Multimodal visual language models are gaining prominence in open-world applications, driven by advancements model architectures, training techniques, and high-quality data. However, their performance is often limited insufficient task-specific data, leading to poor generalization biased outputs. Existing efforts increase task diversity fine-tuning datasets hindered the labor-intensive process of manual labeling, which typically produces only a few hundred types. To address this, we propose...

Enhancing the fine-grained instance spatiotemporal motion perception capabilities of Video Large Language Models is crucial for improving their temporal and general video understanding. However, current models struggle to perceive detailed complex motions. To address these challenges, we have made improvements from both data model perspectives. In terms data, meticulously curated iMOVE-IT, first large-scale instance-motion-aware instruction-tuning dataset. This dataset enriched with...

The explosively growing demands for mobile traffic services bring both challenges and opportunities to wireless networks. Wireless network virtualization is proposed as the main evolution path toward forthcoming fifth generation (5G) cellular In this paper, we propose a software defined virtualized (SDV) architecture enabling multi-flow transmission with multiple infrastructure providers (InPs) virtual operators (MVNOs). order ensure heterogeneity, formulate resource allocation problem...

The explosively growing demands for mobile traffic service bring both challenges and opportunities to wireless networks, among which, network virtualization is proposed as the main evolution towards 5G. In this paper, we first propose a Software Defined Network (SDN) based architecture enabling multi-flow transmission in order save capital expenses (CapEx) operation (OpEx) significantly with multiple Infrastructures Providers (InPs) Mobile Virtual Operators (MVNOs). We formulate virtual...

To accommodate the explosively growing demands for mobile traffic service, wireless network virtualization is proposed as main evolution towards 5G. In this work, a novel contract theoretic incentive mechanism to study how manage resources and provide services users in virtualized networks. We consider that infrastructure providers (InPs) own physical networks virtual operator (MVNO) has service information of needs lease radio providing services. particular, we utilize approach model...

In this paper, resource allocation (RA) problem in heterogeneous Software Defined Network (SDN) with infrastructure sharing platform among multiple network service providers (NSPs) is studied. The considered modeled as a reverse combinatorial auction (R-CA) game, which takes competitiveness and fairness of different NSPs into account. RA associated personal QoS requirement optimized by maximizing the social welfare, demonstrated to be total system throughput. By exploiting properties...

Wireless body area networks (WBANs) have been deployed in numerous applications, where the most common communication technology is Bluetooth. Bluetooth uses numeric comparison protocol (NCP) to negotiate session keys based on elliptic curve cryptography (ECC) and Out-of-Band (OoB) channels. However, scalar multiplication of ECC a heavy computing operation for devices WBANs. To address this issue, we propose lightweight secure NCP (LSNCP) which requires less than New logic expressions rules...

The rapidly increasing mobile traffic demand poses both new communication requirements and challenges on existing networks in terms of technologies business models. Wireless network virtualization is a promising technology to provide service-based architecture contract theory powerful framework from microeconomics for providing tools model incentive mechanisms. In this work, novel theoretic mechanism proposed study how services multiple users the wireless virtualized networks. Infrastructure...

The use of artificial intelligence and machine learning is recognized as the key enabler for 5G mobile networks which would allow service providers to tackle network complexity ensure security, reliability allocation necessary resources their customers in a dynamic, robust trustworthy way. Dependability future generation on accurate timely performance its components means that disturbance functionality these may have negative impact entire network. As result, there an increasing concern...

The introduction of 5G technology along with the exponential growth in connected devices is expected to cause a challenge for efficient and reliable network resource allocation. Network providers are now required dynamically create deploy multiple services which function under various requirements different vertical sectors while operating on top same physical infrastructure. recent progress artificial intelligence machine learning theorized be potential answer arising allocation challenges....

In this paper, resource allocation for energy efficiency in heterogeneous Software Defined Network (SDN) with multiple network service providers (NSPs) is studied. The considered problem modeled as a reverse combinatorial auction game, which takes different quality of (QoS) requirements into account. selection associated power optimized by maximizing the data transmission. By exploiting properties fractional programming, resulting non-convex Winner Determination Problem (WDP) transformed an...

Distributed deep learning (DDL) uses a cluster of servers to train models in parallel. This has been applied multiplicity problems, e.g. online advertisement, friend recommendations. However, the distribution training means that communication network becomes key component system performance. In this paper, we measure Alibaba's DDL system, with focus on understanding bottlenecks introduced by network. Our finding is communications overhead surprisingly large impact To explore this, analyse...

Software-defined networking (SDN) faces many of the same security threats as traditional networks. The separation SDN control plane and data makes controller more vulnerable to cyber attacks. conventional "perimeter defense" network model cannot prevent lateral movement attacks caused by malicious insider users or hardware software vulnerabilities. "zero trust architecture" has become a new protect enterprise security. In this article, we propose an intelligent zero-trust framework IZTSDN...

High dynamic range (HDR) displays are enabling new advances in visual psychophysics, but commercial HDR both expensive, and difficult to calibrate colorimetrically. Homebrew incorporating LCD panels digital projectors relatively inexpensive can be calibrated, building such requires sophisticated technical skills. We have developed a low-cost, color-calibrated display for vision research that constructed used by researchers without the need specialized equipment or advanced engineering...

In outdoor low-level vision systems, not only is the resolution of imaging system important, but rain corrupts visibility scenes and may cause computer systems to fail. We present a deep convolutional neural network (CNN) architecture for simultaneously performing single-image super-resolution removal. Instead learning an end-to-end mapping between low-resolution rainy images high-resolution clean in original image space, we train our detail i.e., space obtained by high-pass filtering image....

Reinforcement learning (RL) has been widely used in training large language models~(LLMs) for preventing unexpected outputs, \eg reducing harmfulness and errors. However, existing RL methods mostly adopt the instance-level reward, which is unable to provide fine-grained supervision complex reasoning tasks, can not focus on few key tokens that lead incorrectness. To address it, we propose a new method named \textbf{RLMEC} incorporates generative model as reward model, trained by erroneous...

The Mixture-of-Experts (MoE) has gained increasing attention in the study of Large Vision-Language Models (LVLMs). It uses a sparse model to replace dense model, achieving comparable performance while activating fewer parameters during inference, thus significantly reducing inference cost. Existing MoE methods LVLMs encourage different experts handle tokens, and they employ router predict routing for each token. However, predictions are based solely on sample features do not truly reveal...

Large language models often encounter challenges with static knowledge and hallucinations, which undermine their reliability. Retrieval-augmented generation (RAG) mitigates these issues by incorporating external information. However, user queries frequently contain noise intent deviations, necessitating query rewriting to improve the relevance of retrieved documents. In this paper, we introduce DMQR-RAG, a Diverse Multi-Query Rewriting framework designed performance both document retrieval...

Mobile stations supporting the 802.11u standard can access WLAN automatically when they are within coverage of network service provided by this WLAN. To achieve goal, need to keep "on" states includingidleandactiveall time. However, studies have noted that idleness often lead considerable power consumption. Although conventional saving mode (PSM) provide energy effect some extent, its own disadvantage leads lower efficiency number accessing target In paper, we propose a Schedule-Aware PSM...

Deep neural networks have gained great success due to the increasing amounts of data, and diverse effective network designs. However, it also brings a heavy computing burden as amount training data is proportional time. In addition, well-behaved model requires repeated trials different structure designs hyper-parameters, which may take large time even with state-of-the-art (SOTA) hyper-parameter optimization (HPO) algorithms architecture search (NAS) algorithms. this paper, we propose an...