The rapid expansion and diversification of internet traffic pose a pressing demand for accurate classification of diverse application streams in network management and security monitoring. This paper introduces a deep...
详细信息
Atmospheric particulate matter pollution has attracted much wider attention *** recent years,the development of atmospheric particle collection techniques has put forwards new demands on the real-time source apportion...
详细信息
Atmospheric particulate matter pollution has attracted much wider attention *** recent years,the development of atmospheric particle collection techniques has put forwards new demands on the real-time source apportionments *** demands are summarized,in this paper,as how to set up new restraints in apportionment and how to develop a non-linear regression model to process complicated circumstances,such as the existence of secondary source and similar *** this study,we firstly analyze the possible and potential restraints in single particle source apportionment,then propose a novel three-step self-feedback long short-term memory(SF-LSTM)network for approximating the source *** proposed deep learning neural network includes three modules,as generation,scoring and refining,and regeneration *** from the scoring modules,SF-LSTM implants four loss functions representing four restraints to be followed in the apportionment,meanwhile,the regeneration module calculates the source contribution in a non-linear *** results show that the model outperforms the conventional regression methods in the overall performance of the four evaluation indicators(residual sum of squares,stability,sparsity,negativity)for the ***,in short time-resolution analyzing,SF-LSTM provides better results under the restraint of stability.
Protein language models are currently experiencing a surge in demand owing to their remarkable accuracy in protein structure prediction. Nevertheless, their applications are hindered by the significant computation and...
详细信息
ISBN:
(数字)9798350359312
ISBN:
(纸本)9798350359329
Protein language models are currently experiencing a surge in demand owing to their remarkable accuracy in protein structure prediction. Nevertheless, their applications are hindered by the significant computation and memory requirements. The existing optimization strategies primarily focus on computational efficiency while often neglecting memory optimization, thereby restricting their suitability for devices with limited resources. In this paper, we propose MEFold, a novel memory-efficient optimization framework for protein language models that enables efficient inference on resource-constrained devices. MEFold consists of Look-up Table Chunk and Fine-grained Quantization. Look-up Table Chunk reduces the memory of intermediate activations by chunk and avoids the overhead of obtaining the optimal chunk size configuration through pre-computing. For the memory of model parameters, Fine-grained Quantization, delicately controls the scope of quantization to ensure that memory reduction is achieved while preventing declines in accuracy and computational speed. Experimental results show that, compared to the original model, for protein sequences ranging from 74 to 1024 in length, our method significantly reduces the peak memory during inference from 14.7-54.2GB to 6.0-14.4GB, while minimizing the impact on inference latency. On CASP14 and CAMEO datasets, the accuracy loss compared to the original model is below 1%. Moreover, our optimization provides various memory-saving alternatives. Our code is available at https://***/llwx593/MEFold.
The protection of model intellectual property is becoming an increasingly important issue. However, the existing methods for protecting model ownership, although effective, have limitations. Firstly, they primarily fo...
The protection of model intellectual property is becoming an increasingly important issue. However, the existing methods for protecting model ownership, although effective, have limitations. Firstly, they primarily focus on classification models, and secondly, most of the proposed methods reduce the model's utility. To overcome these shortcomings, this paper proposes a task-agnostic model ownership verification framework based on feature fingerprint, called TMOVF, which separates ownership verification from model task. Our key idea is that model knowledge can be uniquely characterized by the extracted features, which may be high-dimensional, complicated, and difficult to compare for each input sample. Nevertheless, these features contain inherent information that cannot be ignored in cases of piracy. To measure the inheritance of our fingerprint, we introduce outlier detection into model ownership verification, which is a first in the field. By reconstructing the outlier detection algorithm, we extract the feature fingerprints of the victim model and the suspicious model, and compute the outliers of their feature fingerprints. By comparing the results, we can verify the ownership of the models. We conduct extensive experiments to evaluate our framework and demonstrate the inheritability of feature fingerprints in stolen models. Our experiments show that the framework is effective in verifying ownership, regardless of the model task. Additionally, our results demonstrate that our framework is more effective than existing methods.
Federated Learning (FL) is a privacy-preserving machine learning technique that trains models on client devices and only uploads new model gradients to servers for aggregation. However, transmitting the true gradients...
详细信息
For pursuing accurate skeleton-based action recognition, many existing graph-based approaches deploy the higher-order polynomials of the skeletal adjacency matrix to model the node correlations of distant neighbours. ...
详细信息
ISBN:
(数字)9798350368741
ISBN:
(纸本)9798350368758
For pursuing accurate skeleton-based action recognition, many existing graph-based approaches deploy the higher-order polynomials of the skeletal adjacency matrix to model the node correlations of distant neighbours. To further capture robust graphical patterns, a novel multi-scale graph convolution operator is proposed, which enables to aggregate multi-scale dependencies and capture long-range joint relationships on human skeleton graph. Additionally, a novel corrective contrastive learning strategy is proposed, which aims to distinguish the representative clues and calibrate the confused action clips in the feature space. Comprehensive experiments validate the effectiveness and superiority of our proposed method over state-of-the-art approaches on three large-scale datasets: NTU RGB+D 60, NTU RGB+D 120, and Kinetics Skeleton 400.
Feature matching is widely applied in the image processing field. However, both traditional feature matching methods and previous deep learning-based methods struggle to accurately match the features with severe defor...
详细信息
The utilization of artificial intelligence has had a profound impact on diverse areas within the medical field day by day. Specifically, the identification and management of gliomas, a specific category of brain tumor...
ISBN:
(纸本)9798400708138
The utilization of artificial intelligence has had a profound impact on diverse areas within the medical field day by day. Specifically, the identification and management of gliomas, a specific category of brain tumors characterized by a challenging prognosis, heavily depend on the computer-assisted analysis of magnetic resonance imaging (MRI) scans. In the present study, we introduce a novel approach based on neural architecture search (NAS) for accurately segmenting brain tumors utilizing multimodal volumetric MRI scans. Our method employs three classes of candidate operations for different cells, with each operation having a learnable probabilistic parameter. By iteratively updating the operation weights and other network parameters, we discover optimal structure for the encoder and decoder cells. Additionally, we introduce an attention module connection to the automatic search, complementing the connection between the encoder cells and decoder cells for brain MRI processing. Through extensive experiments conducted on the BraTS 2019 dataset, we validate the effectiveness and scalability of our proposed algorithm. Our approach not only relieves the burden of manual architecture design but also achieves competitive performance in terms of brain tumor segmentation.
The development of positioning technology and mobile intelligent devices has given rise to plentiful location-based services and generated much trajectory data. The publication and analysis of trajectory data bring co...
详细信息
ISBN:
(纸本)9798400707964
The development of positioning technology and mobile intelligent devices has given rise to plentiful location-based services and generated much trajectory data. The publication and analysis of trajectory data bring convenience to users, but also carry the risk of sensitive information being exposed. Existing trajectory data publication models mostly only consider the processing of static trajectory data and do not effectively combine spatiotemporal properties, and the problem of dishonest third parties unconsciously collecting data from other parties also occurs from time to time. This paper proposes a new privacy-preserving trajectory data publication hierarchical model based on secure two-party computation. The user layer, the cloud layer, and the local server layer collaborate to complete the trajectory publishing process. Experimental statistics have proved that the proposed method has better privacy and data availability.
Cardiac image segmentation is critical for medical diagnosis and treatment planning. Traditional approaches often face accuracy challenges. In this study, we propose a deep learning-based method that incorporates arch...
ISBN:
(纸本)9798400708138
Cardiac image segmentation is critical for medical diagnosis and treatment planning. Traditional approaches often face accuracy challenges. In this study, we propose a deep learning-based method that incorporates architectural improvements and optimization techniques to overcome these limitations. Our method integrates skip connections, a spatial attention mechanism, and label smoothing for enhanced segmentation performance. Experimental results on the CAMUS dataset show that our approach surpasses baseline models, achieving superior segmentation accuracy. Specifically, our method increases the mean Intersection over Union (mIoU) from 0.8141 (U-Net) to 0.8428 (Residual Attention U-Net) and the mean Dice score from 0.8948 (U-Net) to 0.9127 (Residual Attention U-Net). The proposed method has potential applications in medical diagnosis, disease prevention, and treatment planning, emphasizing its practical significance in cardiac image segmentation.
暂无评论