Search

Abstract With the development of deep neural network (DNN) techniques, applications of DNNs show state-of-art performance. In the cloud edge collaborative mode, edge devices upload the raw data, such as texts, images, and videos, to the cloud for processing. Then, the cloud returns prediction or classification results. Although edge devices take advantage of the powerful performance of DNN, there are also colossal privacy protection risks. DNN partition strategy can effectively solve the privacy problems by offload part of the DNN model to the edge, in which the encoded features are transmitted rather than original data. We explore the relationship between privacy and the intermedia result of the DNN. The more parts offloaded to the edge, the more abstract features we can have, indicating more conducive to privacy protection. We propose a privacy protection approach based on a maximum DNN partition strategy. Besides, a mix-precision quantization approach is adopted to reduce the energy use of edge devices. The experiments show that our method manages to increase at most 20% model privacy in various DNN architecture. Through the energy-aware mixed-precision quantization approach, the model’s energy consumption is reduced by at most 5x comparing to the typical edge-cloud solution.

Multi-modal 3D object detection has achieved remarkable progress, but it is often limited in practical industrial production because of its high cost and low efficiency. The multi-view camera-based method provides a feasible solution due to its low cost. However, camera data lacks geometric depth, and only using camera data to obtain high accuracy is challenging. This paper proposes a multi-modal Bird-Eye-View (BEV) distillation framework (MMDistill) to make a trade-off between them. MMDistill is a carefully crafted two-stage distillation framework based on teacher and student models for learning cross-modal knowledge and generating multi-modal features. It can improve the performance of unimodal detectors without introducing additional costs during inference. Specifically, our method can effectively solve the cross-gap caused by the heterogeneity between data. Furthermore, we further propose a Light Detection and Ranging (LiDAR)-guided geometric compensation module, which can assist the student model in obtaining effective geometric features and reduce the gap between different modalities. Our proposed method generally requires fewer computational resources and faster inference speed than traditional multi-modal models. This advancement enables multi-modal technology to be applied more widely in practical scenarios. Through experiments, we validate the effectiveness and superiority of MMDistill on the nuScenes dataset, achieving an improvement of 4.1% mean Average Precision (mAP) and 4.6% NuScenes Detection Score (NDS) over the baseline detector. In addition, we also present detailed ablation studies to validate our method. [ABSTRACT FROM AUTHOR]

The physiological similarities between mice and humans make them vital animal models in biological and medical research. This paper explores the application of artificial intelligence (AI) in analyzing mice behavior, emphasizing AI's potential to identify and classify these behaviors. Traditional methods struggle to capture subtle behavioral features, whereas AI can automatically extract quantitative features from large datasets. Consequently, this study aims to leverage AI to enhance the efficiency and accuracy of mice behavior analysis. The paper reviews various applications of mice behavior analysis, categorizes deep learning tasks based on an AI pyramid, and summarizes AI methods for addressing these tasks. The findings indicate that AI technologies are increasingly applied in mice behavior analysis, including disease detection, assessment of external stimuli effects, social behavior analysis, and neurobehavioral assessment. The selection of AI methods is crucial and must align with specific applications. Despite AI's promising potential in mice behavior analysis, challenges such as insufficient datasets and benchmarks remain. Furthermore, there is a need for a more integrated AI platform, along with standardized datasets and benchmarks, to support these analyses and further advance AI-driven mice behavior analysis. [ABSTRACT FROM AUTHOR]

The automatic recognition and quantitative analysis of abnormal behavior in mice play a crucial role in behavioral observation experiments in neuroscience, pharmacology, and toxicology. Due to the challenging definition of abnormal behavior and difficulty in collecting training samples, directly applying behavior recognition methods to identify abnormal behavior is often infeasible. This paper proposes ABNet, an AI-empowered abnormal action recognition approach for mice. ABNet utilizes an enhanced Spatio-Temporal Graph Convolutional Network (ST-GCN) as an encoder; ST-GCN combines graph convolution and temporal convolution to efficiently capture and analyze spatio-temporal dynamic features in graph-structured data, making it suitable for complex tasks such as action recognition and traffic prediction. ABNet trains the encoding network with normal behavior samples, then employs unsupervised clustering to identify abnormal behavior in mice. Compared to the original ST-GCN network, the method significantly enhances the capabilities of feature extraction and encoding. We conduct comprehensive experiments on the Kinetics-Skeleton dataset and the mouse behavior dataset to evaluate and validate the performance of ABNet in behavior recognition and abnormal motion detection. In the behavior recognition experiments conducted on the Kinetics-Skeleton dataset, ABNet achieves an accuracy of 32.7% for the top one and 55.2% for the top five. Moreover, in the abnormal behavior analysis experiments conducted on the mouse behavior dataset, ABNet achieves an average accuracy of 83.1%. [ABSTRACT FROM AUTHOR]

Multi-modal fusion technology gradually become a fundamental task in many fields, such as autonomous driving, smart healthcare, sentiment analysis, and human-computer interaction. It is rapidly becoming the dominant research due to its powerful perception and judgment capabilities. Under complex scenes, multi-modal fusion technology utilizes the complementary characteristics of multiple data streams to fuse different data types and achieve more accurate predictions. However, achieving outstanding performance is challenging because of equipment performance limitations, missing information, and data noise. This paper comprehensively reviews existing methods based on multi-modal fusion techniques and completes a detailed and in-depth analysis. According to the data fusion stage, multi-modal fusion has four primary methods: early fusion, deep fusion, late fusion, and hybrid fusion. The paper surveys the three major multi-modal fusion technologies that can significantly enhance the effect of data fusion and further explore the applications of multi-modal fusion technology in various fields. Finally, it discusses the challenges and explores potential research opportunities. Multi-modal tasks still need intensive study because of data heterogeneity and quality. Preserving complementary information and eliminating redundant information between modalities is critical in multi-modal technology. Invalid data fusion methods may introduce extra noise and lead to worse results. This paper provides a comprehensive and detailed summary in response to these challenges. [ABSTRACT FROM AUTHOR]

Blockchain as a Service (BaaS) combines features of cloud computing and blockchain, making blockchain applications more convenient and promising. Although current BaaS platforms have been widely adopted by both industry and academia, concerns arise regarding their performance, especially in job allocation. Existing BaaS job allocation strategies are simple and do not guarantee load balancing due to the dynamic nature and complexity of BaaS job execution. In this paper, we propose a deep reinforcement learning-based algorithm, Balanced-DRL, to learn an optimized allocation strategy in BaaS based on analyzing the execution process of BaaS jobs and a set of job scale characteristics. Following extensive experiments with generated job request workloads, the results show that Balanced-DRL significantly improves BaaS performance, achieving a 5% to 8% increase in job throughput and a 5% to 20% decrease in job latency. [ABSTRACT FROM AUTHOR]

The personalization of autonomous vehicles or advanced driver assistance systems has been a widely researched topic, with many proposals aiming to achieve human-like or driver-imitating methods. However, these approaches rely on an implicit assumption that all drivers prefer the vehicle to drive like themselves, which may not hold true for all drivers. To address this issue, this study proposes an online personalized preference learning method (OPPLM) that utilizes a pairwise comparison group preference query and the Bayesian approach. The proposed OPPLM adopts a two-layer hierarchical structure model based on utility theory to represent driver preferences on the trajectory. To improve the accuracy of learning, the uncertainty of driver query answers is modeled. In addition, informative query and greedy query selection methods are used to improve learning speed. To determine when the driver's preferred trajectory has been found, a convergence criterion is proposed. To evaluate the effectiveness of the OPPLM, a user study is conducted to learn the driver's preferred trajectory in the curve of the lane centering control (LCC) system. The results show that the OPPLM can converge quickly, requiring only about 11 queries on average. Moreover, it accurately learned the driver's favorite trajectory, and the estimated utility of the driver preference model is highly consistent with the subject evaluation score. [ABSTRACT FROM AUTHOR]