Variable partial-update NLMS algorithms with data-selective updating

We note that some existing algorithms are based on the normalized least-mean square （NLMS） algorithm and aim to reduce the computational complexity of NLMS all inherited from the solution of the same optimization problem, but with different constraints. A new constraint is analyzed to substitute an extra searching technique in the set-membership partial-update NLMS algorithm （SM-PU-NLMS） which aims to get a variable number of updating coefficients for a further reduction of computational complexity. We get a closed form expression of the new constraint without extra searching technique to generate a novel set-membership variable-partial-update NLMS （SM-VPU-NLMS） algorithm. Note that tile SM-VPU-NLMS algorithm obtains a faster convergence and a smaller mean-squared error （MSE） than the existing SM-PU-NLMS. It is pointed out that the closed form expression can also be applied to the conventional variable-step-size partial-update NLMS （VSS-PU-NLMS） algorithm. The novel variable-step-size variable-partial-update NLMS （VSS-VPU-NLMS） algorithm is also verified to get a further computational complexity reduction. Simulation results verify that our analysis is reasonable and effective.     

融合交互意图的图神经网络协同过滤算法

传统的协同过滤方法很大程度上以一种统一的方式进行建模,未从交互意图的细粒度上考虑建模关系。为此提出一种融合交互意图的图神经网络协同过滤算法（INTNGCF）。首先,通过将边分解成多个潜在空间来识别潜在意图;其次,利用交互意图融合层确定这些潜在意图的重要性;最后,生成用户对项目的预测评分。在三个真实数据集上与七种基线模型进行对比实验,结果表明,提出的算法与其他先进的推荐模型相比具有一定的性能优势。     

基于个性化推荐系统的视频App的设计

近年来,各类视频应用上内容越来越丰富,页面上与当前用户无关的内容也越来越多。因此,市面上出现了多种不同的推荐算法来进行内容推荐。但是,不是每种推荐算法都能够解决所有的问题。基于个性化推荐系统的视频App,融合了多种推荐方法。首先为了解决推荐系统的冷启动问题,采用了基于统计学的推荐方式,同时,采用基于协同过滤的推荐算法,计算视频和用户间的隐藏特征,最后还有实时推荐模块,能够根据用户近期的行为对推荐内容进行调整。     

融合引导滤波和小波变换的水下图像增强算法

针对水下环境对比度低和极其模糊的视觉效果问题,提出了一种融合引导滤波和小波变换的水下图像增强算法。利用引导滤波将图像分为基础部分和细节部分,将小波变换应用于基础部分,对产生的低频图像进行限制对比度直方图均衡,高频部分分别使用改进的对数函数增强,然后将处理后的高低频部分进行小波逆变换。细节部分进行增益计算;最后将处理后的基础部分和细节部分进行加权融合,再对融合图像进行限制对比度直方图均衡,得到目标图像。仿真实验表明,该方法在保留图像细节的同时很好地克服了光照不均匀对图像的影响.     

基于迁移学习的卷积神经网络通道剪枝

卷积神经网络在计算机视觉等多个领域应用广泛,然而其模型参数量众多、计算开销庞大,导致许多边缘设备无法满足其存储与计算资源要求。针对其边缘部署困难,提出使用迁移学习策略改进基于BN层缩放因子通道剪枝方法的稀疏化过程。本文对比不同层级迁移方案对稀疏化效果与通道剪枝选取容限的影响;并基于网络结构搜索观点设计实验,探究其精度保持极限与迭代结构的收敛性。实验结果表明,对比原模型,采用迁移学习的通道剪枝算法,在精度损失不超过0.10的前提下,参数量减少89.1%,模型存储大小压缩89.3%;对比原剪枝方法,将剪枝阈值从0.85提升到0.97,进一步减少参数42.6%。实验证明,引入迁移策略更易实现充分的稀疏化,提高通道剪枝阈值选取容限,实现更高压缩率;并在迭代剪枝的网络结构搜索过程中,提供更高效的搜索起点,利于快速迭代趋近至搜索空间的一个网络结构局部最优解。     

A joint introduction to Gaussian Processes and Relevance Vector Machines with connections to Kalman filtering and other kernel smoothers

The expressive power of Bayesian kernel-based methods has led them to become an important tool across many different facets of artificial intelligence, and useful to a plethora of modern application domains, providing both power and interpretability via uncertainty analysis. This article introduces and discusses two methods which straddle the areas of probabilistic Bayesian schemes and kernel methods for regression: Gaussian Processes and Relevance Vector Machines. Our focus is on developing a common framework with which to view these methods, via intermediate methods a probabilistic version of the well-known kernel ridge regression, and drawing connections among them, via dual formulations, and discussion of their application in the context of major tasks: regression, smoothing, interpolation, and filtering. Overall, we provide understanding of the mathematical concepts behind these models, and we summarize and discuss in depth different interpretations and highlight the relationship to other methods, such as linear kernel smoothers, Kalman filtering and Fourier approximations. Throughout, we provide numerous figures to promote understanding, and we make numerous recommendations to practitioners. Benefits and drawbacks of the different techniques are highlighted. To our knowledge, this is the most in-depth study of its kind to date focused on these two methods, and will be relevant to theoretical understanding and practitioners throughout the domains of data-science, signal processing, machine learning, and artificial intelligence in general.     

基于局部概率抽样的标签噪声过滤方法

分类学习任务中,在获取数据的过程中会不可避免地产生噪声,特别是标签噪声的存在不仅使得学习模型更复杂,而且容易造成过拟合并导致分类器泛化能力的下降。标签噪声过滤算法虽然在一定程度上可以解决上述问题,但是仍然存在噪声识别能力较差、分类效果不够理想以及过滤效率低等问题。针对这些问题,提出一种基于标签置信度分布的局部概率抽样方法来进行标签噪声过滤。首先利用随机森林分类器对样本的标签进行投票,从而获取每个样本的标签置信度;然后根据标签置信度的大小,将样本划分为易识别样本和难识别样本;最后分别采用不同的过滤策略对样本进行过滤。实验结果表明,在标签噪声存在的情况下,所提方法在大多数案例上能够保持较高的噪声识别能力,并且在分类泛化性能上也具有明显优势。     

融入时间的兴趣点协同推荐算法

兴趣点（POI）推荐可以帮助用户发现其没有访问过但可能感兴趣的地点,是重要的基于位置的服务之一。时间在POI推荐中是一个重要因素,而现有POI推荐模型并没有较好地考虑时间因素,因此通过考虑时间因素来提出融入时间的POI协同推荐（TUCF）算法,从而提高POI推荐的效果。首先,分析基于位置的社交网络（LBSN）的用户签到数据,以探索用户签到的时间关系;然后,利用时间关系对用户签到数据进行平滑处理,以融入时间因素并缓解数据稀疏性;最后,根据基于用户的协同过滤方法,在不同时间推荐不同POI给用户。在真实签到数据集上的实验结果表明,与基于用户的协同过滤（U）算法相比,TUCF算法的精确率和召回率分别提高了63%和69%;与具有平滑增强时间偏好的协同过滤（UTE）算法相比,TUCF算法的精确率和召回率分别提高了8%和12%;并且TUCF算法的平均绝对误差（MAE）比U算法和UTE算法分别减小了1.4%和0.5%。     

基于均衡接近度灰关联的Slope One算法

作为一种基于项目的协同过滤推荐算法,Slope One算法易于实现且高效。但由于Slope One算法未考虑用户相似性,导致其在处理涉及用户关系的个性化推荐任务时性能不高。针对以上问题,提出了改进的Slope One算法。提出一种均衡接近度灰关联分析方法计算用户之间的均衡接近度,利用均衡接近度度量用户间的相似程度,然后将均衡接近度值融入到Slope One算法中进行评分预测,在MovieLens和Epinions数据集下的对比实验表明,该算法具有更低的平均绝对误差（MAE）和均方根误差（RMSE）,提高了预测的准确度和推荐质量。     

Modified BM3D algorithm for image denoising using the directed diffusion equation

In order to preserve the edges and details of the image better, a modified block-matching and 3D filtering algorithm based on the directed diffusion is proposed. First, the block-matching and 3D filtering result is divided into three perpendicular subbands by using the wavelet decomposition. Second, by adding the diffusion operator and diffusion coefficients to the directed diffusion equation, the diffusion speed can be accelerated and edges can not be blurred quickly. Then we diffuse the subband of the high-frequency, of which the wavelet coefficients are less than the threshold to the corresponding subband of the noisy image to get a new estimate. Lastly, by replacing the corresponding subband of the block-matching and 3D filtering estimated image with the new one, we obtain the improved denoising result. Experimental results show that the improved algorithm achieves better performance than the original block-matching and 3D filtering algorithm in terms of both removing noise and preserving the edges and details of the image.