China
Africa
Explore chapters and articles related to this topic
Real-Time Detection of Facial Expressions Using K-NN, SVM, Ensemble Classifier and Convolution Neural Networks
Published in Saravanan Krishnan, Ramesh Kesavan, B. Surendiran, G. S. Mahalakshmi, Handbook of Artificial Intelligence in Biomedical Engineering, 2021
A. Sharmila, B. Bhavya, K. V. N. Kavitha, P. Mahalakshmi
Pantic and Rothkrantz (2004) developed an algorithm that recognized facial expressions in frontal and profile views. They took a sample of 25 subjects in an MMI database. They used a rule-based classifier that had an accuracy of 86%. They proposed a way to do automatic coding in profile images but not in real time. This involved the extraction of frontal and profile facial points. Buciu and Pitas (2004) performed principal component analysis (PCA) for a comparison purpose. Local non-negative matrix factorization (LNMF) outperformed both PCA and non-negative matrix factorization (NMF), whereas NMF performed the poorest. They discovered that the cumulative learning system (CSM) classifier is more reliable than Matthews correlation coefficient (MCC) and gives better recognition. They used the Cohn–Kanade database that had a sample size of 164 samples and the JAFFE database that had 150 samples.
Literature Survey
Published in Arun Reddy Nelakurthi, Jingrui He, Social Media Analytics for User Behavior Modeling, 2020
Arun Reddy Nelakurthi, Jingrui He
Non-negative matrix factorization (NMF) is widely used for co-clustering problems. Li and Ding (2006) demonstrated a NMF framework for document-word co-clustering. Cai et al. (2011) improved the framework proposed by Li and Ding (2006) by adding a graph regularizer which captures geometric information embedded in the data. Gu et al. (2011) proposed an orthogonal framework to fix scaling problem in Cai et al. (2011). Wang et al. (2015) proposed an NMF based Dual Knowledge Transfer approach for cross-language Web page classification. Our approach differs from previous works as we jointly factor user-keyword matrices from multiple social networks to learn latent features on the combined set of keywords from all the social networks and users from each social network. Chakraborty and Sycara (2015) proposed a constrained NMF framework for community detection in social networks which is closely related to our work. Our problem is different from the community detection problem, which finds communities of closely related users inside a social network.
Big Data in Prostate Cancer
Published in Ayman El-Baz, Jasjit S. Suri, Big Data in Multimodal Medical Imaging, 2019
Islam Reda, Ashraf Khalil, Mohammed Ghazal, Ahmed Shalaby, Mohammed Elmogy, Ahmed Aboelfetouh, Ali Mahmoud, Mohamed Abou El-Ghar, Ayman El-Baz
The proposed CAD system summarized in Figure 11.1 performs sequentially three steps. First, the prostate is segmented using our previously developed geometric deformable model (level-set) as described in [39]. This model is guided by a stochastic speed function that is derived using non-negative matrix factorization (NMF). The NMF attributes are calculated using information from the MRI intensity, a probabilistic shape model, and the spatial interactions between prostate voxels. The proposed approach reaches 86.89% overall Dice similarity coefficient and an average Hausdorff distance of 5.72 mm, indicating high segmentation accuracy. Details of this approach and comparisons with other segmentation approaches can be found in [39]. Afterwards, global features describing the water diffusion inside the prostate tissue are extracted based on the cumulative distribution functions of the ADC maps. Finally, a two-stage structure of stacked non-negativity constraint auto-encoder (SNCAE) is trained to classify the prostate tumor as benign or malignant based on the CDFs constructed in the previous step and the blood test-based probabilities. The latter two steps of the proposed CAD system are discussed in the following sections.
VoteSumm: A Multi-Document Summarization Scheme Using Influential Nodes of Multilayer Weighted Sentence Network
Published in IETE Technical Review, 2023
Raksha Agarwal, Niladri Chatterjee
In the last one decade or so, multi-document text summarization has gained significant attention. For illustration, Park et al. [1] used clustering and non-negative matrix factorization for muti-document text summarization. Agarwal and Chatterjee [2] used Word Graph-based multi-sentence compression for multi-document text abstraction. However, application of network-based representation for this task has not been pursued in a significant way in the past. The present work proposes a multi-document summarization scheme, named VoteSumm, for the abovesaid purpose. The proposed scheme works by identifying the influential nodes of a weighted multilayer sentence network using the VoteRank [3,4] technique. Sun et al. [4] used VoteRank for identifying potential spreader nodes in email networks, social media networks and epidemic networks.
Matrix Factorization in Recommender Systems: Algorithms, Applications, and Peculiar Challenges
Published in IETE Journal of Research, 2021
Non-negative matrix factorization [48–50] is notable among other MF. NMF grew out of Principal Component Analysis (PCA) and it has become an extensively used technique for exploring high-dimensional data, as it automatically extracts sparse and significant features from a set of non-negative data vectors. Given a non-negative matrix , where then the NMF factorizes the matrix into two distinct nonnegative matrices and as seen in Figure 3, which gives the following: it is possible to reconstruct the initial matrix using and as: the Frobenius norm can be utilized to compute the dissimilarity between and as follows [50]:
Multi-frame moving video detection algorithm for IOT based on Gauss Monte Carlo particle filter
Published in International Journal of Computers and Applications, 2020
Song Tao, Zhuang Lei, Jing Chenkai
Collect the foreground features of each short-term window and assume that is T set feature matrix. The target is to learn latent variable factors in . A simple way to perform local decomposition is non-negative matrix factorization (NNMF): where and are non-negative matrices. The column vector of contains latent variable factors and contains the contributory coefficient of each factor to the original data in . Local or additive decomposition can be obtained on the basis of non-negativity of . Each column of can be expressed as: However, the NNMF method requires the K number of latent variables to be given in advance, and the K number is difficult to be obtained, which affects the universality of the algorithm.