Transformers excel at sequence modeling but face quadratic complexity, while linear attention offers improved efficiency but often compromises recall accuracy over long contexts.In this work, we introduce Native Hybrid Attention (NHA), a novel hybrid architecture of linear and full attention that integrates both intra & inter-layer hybridization into a unified layer design.NHA maintains longterm context in key-value slots updated by a linear RNN, and augments them with shortterm tokens from a sliding window.A single softmax attention operation is then applied over all keys and values, enabling pertoken and per-head context-dependent weighting without requiring additional fusion parameters.The inter-layer behavior is controlled through a single hyperparameter, the sliding window size, which allows smooth adjustment between purely linear and full attention while keeping all layers structurally uniform.Experimental results show that NHA surpasses Transformers and other hybrid baselines on recall-intensive and commonsense reasoning tasks.Furthermore, pretrained LLMs can be structurally hybridized with NHA, achieving competitive accuracy while delivering significant efficiency gains.Code is available at https://github.com/JusenD/NHA.

State Space Models (SSMs) have emerged as a powerful framework for sequence modeling in recent years. By approximating continuous dynamical systems and applying discretization techniques, SSMs are particularly well-suited for modeling time-series data. However, despite their growing popularity, most existing applications of SSMs in time-series forecasting treat the models as black boxes. Besides, the underlying mechanisms that contribute to their effectiveness remain unclear, and common claims regarding their advantages in efficiency and expressiveness are not fully substantiated. To address these gaps, this paper establishes a theoretical connection between SSMs and classical spectral transformations from signal processing, thereby providing a more interpretable foundation. Furthermore, we conduct comprehensive ablation studies to examine the properties of different SSM configurations. Our goal is to offer both theoretical insight and empirical guidance for future research on SSM-based approaches in time-series forecasting.

Transformers with linear recurrent modeling offer linear-time training and constant-memory inference. Despite their demonstrated efficiency and performance, pretraining such non-standard architectures from scratch remains costly and risky. The linearization of large language models (LLMs) transforms pretrained standard models into linear recurrent structures, enabling more efficient deployment. However, current linearization methods typically introduce additional feature map modules that require extensive fine-tuning and overlook the gating mechanisms used in state-of-the-art linear recurrent models. To address these issues, this paper presents Liger, short for Linearizing LLMs to gated recurrent structures. Liger is a novel approach for converting pretrained LLMs into gated linear recurrent models without adding extra parameters. It repurposes the pretrained key matrix weights to construct diverse gating mechanisms, facilitating the formation of various gated recurrent structures while avoiding the need to train additional components from scratch. Using lightweight fine-tuning with Low-Rank Adaptation (LoRA), Liger restores the performance of the linearized gated recurrent models to match that of the original LLMs. Additionally, we introduce Liger Attention, an intra-layer hybrid attention mechanism, which significantly recovers 93% of the Transformer-based LLM performance at 0.02% pre-training tokens during the linearization process, achieving competitive results across multiple benchmarks, as validated on models ranging from 1B to 8B parameters.

Depression detection remains challenged by generalized modeling approaches that fail to account for individual heterogeneity. To address this, the Multimodal Personality-aware Depression Detection (MPDD) Challenge introduced personalized features into the modeling process, aiming to better capture individual variability. However, the baseline models still exhibit two critical limitations: the neglect of textual semantics embedded in audio, and inconsistent predictions for the same subject across tasks and samples. Motivated by these limitations, we introduce HOPE (Hierarchical fusion for Optimized and Personality-aware Estimation of Depression), a unified framework for consistent, subject-level depression estimation. HOPE first employs a Latent Semantic Projection (LSP) module to reconstruct textual semantics from audio features when transcripts are unavailable. It then introduces a consistency-aware integration mechanism that hierarchically fuses multi-branch predictions to resolve inter-task and inter-sample contradictions. HOPE achieved first place in the MPDD Challenge Young Track, demonstrating strong cross-modal learning capabilities and consistent, subject-level depression prediction.

Recent efficient sequence modeling methods, such as Gated DeltaNet, TTT, and RWKV-7, have achieved performance improvements by supervising the recurrent memory management through the Delta learning rule. Unlike previous state-space models (e.g., Mamba) and gated linear attentions (e.g., GLA), these models introduce interactions between the recurrent state and the key vector, resulting in a bilinear recursive structure. In this paper, we first introduce the concept of Bilinear RNNs with a comprehensive analysis on the advantages and limitations of these models. Then based on the closed-loop control theory, we propose a novel Bilinear RNN variant named Comba, which adopts a scalar-plus-low-rank state transition, with both state feedback and output feedback corrections. We also implement a hardware-efficient chunk-wise parallel kernel in Triton and train models with 340M/1.3B parameters on a large-scale corpus. Comba demonstrates its superior performance and computation efficiency on both language modeling and vision tasks.

The 5G core network (5GC) architecture based on Service-Based Architecture (SBA) has brought unprecedented flexibility and innovation. However, this architecture also comes with potential security challenges. The integration of different signaling protocols and the complexity of virtualization in 5GC have increased security risks within the core network. The concept of zero trust is considered a new solution, and Software-Defined Perimeter (SDP) represents a best practice for zero trust. In this paper, we propose a 5GC-SDP architecture that provides secure communication within the core network through authentication-based methods. Single Package Authorization (SPA) is the key technology of this study. Only Network Functions (NF) that have been authenticated and authorized by SPA can access each other. To the best of our knowledge, this is the first study to combine SDP with StandAlone (SA) 5GC. At the same time, we fully consider that although SPA technology can withstand most DoS attacks, DoS attacks caused by SPA packets will still become a problem. Therefore, we design a SPA enhancement module, and machine learning algorithms are used for SPA-DoS detection. We have conducted practical exploration on the proposed 5GC-SDP architecture and implemented testing on port scanning, DoS, and DDoS attacks. The experiments have shown that 5GC-SDP achieves enhanced protection of the core network by limiting network exposure and implementing fine-grained access control.

In long-term time series forecasting (LTSF) tasks, an increasing number of works have acknowledged that discrete time series originate from continuous dynamic systems and have attempted to model their underlying dynamics. Recognizing the chaotic nature of real-world data, our model, Attraos, incorporates chaos theory into LTSF, perceiving real-world time series as low-dimensional observations from unknown high-dimensional chaotic dynamical systems. Under the concept of attractor invariance, Attraos utilizes non-parametric Phase Space Reconstruction embedding along with a novel multi-resolution dynamic memory unit to memorize historical dynamical structures, and evolves by a frequency-enhanced local evolution strategy. Detailed theoretical analysis and abundant empirical evidence consistently show that Attraos outperforms various LTSF methods on mainstream LTSF datasets and chaotic datasets with only one-twelfth of the parameters compared to PatchTST.

Ethereum leverages ECDSA as the digital signature scheme to validate transactions. From the provable security standpoint, ECDSA built on an 80-bit security Elliptic Curve group can achieve at most 50-bit concrete security, rather than 80-bit security, due to its reduction loss for$2^{30}$signature queries in security analysis. The state-of-the-art ECDSA scheme comes with no de facto formal security guarantee. Although there have been many signatures with higher concrete security, their structures are quite different from ECDSA and a total replacement of the signature field in Ethereum will incur high deployment cost. In this work, we present EthereumX without compromising the signature structure in Ethereum while achieves better security. The security gain is built on top of a new technique named randomness preprocessing module (RPM), which can securely pre-generate and verify randomness with the help of Ethereum. Calling RPM allows to pre-select randomness, which will be used for the subsequent signature, and to verify the randomness, assuring that it is previously generated. We give an instantiation with formal security guarantee and prove that it can be improved to 80-bit concrete security under the same discrete logarithm assumption as ECDSA. From this instantiated scheme, we implement EthereumX via a deployment into a locally simulated network. Experiment results show that EthereumX costs 5 seconds for a block generation which is equal to Ethereum, and generates/verifies at least$17017/10623$transactions per second that is practical enough in application, even if they are slightly slower than Ethereum which generates/verifies at least$17908/11257$transactions per second. We also mention that RMP can be applied to other DL-based signatures for the security improvement.

With expansion of neural network (NN) applications lowering their hardware implementation cost becomes an urgent task especially in back-end applications where the power-supply is limited. Stochastic computing (SC) is a promising solution to realize low-cost hardware designs. Implementation of matrix multiplication has been a bottleneck in previous stochastic neural networks (SC-NNs). In this paper, we introduce spintronic components into the design of SC-NNs. A novel spin-CMOS matrix multiplier is proposed in which the stochastic multiplications are performed by CMOS AND gates while the sum of products is implemented by spintronic compressor gates. The experimental results indicate that compared to the conventional binary implementations the proposed hybrid spin-CMOS architecture can achieve over 125x, 4.5x and 43x; reduction in terms of power, energy and area consumptions, respectively. Moreover, compared to previous CMOS-based SC-NNs, our design saves the power by 3.1x - 7.3x, reduces energy consumption by 3.1x - 7.3x and decreases area by 1.4x - 7.6x while maintaining similar recognition rates.

This paper investigates a synchronous control method with variable proportion for multi-axis servo system based on sliding mode variable structure to solve the inherent shortcomings of parallel synchronous uncoupled control for multi-servo motors. The method is an improvement of the ring-coupled control strategy and is thus applicable to multi-axis proportional synchronous control. No-salient pole permanent magnet synchronous servo motors are taken as the controlled objects. A sliding mode controller is designed according to the state equation of the mathematical model. The rationality of the designed controller is proven by Lyapunov stability theory, and the method is compared with a parallel synchronous control strategy, which is widely used in the object of this research. Finally, the feasibility and superiority of the method in a multi-axis synchronous control system with variable proportion is confirmed by MATLAB simulation and experimental verification. The method is applied to a three-degree-of-freedom servo system, and makes the theoretical research have the corresponding practical engineering application value.

This paper presents a novel approach based on spectral geometry to quantify and visualize non-isometric deformations of 3D surfaces by mapping two manifolds. The proposed method can determine multi-scale, non-isometric deformations through the variation of Laplace-Beltrami spectrum of two shapes. Given two triangle meshes, the spectra can be varied from one to another with a scale function defined on each vertex. The variation is expressed as a linear interpolation of eigenvalues of the two shapes. In each iteration step, a quadratic programming problem is constructed, based on our derived spectrum variation theorem and smoothness energy constraint, to compute the spectrum variation. The derivation of the scale function is the solution of such a problem. Therefore, the final scale function can be solved by integral of the derivation from each step, which, in turn, quantitatively describes non-isometric deformations between two shapes. To evaluate the method, we conduct extensive experiments on synthetic and real data. We employ real epilepsy patient imaging data to quantify the shape variation between the left and right hippocampi in epileptic brains. In addition, we use longitudinal Alzheimer data to compare the shape deformation of diseased and healthy hippocampus. In order to show the accuracy and effectiveness of the proposed method, we also compare it with spatial registration-based methods, e.g., non-rigid Iterative Closest Point (ICP) and voxel-based method. These experiments demonstrate the advantages of our method.

This paper presents a registration-free method based on geometry spectrum for mapping two shapes. Our method can quantify and visualize the surface deformation by the variation of Laplace-Beltrami spectrum of the object. In order to examine our method, we employ synthetic data that has non-isometric deformation. We have also applied our method to quantifying the shape variation between the left and right hippocampus in epileptic human brains. The results on both synthetic and real patient data demonstrate the effectiveness and accuracy of our method.

In order to visualize and analyze complex collective data, complicated geometric structure of each data is desired to be mapped onto a canonical domain to enable map-based visual exploration. This paper proposes a novel volume-preserving mapping and registration method which facilitates effective collective data visualization. Given two 3-manifolds with the same topology, there exists a mapping between them to preserve each local volume element. Starting from an initial mapping, a volume restoring diffeomorphic flow is constructed as a compressible flow based on the volume forms at the manifold. Such a flow yields equality of each local volume element between the original manifold and the target at its final state. Furthermore, the salient features can be used to register the manifold to a reference template by an incompressible flow guided by a divergence-free vector field within the manifold. The process can retain the equality of local volume elements while registering the manifold to a template at the same time. An efficient and practical algorithm is also presented to generate a volume-preserving mapping and a salient feature registration on discrete 3D volumes which are represented with tetrahedral meshes embedded in 3D space. This method can be applied to comparative analysis and visualization of volumetric medical imaging data across subjects. We demonstrate an example application in multimodal neuroimaging data analysis and collective data visualization.

Given an N-point sequence, finding its k largest components in the frequency domain is a problem of great interest. This problem, which is usually referred to as a sparse Fourier Transform, was recently brought back on stage by a newly proposed algorithm called the sFFT. In this paper, we present a parallel implementation of sFFT on both multi-core CPUs and GPUs using a human voice signal as a case study. Using this example, an estimate of k for the 3dB cutoff points was conducted through concrete experiments. In addition, three optimization strategies are presented in this paper. We demonstrate that the multi-core-based sFFT achieves speedups of up to three times a single-threaded sFFT while a GPU-based version achieves up to ten times speedup. For large scale cases, the GPU-based sFFT also shows its considerable advantages, which is about 40 times speedup compared to the latest out-of-card FFT implementations [2].

Parameterization of complex surfaces constitutes a major means of visualizing highly convoluted geometric structures as well as other properties associated with the surface. It also enables users with the ability to navigate, orient, and focus on regions of interest within a global view and overcome the occlusions to inner concavities. In this paper, we propose a novel area-preserving surface parameterization method which is rigorous in theory, moderate in computation, yet easily extendable to surfaces of non-disc and closed-boundary topologies. Starting from the distortion induced by an initial parameterization, an area restoring diffeomorphic flow is constructed as a Lie advection of differential 2-forms along the manifold, which yields equality of the area elements between the domain and the original surface at its final state. Existence and uniqueness of result are assured through an analytical derivation. Based upon a triangulated surface representation, we also present an efficient algorithm in line with discrete differential modeling. As an exemplar application, the utilization of this method for the effective visualization of brain cortical imaging modalities is presented. Compared with conformal methods, our method can reveal more subtle surface patterns in a quantitative manner. It, therefore, provides a competitive alternative to the existing parameterization techniques for better surface-based analysis in various scenarios.