Search

Abstract In recent years, large-scale artificial intelligence (AI) models have become a focal point in technology, attracting widespread attention and acclaim. Notable examples include Google’s BERT and OpenAI’s GPT, which have scaled their parameter sizes to hundreds of billions or even tens of trillions. This growth has been accompanied by a significant increase in the amount of training data, significantly improving the capabilities and performance of these models. Unlike previous reviews, this paper provides a comprehensive discussion of the algorithmic principles of large-scale AI models and their industrial applications from multiple perspectives. We first outline the evolutionary history of these models, highlighting milestone algorithms while exploring their underlying principles and core technologies. We then evaluate the challenges and limitations of large-scale AI models, including computational resource requirements, model parameter inflation, data privacy concerns, and specific issues related to multi-modal AI models, such as reliance on text-image pairs, inconsistencies in understanding and generation capabilities, and the lack of true “multi-modality”. Various industrial applications of these models are also presented. Finally, we discuss future trends, predicting further expansion of model scale and the development of cross-modal fusion. This study provides valuable insights to inform and inspire future future research and practice.

Based on the double-periodic characteristics of a sine waveguide, the dispersion equation, interaction impedance, and transmission losses of slow-wave structures are derived. The dispersion equation indicates that the waves propagating in the sine waveguide should belong to two mode types, which can exist independently rather than as “pseudo-modes.” The dispersion equation, interaction impedance, and transmission losses are obtained using a theoretical model. A 220 GHz traveling wave tube (TWT) is used as an illustrative example to verify the validity of the analytical model. The calculated results show that the dispersion curve is in good agreement with that given by the Ansys high-frequency simulation software over the entire frequency range and that the theoretical numerical calculation time is less than 2% of that of the Ansys software package. In addition, the results demonstrate that the interaction impedance of the symmetric φ0 + π mode is much higher than that of the asymmetric φ0 mode. Therefore, the symmetric φ0 + π mode should be used for propagation in sine waveguide traveling-wave tubes. Moreover, when the conductivity is set at 1.6 × 107 S/m, the loss of the sine waveguide is ∼1.15 dB/cm for 220 GHz. The equivalent conductivity of the metal can be used appropriately in the design of the mm-wave and the THz TWT.

An improved algorithm has been proposed to address the challenges encountered in object detection using visible and thermal infrared images. These challenges include the diversity of object detection perspectives, deformation of the object, occlusion, illumination, and detection of small objects. The proposed algorithm introduces the concept of contrastive learning into the YOLOv5 object detection network. To extract image features for contrastive loss calculation, object and background image regions are randomly cropped from image samples. The contrastive loss is then integrated into the YOLOv5 network, and the combined loss function of both object detection and contrastive learning is used to optimize the network parameters. By utilizing the strategy of contrastive learning, the distinction between the background and the object in the feature space is improved, leading to enhanced object detection performance of the YOLOv5 network. The proposed algorithm has shown pleasing detection results in both visible and thermal infrared images.

Edge computing is a new computing paradigm which distributes tasks to edge networks for processing, it provides effective support for various mobile applications to meet their rapid response requirement. Unmanned Aerial Vehicle (UAV) has been widely used in emergency rescue, mapping, etc., with the advantages of flexible deployment and rapid movement. However, on one hand, mobile applications will be terminated when the battery energy is exhausted. On the other hand, mobile applications will be out of service when mobile devices are out of radio coverage of UAVs. How to achieve low-cost task unloading in the resource limited and location sensitive multi-UAVs edge computing environment is rather challenging. In this paper, we propose a distributed location-aware task offloading scheme, aiming at the above issues. Specifically, we create a nonlinear task allocation problem by combining the limited energy constraints of edge nodes with the random movement of users, where the cost function is divided into static and dynamic costs, respectively. Then, we formulate this problem to a convex optimization one with linear constrains, based on regularization technology. The mathematical proof shows that the scheme can support a parameterized competitive ratio without requiring any prior knowledge of the input task. The simulation results show that the proposed scheme can achieve lower cost edge computing services.

Recent years have seen many advances based on Deep Convolutional Neural Networks (DCNNs) in the tasks of face recognition, most of which are developed to pursue high recognition accuracy. In this paper, we propose a novel Fast FAce Recognizer (Fast-FAR), learning to improve the speed of DCNN-based face recognition model without sacrificing recognition accuracy. Our fundamental insight is that the computation increases exponentially with the depth of a network, the easily identifiable face images can be accurately recognized by the cheap features (pixel values at shallow layers), while the challenging samples that exhibit low quality, large pose variations or occlusions need to be processed by the expensive deep layers. The major contribution of this paper is the Reinforcement Learning Agent (RLA), which is proposed to learn a decision policy determined by a reward function. The policy adaptively decides whether the recognition should be performed at an early layer with a high recognition confidence, or proceeding to the subsequent layers, thus significantly reducing feed-forward cost for the easy faces. According to the extensive experiments on the popular face recognition benchmarks, Fast-FAR reduces the inference time by 14.22%, 20.61%, and 7.84% on the dataset LFW, AgeDB-30 and CFP-FP, respectively.

With the rapid advancement of Internet of Vehicles (IoV) technology, the demands for real-time navigation, advanced driver-assistance systems (ADAS), vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications, and multimedia entertainment systems have made in-vehicle applications increasingly computing-intensive and delay-sensitive. These applications require significant computing resources, which can overwhelm the limited computing capabilities of vehicle terminals despite advancements in computing hardware due to the complexity of tasks, energy consumption, and cost constraints. To address this issue in IoV-based edge computing, particularly in scenarios where available computing resources in vehicles are scarce, a multi-master and multi-slave double-layer game model is proposed, which is based on task offloading and pricing strategies. The establishment of Nash equilibrium of the game is proven, and a distributed artificial bee colonies algorithm is employed to achieve game equilibrium. Our proposed solution addresses these bottlenecks by leveraging a game-theoretic approach for task offloading and resource allocation in mobile edge computing (MEC)-enabled IoV environments. Simulation results demonstrate that the proposed scheme outperforms existing solutions in terms of convergence speed and system utility. Specifically, the total revenue achieved by our scheme surpasses other algorithms by at least 8.98%. [ABSTRACT FROM AUTHOR]