As one kind of distributed machine learning technique, federated learning enables multiple clients to build a model across decentralized data collaboratively without explicitly aggregating the data. Due to its ability...
详细信息
Neural Radiance Fields (NeRF) have been gaining attention as a significant form of 3D content representation. With the proliferation of NeRF-based creations, the need for copyright protection has emerged as a critical...
详细信息
In this paper, we explore the low-complexity optimal bilinear equalizer (OBE) combining scheme design for cell-free massive multiple-input multiple-output networks with spatially correlated Rician fading *** provide a...
详细信息
The daily practice of sharing images on social media raises a severe issue about privacy leakage. To address the issue, privacy-leaking image detection is studied recently, with the goal to automatically identify imag...
详细信息
Due to multi-layer encoding and Inter-layer prediction, Spatial Scalable High-Efficiency Video Coding (SSHVC) has extremely high coding complexity. It is very crucial to improve its coding speed so as to promote wides...
详细信息
Due to multi-layer encoding and Inter-layer prediction, Spatial Scalable High-Efficiency Video Coding (SSHVC) has extremely high coding complexity. It is very crucial to improve its coding speed so as to promote widespread and cost-effective SSHVC applications. In this paper, we have proposed a novel Mode Selection-Based Fast Intra Prediction algorithm for SSHVC. We reveal the RD costs of Inter-layer Reference (ILR) mode and Intra mode have a significant difference, and the RD costs of these two modes follow Gaussian distribution. Based on this observation, we propose to apply the classic Gaussian Mixture Model and Expectation Maximization in machine learning to determine whether ILR is the best mode so as to skip the Intra mode. Experimental results demonstrate that the proposed algorithm can significantly improve the coding speed with negligible coding efficiency loss.
On-ramp merge is a complex traffic scenario in autonomous driving. Because of the uncertainty of the driving environment, most rule-based models cannot solve such a problem. In this study, we design a Deep Reinforceme...
详细信息
Owing to a large amount of multi-modal data in modern medical systems, such as medical images and reports, Medical Vision-Language Pre-training (Med-VLP) has demonstrated incredible achievements in coarse-grained down...
详细信息
Diffusion MRI (dMRI) is widely used for the detection and diagnosis of various pathologies due to its non-invasive nature. However, its signal acquisition process is relatively slow and susceptible to motions. In orde...
详细信息
ISBN:
(数字)9798350387384
ISBN:
(纸本)9798350387391
Diffusion MRI (dMRI) is widely used for the detection and diagnosis of various pathologies due to its non-invasive nature. However, its signal acquisition process is relatively slow and susceptible to motions. In order to accelerate reconstruction process, many deep learning-based methods have been developed in recent years to achieve promising results, while most of them are supervised based methods and require fully sampled data which are difficult to obtain in practical applications to train the network. Self-supervised based methods were proposed to deal with this problem to a certain extent, however, they are devoted only to conventional MRI data, ignoring the correlation between the directions of dMRI data. To address this issue, we propose a self-supervised fast reconstruction method for dMRI based on multi-directional assistance. The method better utilizes the information between dMRI data directions by Partition-Replace operation, cleverly constructing data pairs and designing effective self-supervised strategy for bi-directional prediction, which can overcome the problem of data incompleteness and guide the model to reconstruct unsampled signals in spite of the difficulty of acquiring fully sampled data. Experimental results show that our method achieves better reconstruction in the absence of fully sampled data, even with an 8-fold acceleration factor.
Most of existing neural methods for multi-objective combinatorial optimization (MOCO) problems solely rely on decomposition, which often leads to repetitive solutions for the respective subproblems, thus a limited Par...
详细信息
Frequent data breaches have attracted people’s attention. The dark web forums have become the crucial platform for data breach transactions. While some forums are well-organized with distinct sections, such as a dedi...
详细信息
ISBN:
(数字)9798350391367
ISBN:
(纸本)9798350391374
Frequent data breaches have attracted people’s attention. The dark web forums have become the crucial platform for data breach transactions. While some forums are well-organized with distinct sections, such as a dedicated area for leaks where most threads are related to data breaches, others lack clear segmentation, like Fear Forum. Hackers conduct a variety of discussions there so it is filled with tons of non-data breach information and building a data breach dataset for a classifier requires huge human resources. The content of these threads is diverse and complex, spanning various fields, and containing specific terminology in the dark web. Identifying data breach threads in these forums is extremely challenging for the model. With the influx of new threads every day the static model trained on old data may not perform well when processing new data, which also brings serious difficulty to identification. To address these challenges, we first design Prompt-based Active Learning (PAL) to guide the model to identify through prompts and select high-quality samples for annotation. Then we propose a system that combines PAL and DarkBERT. Our system can not only greatly reduce annotation costs but also continuously learn and update itself with the new informative data selected by PAL. Through rich experiments on a data breach dataset, we demonstrate that our system effectively identifies breach threads and PAL significantly outperforms other baselines. We also search for the optimal prompt templates to improve system performance.
暂无评论