时空视频超分辨率(space-time video super-resolution,STVSR)通过时间和空间2个尺度提升视频的质量,从而实现在视频采集设备、传输或者存储有限的情况下依然能实时地呈现高分辨率和高帧率的视频,满足人们对超高清画质的追求。相比两阶...时空视频超分辨率(space-time video super-resolution,STVSR)通过时间和空间2个尺度提升视频的质量,从而实现在视频采集设备、传输或者存储有限的情况下依然能实时地呈现高分辨率和高帧率的视频,满足人们对超高清画质的追求。相比两阶段方法,一阶段方法实现的是特征层面而非像素层面的帧插值,其在推理速度和计算复杂度上都明显更胜一筹。一些现有的一阶段STVSR方法采用基于像素幻觉的特征插值,这幻化了像素,因此很难应对帧间快速运动物体的预测。为此,提出一种基于光流法的金字塔编码器-解码器网络来进行时间特征插值,实现快速的双向光流估计和更真实自然的纹理合成,在使得网络结构更高效的同时弥补了大运动对光流估计带来的不稳定性。另外,空间模块采用基于滑动窗口的局部传播和基于循环网络的双向传播来强化帧对齐,整个网络称为时间特征细化网络(temporal feature refinement netowrk,TFRnet)。为了进一步挖掘TFRnet的潜力,将空间超分辨率先于时间超分辨率(space-first),在几种广泛使用的数据基准和评估指标上的实验证明了所提出方法TFRnet-sf的出色性能,在总体峰值信噪比(peak signal to noise ratio,PSNR)和结构相似性(structural similarity,SSIM)提升的同时,插入中间帧的PSNR和SSIM也得到提升,在一定程度上缓和了插入的中间帧与原有帧之间PSNR和SSIM差距过大的问题。展开更多
目前矩阵分解推荐系统在集中环境下存在隐私泄露的风险,且更多的数据拥有者不愿提供自身的数据,应用于分布式环境下的联邦矩阵分解推荐系统应用而生。传统的联邦矩阵分解模型在数据稀疏的情况下推荐准确率低,没有考虑用户的兴趣随时间...目前矩阵分解推荐系统在集中环境下存在隐私泄露的风险,且更多的数据拥有者不愿提供自身的数据,应用于分布式环境下的联邦矩阵分解推荐系统应用而生。传统的联邦矩阵分解模型在数据稀疏的情况下推荐准确率低,没有考虑用户的兴趣随时间变化的动态性。本文针对以上问题,引入联邦矩阵分解模型与时间隐语义模型相结合,提出一种融合时间特征的联邦矩阵分解推荐算法TF-FedMF (Federated Matrix Factorization Recommendation Algorithm with Temporal Feature Integration)。该算法在联邦矩阵分解框架中加入时间特征,用于捕捉用户行为随时间变化的趋势,提高了推荐系统的时效性和准确性;同时,结合同态加密对上传的梯度信息进行加密,增强算法的安全性。通过MovieLens数据集进行实验对比,实验结果表明,所提出的算法较其它算法在兼顾用户隐私安全性的同时,具有较高的推荐准确性。At present, the matrix decomposition recommendation system has the risk of privacy leakage in a centralized environment, and more data owners are unwilling to provide their own data. Therefore, the application of the federated matrix decomposition recommendation system in a distributed environment has emerged. The traditional federated matrix decomposition model has low recommendation accuracy when the data is sparse, and does not consider the dynamic nature of user interests changing over time. In view of the above problems, this paper introduces the combination of the federated matrix decomposition model and the temporal latent semantic model, and proposes a federated matrix decomposition recommendation algorithm TF-FedMF (Federated Matrix Factorization Recommendation Algorithm with Temporal Feature Integration) with temporal feature integration. The algorithm adds temporal features to the framework of federated matrix decomposition to capture the trend of user behavior changing over time, thereby improving the timeliness and accuracy of the recommendation system;at the same time, the uploaded gradient information is encrypted by combining homomorphic encryption to enhance the security of the algorithm. An experimental comparison is carried out on the MovieLens dataset. The experimental results show that the proposed algorithm has higher recommendation accuracy than other algorithms while taking into account user privacy and security.展开更多
文摘时空视频超分辨率(space-time video super-resolution,STVSR)通过时间和空间2个尺度提升视频的质量,从而实现在视频采集设备、传输或者存储有限的情况下依然能实时地呈现高分辨率和高帧率的视频,满足人们对超高清画质的追求。相比两阶段方法,一阶段方法实现的是特征层面而非像素层面的帧插值,其在推理速度和计算复杂度上都明显更胜一筹。一些现有的一阶段STVSR方法采用基于像素幻觉的特征插值,这幻化了像素,因此很难应对帧间快速运动物体的预测。为此,提出一种基于光流法的金字塔编码器-解码器网络来进行时间特征插值,实现快速的双向光流估计和更真实自然的纹理合成,在使得网络结构更高效的同时弥补了大运动对光流估计带来的不稳定性。另外,空间模块采用基于滑动窗口的局部传播和基于循环网络的双向传播来强化帧对齐,整个网络称为时间特征细化网络(temporal feature refinement netowrk,TFRnet)。为了进一步挖掘TFRnet的潜力,将空间超分辨率先于时间超分辨率(space-first),在几种广泛使用的数据基准和评估指标上的实验证明了所提出方法TFRnet-sf的出色性能,在总体峰值信噪比(peak signal to noise ratio,PSNR)和结构相似性(structural similarity,SSIM)提升的同时,插入中间帧的PSNR和SSIM也得到提升,在一定程度上缓和了插入的中间帧与原有帧之间PSNR和SSIM差距过大的问题。
文摘目前矩阵分解推荐系统在集中环境下存在隐私泄露的风险,且更多的数据拥有者不愿提供自身的数据,应用于分布式环境下的联邦矩阵分解推荐系统应用而生。传统的联邦矩阵分解模型在数据稀疏的情况下推荐准确率低,没有考虑用户的兴趣随时间变化的动态性。本文针对以上问题,引入联邦矩阵分解模型与时间隐语义模型相结合,提出一种融合时间特征的联邦矩阵分解推荐算法TF-FedMF (Federated Matrix Factorization Recommendation Algorithm with Temporal Feature Integration)。该算法在联邦矩阵分解框架中加入时间特征,用于捕捉用户行为随时间变化的趋势,提高了推荐系统的时效性和准确性;同时,结合同态加密对上传的梯度信息进行加密,增强算法的安全性。通过MovieLens数据集进行实验对比,实验结果表明,所提出的算法较其它算法在兼顾用户隐私安全性的同时,具有较高的推荐准确性。At present, the matrix decomposition recommendation system has the risk of privacy leakage in a centralized environment, and more data owners are unwilling to provide their own data. Therefore, the application of the federated matrix decomposition recommendation system in a distributed environment has emerged. The traditional federated matrix decomposition model has low recommendation accuracy when the data is sparse, and does not consider the dynamic nature of user interests changing over time. In view of the above problems, this paper introduces the combination of the federated matrix decomposition model and the temporal latent semantic model, and proposes a federated matrix decomposition recommendation algorithm TF-FedMF (Federated Matrix Factorization Recommendation Algorithm with Temporal Feature Integration) with temporal feature integration. The algorithm adds temporal features to the framework of federated matrix decomposition to capture the trend of user behavior changing over time, thereby improving the timeliness and accuracy of the recommendation system;at the same time, the uploaded gradient information is encrypted by combining homomorphic encryption to enhance the security of the algorithm. An experimental comparison is carried out on the MovieLens dataset. The experimental results show that the proposed algorithm has higher recommendation accuracy than other algorithms while taking into account user privacy and security.
