查看更多>>摘要：Cross-platform binary code similarity detection aims at detecting whether two or more pieces of bina-ry code are similar or not.Existing approaches that combine control flow graphs(CFGs)-based function representa-tion and graph convolutional network(GCN)-based similarity analysis are the best-performing ones.Due to a large amount of convolutional computation and the loss of structural information,the use of convolution networks will in-evitably bring problems such as high overhead and sometimes inaccuracy.To address these issues,we propose a fast cross-platform binary code similarity detection framework that takes advantage of natural language processing(NLP)and inductive graph neural network(GNN)for basic blocks embedding and function representation respectively by simulating extracting structural features and temporal features.GNN's node-centric and small batch is a suitable training way for large CFGs,it can greatly reduce computational overhead.Various NLP basic block embedding models and GNNs are evaluated.Experimental results show that the scheme with long short term memory(LSTM)for basic blocks embedding and inductive learning-based GraphSAGE(GAE)for function representation outperforms the state-of-the-art works.In our framework,we can take only 45％overhead.Improve efficiency significantly with a small performance trade-off.
查看更多>>摘要：The ability to learn incrementally is critical to the long-term operation of AI systems.Benefiting from the power of few-shot class-incremental learning(FSCIL),deep learning models can continuously recognize new class-es with only a few samples.The difficulty is that limited instances of new classes will lead to overfitting and exacer-bate the catastrophic forgetting of the old classes.Most previous works alleviate the above problems by imposing strong constraints on the model structure or parameters,but ignoring embedding network transferability and classifi-er adaptation(CA),failing to guarantee the efficient utilization of visual features and establishing relationships be-tween old and new classes.In this paper,we propose a simple and novel approach from two perspectives:embedding bias and classifier bias.The method learns an embedding augmented(EA)network with cross-class transfer and class-specific discriminative abilities based on self-supervised learning and modulated attention to alleviate embed-ding bias.Based on the adaptive incremental classifier learning scheme to realize incremental learning capability,guiding the adaptive update of prototypes and feature embeddings to alleviate classifier bias.We conduct extensive experiments on two popular natural image datasets and two medical datasets.The experiments show that our method is significantly better than the baseline and achieves state-of-the-art results.
查看更多>>摘要：Low hardware cost and power consumption in information transmission,processing and storage is an urgent demand for many big data problems,in which the high-dimensional data often be modelled as graph signals.This paper considers the problem of recovering a smooth graph signal by using its low-resolution multi-bit quantized observations.The underlying problem is formulated as a regularized maximum-likelihood optimization and is solved via an expectation maximization scheme.With this scheme,the multi-bit graph signal recovery(MB-GSR)is effi-ciently implemented by using the quantized observations collected from random subsets of graph nodes.The simula-tion results show that increasing the sampling resolution to 2 or 3 bits per sample leads to a considerable perfor-mance improvement,while the energy consumption and implementation costs remain much lower compared to the implementation of high resolution sampling.
查看更多>>摘要：Compressive sensing technique has been widely applied to achieve range-Doppler reconstruction of high frequency radar by utilizing sparse random stepped-frequency(SRSF)signal,which can suppress the complex electromagnetic interference and greatly reduce the coherent processing interval.An important way to improve the performance of sparse signal reconstruction is to optimize the sensing matrix(SM).However,the existing research on the SM optimization needs to design a measurement matrix with superior performance,which needs a large amount of computation and does not consider the influence of the waveform parameters design.In order to improve the supe-rior reconstruction performance,a novel SM optimization approach for SRSF signal is proposed by using two-dimen-sional ambiguity function(TDAF)in this paper.Firstly,based on the two-dimensional sparse reconstruction model of the SRSFs,the internal relationship between the waveform parameters and the SM was derived.Secondly,the SM optimization problem was directly transformed into the waveform design of SRSFs.Furthermore,on the basis of ana-lyzing the relationship between the mutual coherence matrix of SM and the TDAF matrix of SRSFs,the purpose of optimizing the SM can be achieved by designing the TDAF of the SRSFs.Based on this analysis,a sparse waveform optimization method with joint constraints of maximum and mean sidelobes of the TDAF by using the genetic algo-rithm was derived.Compared with the traditional SM optimization method,our method not only avoids generating a new measurement matrix,but also further reduces the complexity of the waveform optimization.Simulation experi-ments verified the effectiveness of the proposed method.
查看更多>>摘要：For a sub-connected hybrid multiple-input multiple-output(MIMO)receiver with K subarrays and N antennas,there exists a challenging problem of how to rapidly remove phase ambiguity in only single time-slot.A di-rection of arrival(DOA)estimator of maximizing received power(Max-RP)is proposed to find the maximum value of K-subarray output powers,where each subarray is in charge of one sector,and the center angle of the sector corre-sponding to the maximum output is the estimated true DOA.To make an enhancement on precision,Max-RP plus quadratic interpolation(Max-RP-QI)method is designed.In the proposed Max-RP-QI,a quadratic interpolation scheme is adopted to interpolate the three DOA values corresponding to the largest three receive powers of Max-RP.To achieve the Cramer Rao lower bound,a Root-MUSIC plus Max-RP-QI scheme is developed.Simulation results show that the proposed three methods eliminate the phase ambiguity during one time-slot and also show low compu-tational complexities.The proposed Root-MUSIC plus Max-RP-QI scheme can reach the Cramer Rao lower bound,and the proposed Max-RP and Max-RP-QI are still some performance losses 2-4 dB compared to the Cramer Rao lower bound.
查看更多>>摘要：As an extension of permutation entropy(PE),coded permutation entropy(CPE)improves the perfor-mance of PE by making a secondary division for ordinal patterns defined in PE.In this study,we provide an explo-ration of the statistical properties of CPE using a finite length Gaussian white noise time series theoretically.By means of the Taylor series expansion,the approximate expressions of the expected value and variance of CPE are de-duced and the Cramér-Rao low bound(CRLB)is obtained to evaluate the performance of the CPE estimator.The results indicate that CPE is a biased estimator,but the bias only depends on relevant parameters of CPE and it can be easily corrected for an arbitrary time series.The variance of CPE is related to the encoding patterns distribution,and the value converges to the CRLB of the CPE estimator when the time series length is large enough.For a finite-length Gaussian white noise time series model,the predicted values can match well with the actual values,which fur-ther validates the statistic theory of CPE.Using the theoretical expressions of CPE,it is possible to better under-stand the behavior of CPE for most of the time series.
查看更多>>摘要：Visible-light indoor positioning is a new generation of positioning technology that can be integrated into smart lighting and optical communications.The current received signal strength(RSS)-based visible-light posi-tioning systems struggle to overcome the interferences of background and indoor-reflected noise.Meanwhile,when en-suring the lighting,it is impossible to use the superposition of each light source to accurately distinguish light source information;furthermore,it is difficult to achieve accurate positioning in complex indoor environments.This study proposes an indoor positioning method based on a combination of power spectral density detection and a neural net-work.The system integrates the mechanism for visible-light radiation detection with RSS theory,to build a back propagation neural network model fitting for multiple reflection channels.Different frequency signals are loaded to different light sources at the beacon end,and the characteristic frequency and power vectors are obtained at the loca-tion end using the Pisarenko harmonic decomposition method.Then,a complete fingerprint database is established to train the neural network model and conduct location tests.Finally,the location effectiveness of the proposed algo-rithm is verified via actual positioning experiments.The simulation results show that,when four groups of sinusoidal waves with different frequencies are superimposed with white noise,the maximum frequency error is 0.104 Hz and the maximum power error is 0.0362 W.For the measured positioning stage,a 0.8 m × 0.8 m × 0.8 m solid wood stereoscopic positioning model is constructed,and the average error is 4.28 cm.This study provides an effective method for separating multi-source signal energies,overcoming background noise,and improving indoor visible-light positioning accuracies.
查看更多>>摘要：Link prediction utilizes accessible network information to complement or predict the network links.Similarity is an important prerequisite for link prediction which means links more likely occurs between two similar nodes.Existing methods utilize the similarity of nodes but neglect of network structure.However the link direction leads to a far more complex structure and contains more information useful than the undirected networks.Most clas-sic methods are difficult to depict the distribution of the network structure with incidental direction so the similarity characteristics of the network structure itself are lost.In this respect,a new method of local structure entropy is pro-posed to depict the directed structural distribution characteristics,which can be used to evaluate the degree of local structural similarity of nodes and then applied to link prediction methods.Experimental results on 8 real directed networks show that this method is effective for both area under the receiver operating characteristic curve(AUC)and ranking-score measures,and improved predictive capacity of the baseline methodology.
查看更多>>摘要：With the diversification of space-based information network task requirements and the dramatic in-crease in demand,the efficient scheduling of various tasks in space-based information network becomes a new chal-lenge.To address the problems of a limited number of resources and resource heterogeneity in the space-based infor-mation network,we propose a bilateral pre-processing model for tasks and resources in the scheduling pre-processing stage.We use an improved fuzzy clustering method to cluster tasks and resources and design coding rules and match-ing methods to match similar categories to improve the clustering effect.We propose a space-based information net-work task scheduling strategy based on an ant colony simulated annealing algorithm for the problems of high latency of space-based information network communication and high resource dynamics.The strategy can efficiently com-plete the task and resource matching and improve the task scheduling performance.The experimental results show that our proposed task scheduling strategy has less task execution time and higher resource utilization than other al-gorithms under the same experimental conditions.It has significantly improved scheduling performance.
查看更多>>摘要：Drug-target interactions(DTIs)prediction plays an important role in the process of drug discovery.Most computational methods treat it as a binary prediction problem,determining whether there are connections be-tween drugs and targets while ignoring relational types information.Considering the positive or negative effects of DTIs will facilitate the study on comprehensive mechanisms of multiple drugs on a common target,in this work,we model DTIs on signed heterogeneous networks,through categorizing interaction patterns of DTIs and additionally ex-tracting interactions within drug pairs and target protein pairs.We propose signed heterogeneous graph neural net-works(SHGNNs),further put forward an end-to-end framework for signed DTIs prediction,called SHGNN-DTI,which not only adapts to signed bipartite networks,but also could naturally incorporate auxiliary information from drug-drug interactions(DDIs)and protein-protein interactions(PPIs).For the framework,we solve the message pass-ing and aggregation problem on signed DTI networks,and consider different training modes on the whole networks consisting of DTIs,DDIs and PPIs.Experiments are conducted on two datasets extracted from DrugBank and relat-ed databases,under different settings of initial inputs,embedding dimensions and training modes.The prediction re-sults show excellent performance in terms of metric indicators,and the feasibility is further verified by the case study with two drugs on breast cancer.
万方数据