Binary analysis, as an important foundational technology, provides support for numerous applications in the fields of software engineering and security research. With the continuous expansion of software scale and the complex evolution of software architecture, binary analysis technology is facing new challenges. To break through existing bottlenecks, researchers have applied artificial intelligence (AI) technology to the understanding and analysis of binary code. The core lies in characterizing binary code, i.e., how to use intelligent methods to generate representation vectors containing semantic information for binary code, and apply them to multiple downstream tasks of binary analysis. In this paper, we provide a comprehensive survey of recent advances in binary code representation technology, and introduce the workflow of existing research in two parts, i.e., binary code feature selection methods and binary code feature embedding methods. The feature selection section includes mainly two parts: definition and classification of features, and feature construction. First, the abstract definition and classification of features are systematically explained, and second, the process of constructing specific representations of features is introduced in detail. In the feature embedding section, based on the different intelligent semantic understanding models used, the embedding methods are classified into four categories based on the usage of text-embedding models and graph-embedding models. Finally, we summarize the overall development of existing research and provide prospects for some potential research directions related to binary code representation technology.
Subdivision is a widely used technique for mesh refinement. Classic methods rely on fixed manually defined weighting rules and struggle to generate a finer mesh with appropriate details, while advanced neural subdivision methods achieve data-driven nonlinear subdivision but lack robustness, suffering from limited subdivision levels and artifacts on novel shapes. To address these issues, this paper introduces a neural mesh refinement (NMR) method that uses the geometric structural priors learned from fine meshes to adaptively refine coarse meshes through subdivision, demonstrating robust generalization. Our key insight is that it is necessary to disentangle the network from non-structural information such as scale, rotation, and translation, enabling the network to focus on learning and applying the structural priors of local patches for adaptive refinement. For this purpose, we introduce an intrinsic structure descriptor and a locally adaptive neural filter. The intrinsic structure descriptor excludes the non-structural information to align local patches, thereby stabilizing the input feature space and enabling the network to robustly extract structural priors. The proposed neural filter, using a graph attention mechanism, extracts local structural features and adapts learned priors to local patches. Additionally, we observe that Charbonnier loss can alleviate over-smoothing compared to L2 loss. By combining these design choices, our method gains robust geometric learning and locally adaptive capabilities, enhancing generalization to various situations such as unseen shapes and arbitrary refinement levels. We evaluate our method on a diverse set of complex three-dimensional (3D) shapes, and experimental results show that it outperforms existing subdivision methods in terms of geometry quality. See https://zhuzhiwei99.github.io/NeuralMeshRefinement for the project page.
Artificial neural networks (ANNs) have made great strides in the field of remote sensing image object detection. However, low detection efficiency and high power consumption have always been significant bottlenecks in remote sensing. Spiking neural networks (SNNs) process information in the form of sparse spikes, creating the advantage of high energy efficiency for computer vision tasks. However, most studies have focused on simple classification tasks, and only a few researchers have applied SNNs to object detection in natural images. In this study, we consider the parsimonious nature of biological brains and propose a fast ANN-to-SNN conversion method for remote sensing image detection. We establish a fast sparse model for pulse sequence perception based on group sparse features and conduct transform-domain sparse resampling of the original images to enable fast perception of image features and encoded pulse sequences. In addition, to meet accuracy requirements in relevant remote sensing scenarios, we theoretically analyze the transformation error and propose channel self-decaying weighted normalization (CSWN) to eliminate neuron overactivation. We propose S3Det, a remote sensing image object detection model. Our experiments, based on a large publicly available remote sensing dataset, show that S3Det achieves an accuracy performance similar to that of the ANN. Meanwhile, our transformed network is only 24.32% as sparse as the benchmark and consumes only 1.46 W, which is 1/122 of the original algorithm’s power consumption.
To enhance information security during transmission over public channels, images are frequently employed for binary data hiding. Nonetheless, data are vulnerable to distortion due to Joint Photographic Experts Group (JPEG) compression, leading to challenges in recovering the original binary data. Addressing this issue, this paper introduces a pioneering method for binary data hiding that leverages a combined spatial and channel attention Transformer, termed SCFformer, to withstand JPEG compression. This method employs a novel discrete cosine transform (DCT) quantization truncation mechanism during the hiding phase to bolster the stego image’s resistance to JPEG compression, using spatial and channel attention to conceal information in less perceptible areas, thereby enhancing the model’s resistance to steganalysis. In the extraction phase, the DCT quantization minimizes secret image loss during compression, facilitating easier information retrieval. The incorporation of scalable modules adds flexibility, allowing for variable-capacity data hiding. Experimental findings validate the high security, large capacity, and high flexibility of our scheme, alongside a marked improvement in binary data recovery post-JPEG compression, underscoring our method’s leading-edge performance.
To solve the privacy leakage and identity island problems in cross-chain interaction, we propose an anti-quantum cross-chain identity authentication approach based on dynamic group signature (DGS-AQCCIDAA) for smart education. The relay-based cross-chain model promotes interconnection in heterogeneous consortium blockchains. DGS is used as the endorsement strategy for cross-chain identity authentication. Our approach can ensure quantum security under the learning with error (LWE) and inhomogeneous small integer solution (ISIS) assumptions, and it uses non-interactive zero-knowledge proof (NIZKP) to protect user identity privacy. Our scheme has low calculation overhead and provides anonymous cross-chain identity authentication in the smart education system.
A federated fusion algorithm of joint multi-Gaussian mixture multi-Bernoulli (JMGM-MB) filters is proposed to achieve optimal fusion tracking of multiple uncertain maneuvering targets in a hierarchical structure. The JMGM-MB filter achieves a higher level of accuracy than the multi-model Gaussian mixture MB (MM-GM-MB) filter by propagating the state density of each potential target in the interactive multi-model (IMM) filtering manner. Within the hierarchical structure, each sensor node performs a local JMGM-MB filter to capture survival, newborn, and vanishing targets. A notable characteristic of our algorithm is a master filter running on the fusion node, which can help identify the origins of state estimates and supplement missed detections. The outputs of all filters are associated into multiple groups of single-target estimates. We rigorously derive the optimal fusion of IMM filters and apply it to merge associated single-target estimates. This optimality is guaranteed by the covariance upper-bounding technique, which can truly eliminate correlations among filters. Simulation results demonstrate that the proposed algorithm outperforms the existing centralized and distributed fusion algorithms in linear and heterogeneous scenarios, and the relative weights of the master and local filters can be adjusted flexibly.
Transformer-based models like large language models (LLMs) have attracted significant attention in recent years due to their superior performance. A long sequence of input tokens is essential for industrial LLMs to provide better user services. However, memory consumption increases quadratically with the increase of sequence length, posing challenges for scaling up long-sequence training. Current parallelism methods produce duplicated tensors during execution, leaving space for improving memory efficiency. Additionally, tensor parallelism (TP) cannot achieve effective overlap between computation and communication. To solve these weaknesses, we propose a general parallelism method called memory-efficient tensor parallelism (METP), designed for the computation of two consecutive matrix multiplications and a possible function between them (O = f(AB) C), which is the kernel computation component in Transformer training. METP distributes subtasks of computing O to multiple devices and uses send/recv instead of collective communication to exchange submatrices for finishing the computation, avoiding producing duplicated tensors. We also apply the double buffering technique to achieve better overlap between computation and communication. We present the theoretical condition of full overlap to help instruct the long-sequence training of Transformers. Suppose the parallel degree is p; through theoretical analysis, we prove that METP provides O(1/p3) memory overhead when not using FlashAttention to compute attention and could save at least 41.7% memory compared to TP when using FlashAttention to compute multi-head self-attention. Our experimental results demonstrate that METP can increase the sequence length by 2.38–2.99 times compared to other methods when using eight A100 graphics processing units (GPUs).
Accurate short-term traffic prediction is essential for improving the efficiency of data transmission in low Earth orbit (LEO) satellite networks. However, traffic values may be missing due to collector failures, transmission errors, and memory failures in complex space environments. Incomplete traffic time series prevent the efficient utilization of data, which can significantly reduce the traffic prediction accuracy. To overcome this problem, we propose a novel spatio-temporal correlation-based incomplete time-series traffic prediction (ITP-ST) model, which consists of two phases: reconstituting incomplete time series by missing data imputation and making traffic prediction based on the reconstructed time series. In the first phase, we propose a novel missing data imputation model based on the improved denoising autoencoder (IDAE-MDI). Specifically, we combine DAE with the Gramian angular summation field (GASF) to establish the temporal correlation between different time intervals and extract the structural patterns from the time series. Taking advantage of the unique spatio-temporal correlation of the LEO satellite network traffic, we focus on improving the missing data initialization method for DAE. In the second phase, we propose a traffic prediction model based on a multi-channel attention convolutional neural network (TP-CACNN) by combining the spatio-temporally correlated traffic of the LEO satellite network. Finally, to achieve the ideal structure of these models, we use the multi-verse optimizer (MVO) algorithm to select the optimal combination of model parameters. Experiments show that the ITP-ST model outperforms the baseline models in terms of traffic prediction accuracy at different data missing rates, which demonstrates the effectiveness of our proposed model.
Recent studies have shown that system capacity is very important for cellular networks. In this paper, we consider maximizing the weighted sum-rate of the cellular network downlink and uplink, where each cell consists of a full-duplex (FD) base station (BS) and half-duplex (HD) users. Federated learning (FL) can train models in the absence of centralized data, which can achieve privacy protection of user data. A low Earth orbit (LEO) satellite edge computing system (LSECS) can be formed by placing the mobile edge computing (MEC) servers on LEO satellites, which greatly increases the processing capacities of the satellites. Therefore, we consider a combination of FL and MEC and propose an FL-based computation offloading algorithm to maximize the weighted sum-rate while ensuring the security of user data. We consider solving the sub-channel assignment and power allocation problems using deep reinforcement learning (DRL) algorithms with excellent global search capabilities. The simulation results show that our proposed algorithm achieves the maximum weighted sum-rate compared with the baseline algorithms and excellent convergence.
Automatic modulation classification (AMC) serves a challenging yet crucial role in wireless communications. Despite deep learning-based approaches being widely used in signal processing, they are challenged by signal distribution variations, especially in various channel conditions. In this paper, we introduce an adversarial transfer framework named frequency-learning adversarial networks (FLANs) based on transfer learning for cross-scenario signal classification. This method uses the stability in the frequency spectrum by introducing a frequency adaptation (FA) technique to incorporate target channel information into source-domain signals. To address the unpredictable interference in the channel, a fitting channel adaptation (FCA) module is used to reduce the difference between the source and target domains caused by variations in the channel environment. Experimental results illustrate that FLANs outperforms state-of-the-art transfer approaches, demonstrating an improved top-1 classification accuracy by about 5.2 percentage points in high signal-to-noise ratio (SNR) scenes on a cross-scenario real collected dataset CSRC2023.