Variational autoencoder (VAE)-based frameworks possess a natural advantage in modeling the shared and private information inherent in multimodal data. However, current models focus on improving the quality of shared representations from the reconstruction perspective, lacking explicit mechanisms to model their underlying semantic structure. In this paper, we propose the multimodal Gaussian mixture variational autoencoder with consistency regularizations, which introduces a Gaussian mixture prior over the shared latent space to enhance its semantic structure and encourage the formation of cluster-aware latent representations. To address the cross-modal inconsistency problem under missing modality conditions, we propose a cluster-guided regularization strategy that enforces the cross-modal consistency using the pseudo-category labels from unsupervised clustering. Additionally, we design a self-supervised contrastive regularization strategy to align semantically similar representations across modalities. Extensive experiments on MNIST-SVHN and MNIST-CDCB datasets demonstrate that our method significantly outperforms prior state-of-the-art models in generation, classification, and retrieval tasks.

Test-time scaling, which is also often referred to as slow-thinking, has been demonstrated to enhance multi-step reasoning in large language models (LLMs). However, despite its widespread utilization, the mechanisms underlying slow-thinking methods remain poorly understood. This paper explores the mechanisms of external slow-thinking from a theoretical standpoint. We begin by examining the snowball error effect within the LLM reasoning process and connect it to the likelihood of correct reasoning using information theory. Building on this, we show that external slow-thinking methods can be interpreted as strategies to mitigate the error probability. We further provide a comparative analysis of popular external slow-thinking approaches, ranging from simple to complex, highlighting their differences and interrelationships. Our findings suggest that the efficacy of these methods is not primarily determined by the specific framework employed, and that expanding the search scope or the model’s internal reasoning capacity may yield more sustained improvements in the long term. We open-source our code at https://github.com/ZyGan1999/Snowball-Errors-and-Probability.

Multimodal image matching is fundamental in computer vision. However, existing methods often struggle to achieve effective cross-modal feature fusion, especially under scale variations and complex scenarios. To this end, we propose PMCMatcher, a Parallel Multi-Scale Cascaded Transformer-Based Network for multimodal feature matching. The core module, the Parallel Multi-Scale Cascaded Transformer, achieves deep interaction and progressive multi-scale fusion through the Selective Multi-Head Linear Attention module and the Cascaded Fusion mechanism. Additionally, the Dynamic Local Feature Enhancement module significantly strengthens the extraction of details by adaptively adjusting convolution kernel weights. To further improve matching accuracy, the Refinement Layer is incorporated to gradually optimize the matching process, enhancing the model’s accuracy and robustness in cross-modal scenarios. Experimental results demonstrate that the proposed method achieves state-of-the-art performance on four representative multimodal image matching datasets, highlighting its superior generalization ability and precise matching accuracy.

Self-knowledge distillation (SKD) enables single-model training by distilling knowledge from the model's own output, eliminating the need for a separate teacher network required in conventional distillation methods. However, current SKD methods focus mainly on replicating common features in the student model, neglecting the extraction of key features that significantly enhance student learning. Inspired by this, we devise a self-knowledge distillation framework entitled Self-Distillation training via Proximal Gradient Optimization or SDPGO, which utilizes gradient information to identify and assign greater weight to features that significantly impact classification performance, enabling the network to learn the most relevant features during training. Specifically, the proposed framework refines the gradient information into a dynamically changing weighting factor to evaluate the distillation knowledge via the dynamic weight adjustment scheme. Meanwhile, we devise the sequential iterative learning module to dynamically optimize knowledge transfer by leveraging historical predictions and real-time gradients, stabilizing training through mini-batch-based KL divergence refinement while adaptively prioritizing task-critical features for efficient self-distillation. Comprehensive experiments on image classification, object detection, and semantic segmentation demonstrate that our method consistently surpasses recent state-of-the-art knowledge distillation techniques. Code is available at: https://github.com/nanxiaotong/SDGPO.

In this paper, we study the mistake bound of online kernel learning on a budget. We propose a new budgeted online kernel learning model, called Ahpatron, which significantly improves the mistake bound of previous work and resolves an open problem related to upper bounds of hypothesis space constraints. We first present an aggressive variant of Perceptron, named AVP, a model without budget, which uses an active updating rule. Then we design a new budget maintenance mechanism, which removes a half of examples, and projects the removed examples onto a hypothesis space spanned by the remaining examples. Ahpatron adopts the above mechanism to approximate AVP. Theoretical analyses prove that Ahpatron has tighter mistake bounds, and experimental results show that Ahpatron outperforms the state-of-the-art algorithms on the same or a smaller budget.

Nonsingular worst-case noise in multiuser broadcast (or “downlink”) communication channels effectively characterizes the users' rate sum. This paper elaborates upon that worst-case-noise characterization for the special singular case. This elaboration is synergistic with the cognitive-radio concept of primary and secondary broadcast-channel users, allowing a mathematical, channel-dependent, and intuitively appealing labeling of user components. The multidimensional singular case adds additional term to the well-known worst-case-noise equation, which term is zero only when the noise is nonsingular. This generalizes a 2004 singular-worst-case-noise result due to Yu [1]. This generalized concept then also enables a new dual multiple-access characterization of a worst-case user-input autocorrelation. Examples appear, along with implications for evolving next-generation cellular-network optimization and policy, focusing on the channel-dependent primary/secondary user-component interpretation.

While constructing polar codes for successive-cancellation decoding can be implemented efficiently by sorting the bit channels, finding optimal polar codes for cyclic-redundancy-check-aided successive-cancellation list (CA-SCL) decoding in an efficient and scalable manner still awaits investigation. This paper first maps a polar code to a unique heterogeneous graph called the polar-code-construction message-passing (PCCMP) graph. Next, a heterogeneous graph-neural-network-based iterative message-passing (IMP) algorithm is proposed which aims to find a PCCMP graph that corresponds to the polar code with minimum frame error rate under CA-SCL decoding. This new IMP algorithm’s major advantage lies in its scalability power. That is, the model complexity is independent of the blocklength and code rate, and a trained IMP model over a short polar code can be readily applied to a long polar code’s construction. Numerical experiments show that IMP-based polar-code constructions outperform classical constructions under CA-SCL decoding. In addition, when an IMP model trained on a length-128 polar code directly applies to the construction of polar codes with different code rates and blocklengths, simulations show that these polar-code constructions deliver comparable performance to the 5G polar codes.

Abstract Copy–move forgery poses a significant threat to social life and has aroused much attention in recent years. Although many copy‐move forgery detection (CMFD) methods have been proposed, the most existing CMFD methods are short of adaptability in detecting images, which leads to the limitation on detection effects. To solve this problem, the paper proposes a novel keypoint‐based CMFD method: second‐keypoint matching and double adaptive filtering (SMDAF). Motivated by image matching based on keypoint, the second‐keypoint matching method is designed to match keypoints extracted from copy–move forgery images, which can be used for both the single‐CMFD and the multiple‐CMFD. Then, a double adaptive filter (DAF) based on the AdaLAM algorithm and the KANN‐DBSCAN clustering algorithm to filter wrong keypoint matches adaptively are proposed, according to the distinct distribution of keypoints in each image. Finally, the forgery regions are presented by finding their convex hulls and padding them. Compared with existing methods, extensive experiments show that the SMDAF method significantly provides more efficiency in detecting images under simulated real‐world conditions, has better robustness when facing images with different post‐treatment attacks, and is more effective in distinguishing images that look copy–move forged but are real.

This paper formulates the polar-code construction problem for the successive-cancellation list (SCL) decoder as a maze-traversing game, which can be solved by reinforcement-learning techniques. The proposed method provides a novel technique for polar-code construction that no longer depends on sorting and selecting bit-channels by reliability, as in most current algorithms. Instead, this technique decides whether the input bits should be frozen in a purely sequential manner. The equivalence of optimizing the polar-code construction for the SCL decoder under this technique and maximizing the expected reward of traversing a maze is drawn. Simulation results show that the standard polar-code constructions that are designed for the successive-cancellation decoder are no longer optimal for the SCL decoder with respect to the frame error rate (FER). In contrast, the proposed game-based construction method finds code constructions that have similar or lower FER for various code lengths and various list sizes of the SCL decoder, compared to the state-of-the-art construction methods. The advantage of the game-based constructions over the standard constructions increases with the channel signal-to-noise ratio and the list size of SCL decoding. Moreover, the learning is highly efficient in terms of the number of required training samples and computational operations.

This paper formulates the polar-code construction problem for the successive-cancellation list (SCL) decoder as a maze-traversing game, which can be solved by reinforcement learning techniques. The proposed method provides a novel technique for polar-code construction that no longer depends on sorting and selecting bit-channels by reliability. Instead, this technique decides whether the input bits should be frozen in a purely sequential manner. The equivalence of optimizing the polar-code construction for the SCL decoder under this technique and maximizing the expected reward of traversing a maze is drawn. Simulation results show that the standard polar-code constructions that are designed for the successive-cancellation decoder are no longer optimal for the SCL decoder with respect to the frame error rate. In contrast, the simulations show that, with a reasonable amount of training, the game-based construction method finds code constructions that have lower frame-error rate for various code lengths and decoders compared to standard constructions.

An adaptive bandwidth allocation mechanism for the calendar associated with the Flexible Ethernet (FlexE) standard is proposed. The proposed method bases the FlexE calendar design on the clients' real transmit data rates. In particular, the proposed method treats clients with very low bandwidth utilization as minor clients and allows them to transmit in an opportunistic manner. Experiments on real Ethernet packet traces indicate that by using the proposed calendar scheme to allocate bandwidth to clients, the total required FlexE bandwidth can be reduced by up to 60% while meeting packet drop requirements.

This paper proposes to use a deep neural network (DNN)- based symbol detector for mmWave systems such that channel state information (CSI) acquisition can be bypassed. In particular, we consider a sliding bidirectional recurrent neural network (BRNN) architecture that is suitable for the long memory length of typical mmWave channels. The performance of the DNN detector is evaluated in comparison to that of the Viterbi detector. The results show that the performance of the DNN detector is close to that of the optimal Viterbi detector with perfect CSI, and that it outperforms the Viterbi algorithm with CSI estimation error. Further experiments show that the DNN detector is robust to a wide range of noise levels and varying channel conditions, and that a pretrained detector can be reliably applied to different mmWave channel realizations with minimal overhead.

In a social network, influence maximization is the problem of identifying a set of users that own the maximum influence ability across the network. In this paper, a novel credit distribution (CD) based model, termed as the multi-action CD (mCD) model, is introduced to quantify the influence ability of each user. Compared to existing models, the new model can work with practical datasets where one type of action is recorded for multiple times. Based on this model, influence maximization is formulated as a submodular maximization problem under a knapsack constraint, which is NP-hard. An efficient streaming algorithm is developed to achieve$(\frac{1}{3}-\epsilon)$approximation of the optimality. Experiments conducted on real Twitter dataset demonstrate that the mCD model enjoys high accuracy compared to the conventional CD model in estimating the total number of people who get influenced in a social network. Furthermore, compared to the greedy algorithm, the proposed single-pass streaming algorithm achieves similar performance in terms of influence maximization, while running several orders of magnitude faster.

Recently, owing to the advances in the self-interference cancellation technology, the in-band full-duplex (FD) capability has been demonstrated at Wi-Fi range. However, the simultaneous uplink (UL) and downlink (DL) transmission may lead to inter-user interference (IUI) and result in decoding failure. Spectrum efficiency should also be considered in the construction process of the FD transmission. In this paper, we propose a hybrid half-duplex/FD MAC protocol based on a two-fold RTS/CTS contention resolution mechanism, in order to fully exploit the channel access opportunities provided by the simultaneous UL and DL transmissions. The noteworthy features of the proposed protocol lie in the following two aspects. First, the protocol provides the flexibility for the AP to decide the probability of constructing FD transmission, and then it adopts a second-fold of RTS/CTS mechanism to prevent the constructed transmission from being affected by the IUI. The second-fold contention and the probability of constructing FD transmission are optimized separately to maximize the spectrum efficiency given different transmission demands. Simulation results show that the proposed MAC protocol achieves higher capacity compared with previous works, and the hybrid characteristic enables the FD Wi-Fi networks to meet with different system requirements.

Device-to-Device (D2D) communication, which enables direct communication between nearby mobile devices, is an attractive technique to improve spectrum efficiency by reusing licensed spectrum. Nowadays, LTE-unlicensed (LTE-U) emerges to extend the cellular network to the unlicensed spectrum to alleviate the spectrum scarcity issue. In this paper, D2D communication is allowed to work in unlicensed spectrum (D2D-U) as an underlay of the cellular network for further booming the network capacity. A sensing-based protocol is designed to support the unlicensed channel access for both LTE and D2D users, based on which we investigate the subchannel allocation problem to maximize the total sum rate while taking into account their interference to the existing Wi-Fi systems. Specifically, we formulate the subchannel allocation as a many-to-many matching problem with externalities, and develop an iterative usersubchannel swap algorithm. Analytical and simulation results show that the proposed D2D-U scheme can significantly improve the network capacity.

Device-to-device (D2D) communication, which enables direct communication between nearby mobile devices, is an attractive add-on component to improve spectrum efficiency and user experience by reusing licensed cellular spectrum in 5G system. In this paper, we propose to enable D2D communication in unlicensed spectrum (D2D-U) as an underlay of the uplink LTE network for further booming the network capacity. A sensing-based protocol is designed to support the unlicensed channel access for both LTE and D2D users. We further investigate the subchannel allocation problem to maximize the sum rate of LTE and D2D users while considering their interference to the existing Wi-Fi systems. Specifically, we formulate the subchannel allocation as a many-to-many matching problem with externalities, and develop an iterative user-subchannel swap algorithm. Analytical and simulation results show that the proposed D2D-U scheme can significantly improve the system sum rate.

Recently, some CSMA/CD-alike protocols have been proposed for full-duplex (FD) WiFi networks, in which users are able to monitor the channel and transmit data simultaneously so as to avoid data collisions and improve the spectrum efficiency. However, along with its benefits, the residual self- interference (RSI) brought by FD becomes a new challenge for the network. As the RSI increases with the transmit power, the sensing performance degrades. Thus, on the one hand, increased transmit power of individual user will promote the system throughput. On the other hand, as the user who raises its transmit power suffers from more detection failure, its transmit probability will be higher and leads to a suppression of that of other users, so the system fairness may decrease. Consequently, a tradeoff between the system throughput and fairness emerges. In this paper, we provide theoretical analysis of the system throughput and fairness, and reveal their relationship with the power profile of users in FD WiFi networks. We formulate the power control problem in FD WiFi networks as a non-cooperative game, for which we propose a distributed power control mechanism considering both system throughput and fairness. Simulation results validate the fairness-throughput tradeoff for the proposed power control mechanism.

Featured by centralized processing and cloud-based infrastructure, cloud radio access network (C-RAN) has emerged as a promising solution to handle the data proliferation in future wireless networks. However, the attractive capacity enhancement brought by large-scale centralized processing comes along with increased computing resource requirement in the baseband unit (BBU) pool. Thus, computing resource as another dimension of manageable resource needs to be considered in resource allocation and C-RAN system design. In this paper, we first characterize the relationship between PHY transmission characteristics and the required computing resource in the BBU pool. Based on this, we propose a feasible algorithm to maximize the network sum-rate under limited computing resource constraint, which is a binary-integer non-linear programming (BINLP) problem with non-convex constraints in nature. Numerical results show the significant impact of computing resource on both user-RRH association strategy and the achievable sum-rate performance.

The idea of in-band full-duplex (FD) communications revives in recent years owing to the significant progress in the self-interference cancellation and hardware design techniques, which offers the potential to double spectral efficiency. However, the adaptations from lower to upper layers are highly demanded in the design of FD communication systems. In this paper, we first propose a novel medium access control (MAC) protocol on a single channel using FD techniques that allows transmitters to monitor the channel usage while transmitting, and backoff when collision happens. Specifically, imperfect sensing brought by residual self-interference (RSI) in the physical layer is taken into account in the design of the protocol, and throughput is analytical derived. Then, we extend the protocol to multichannel scenario and propose a multichannel selection mechanism based on the rate and occupancy history of each channel for FD users. Simulation results show the effectiveness of the channel selection protocol and indicate that the total throughput of the proposed FD-MAC protocol can significantly outperforms the greedy channel selection strategy based on the CSMA access mechanism.

In the dynamic spectrum access (DSA) paradigm for cognitive radio networks (CRNs), one of the commonly used Medium Access Control (MAC) schemes is designed on basis of the popular carrier sensing multiple access with collision avoidance. However, this proposal suffers from two major problems that may significantly decrease the system performance: (1) collision among the secondary users (SUs) can hardly be detected, thus leading to the secondary transmission failures, and (2) SUs cannot abort transmission when collision occurs, making the long collision duration possible. In this paper, we propose a new cognitive MAC protocol for efficient DSA based on full-duplex CRNs (FD-CRNs), where SUs are able to perform simultaneous spectrum sensing and data transmission owing to full-duplex techniques. Specifically, SUs can detect the collision during transmission, so as to reduce the collision time and improve secondary network performance. Analytical results include the derivations of key design parameters such as the collision ratio with the PU, spectrum usage ratio, optimal contention window size, and the performance comparisons with the conventional DSA in half-duplex CRNs (HD-CRNs), which are further confirmed by simulation results.

Recently, the development in full-duplex communications has offered a great opportunity to perform simultaneous spectrum sensing and spectrum access in cognitive radio networks. In this paper, we consider a cognitive cellular network, in which the secondary base station (SBS) is a full-duplex device that can simultaneously sense the primary spectrum and transmit to the secondary users. We show that the power allocation of the SBS can affect both the sensing performance and the transmission capacity, and thus, we jointly consider the power allocation problem in the spectrum management process. First, we formulate the considered problem as a 3-dimensional matching problem and prove its NP-hardness. Then, we propose an approximate solution by extending a 2-dimensional matching algorithm. The simulation results show that the proposed algorithm can highly increase the secondary throughput of the SBS, compared with the greedy algorithm and the random algorithm.

With the booming exploitation of WiFi networks based on the conventional CSMA/CA access scheme, the frequent long collision period becomes unbearable. To better utilize the WiFi spectrum, in this poster, we propose a novel WiFi protocol with the assistance of full-duplex (FD) technique that allows users to sense the spectrum while transmitting, and stop transmission once collision is detected. Analytical and simulated results show that the average collision length is sharply reduced and the normalized throughput can be significantly improved by the proposed FD-WiFi protocol compared with the conventional CSMA/CA.

In traditional cognitive radio networks, secondary users (SUs) typically access the spectrum of primary users (PUs) by a two-stage "listen-before-talk" (LBT) protocol, i.e., SUs sense the spectrum holes in the first stage before transmit in the second stage. In this paper, we propose a novel "listen-and-talk" (LAT) protocol with the help of the full-duplex (FD) technique that allows SUs to simultaneously sense and access the vacant spectrum. Analysis of sensing performance and SU's throughput are given for the proposed LAT protocol. And we find that due to self-interference caused by FD, increasing transmitting power of SUs does not always benefit to SU's throughput, which implies the existence of a power-throughput tradeoff. Besides, though the LAT protocol suffers from self-interference, it allows longer transmission time, while the performance of the traditional LBT protocol is limited by channel spatial correction and relatively shorter transmission period. To this end, we also present an adaptive scheme to improve SUs' throughput by switching between the LAT and LBT protocols. Numerical results are provided to verify the proposed protocol and the theoretical results.