intrinsic subspace is learnt via the transformation matrix. Experimental results on both synthetic and benchmark datasets verified that our proposed method outputted more effective clustering result than the previous clustering methods.

This paper proposes a graph feature selection method for dementia diagnosis, by adding the information inherent in the observations into a sparse multi-task learning framework. Specifically, this paper first defines two relations (i.e., the feature-feature relation and the sample-sample relation, respectively) based on the prior knowledge in the data. The feature-feature selection enforces the similarity relationship between features to be preserved in the coefficient matrix while the sample-sample relation is designed to preserve the relation between samples invariant in the predicted space. Then we embed these two kinds of relations into a multi-task learning framework (i.e., a least square loss function plus an l2-norm regularization term) to conduct feature selection. Furthermore, we feed the reduced data into Support Vector Machine (SVM) for conducting the identification of Alzheimer's Disease (AD). Finally, the experimental results on a subset of the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset showed the effectiveness of the proposed method in terms of classification accuracy, by comparing with the state-of-the-art methods, including k Nearest Neighbor (kNN), ridge regression, SVM, and so on. Refereed/Peer-reviewed

since amount of unlabeled and high-dimensional datasets need to be preprocessed, unsupervised learning plays a more and more important role in machine learning field. This paper proposed a novel unsupervised feature selection algorithm that can select informative features from dataset without label, by mixing two sparse representation and self-representation loss function into a unified framework. That is, we use self-representation loss function to represent every feature with remainder features and achieve minimum reconstruction mirror, and then utilize l 2 , 1-norm regularization term and l 1-norm regularization term simultaneously to enforce coefficient matrix to be sparse, such that filter redundant and irrelative features in order to conduct feature selection, where l 2 , 1-norm regularization can enforce group sparsity while l 1-norm regularization enforce element sparsity. By this way that utilize both of sparse representation terms, we can choose representative features more accurately. At final, we feed reduced data into support vector machine (SVM) to conduct classification accuracy, which is main assessment criteria to validate performance of algorithm. Extensive experiments on synthetic datasets and real-world datasets have exhibited that our proposed method outperform most of common-used methods, such as PCA, LPP and so on. Refereed/Peer-reviewed

This paper proposes a new k Nearest Neighbor (kNN) algorithm based on sparse learning, so as to overcome the drawbacks of the previous kNN algorithm, such as the fixed k value for each test sample and the neglect of the correlation of samples. Specifically, the paper reconstructs test samples by training samples to learn the optimal k value for each test sample, and then uses kNN algorithm with the learnt k value to conduct all kinds of tasks, such as classification, regression, and missing value imputation. The rationale of the proposed method is that different test samples should be assigned different k values in kNN algorithm, and learning the optimal k value for each test sample should be taken the correlation of data into account. To this end, in the reconstruction process, the proposed method is designed to achieve the minimal reconstruction error via a least square loss function, and employ an ℓ1-norm regularization term to create the element-wise sparsity in the reconstruction coefficient, i.e., sparsity appearing in the element of the coefficient matrix. For achieving effectiveness, the Locality Preserving Projection (LPP) is employed to keep the local structures of data. Finally, the experimental results on real datasets, and the experimental results show that the proposed kNN algorithm is better than the state-of-the-art algorithms in terms of different learning tasks, such as classification, regression, and missing value imputation. Refereed/Peer-reviewed