Singular-Value-Decomposition Analysis of Associative Memory in a Neural Network

Kumamoto, Tatsuya; Suzuki, Mao; Matsueda, Hiroaki

doi:10.7566/jpsj.86.024005

Cited by 7 publications

(4 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This method can deal better with light changes and subtle movements in the scene, such as water waves and branches swinging, but it requires a large number of images without moving objects to train the model. A key pose extraction system for weightlifting video was designed by literature [ 24 ]. A display module and a key frame extraction module are included in the system.…”

Section: Related Workmentioning

confidence: 99%

Sports Video Athlete Detection Based on Associative Memory Neural Network

Yang

2022

Computational Intelligence and Neuroscience

View full text Add to dashboard Cite

Aiming at the detection of athletes in sports videos, an automatic detection method based on AMNN is proposed. The background image from the image sequence is obtained, the moving area is extracted, and the color information of pixels to extract the green stadium from the background image is used. In order to improve the accuracy of athletes’ detection, the texture similarity measurement method is used to eliminate the shadow in the movement area, the morphological method is used to eliminate the cracks in the area, and the noise outside the stadium is removed according to the stadium information. Combined with the images of nonathletes, a training set is constructed to train the NN classifier. For the input image frames, image pyramids of different scales are constructed by subsampling and the positions of several candidate athletes are detected by NN. The center of gravity of candidate athletes is calculated, a representative candidate athlete is obtained, and then, the final athlete position through a local search process is determined. Experiments show that the system can accurately detect the motion shape of moving targets, can process images in real time, and has good real-time performance.

show abstract

Section: Related Workmentioning

confidence: 99%

Sports Video Athlete Detection Based on Associative Memory Neural Network

Yang

2022

Computational Intelligence and Neuroscience

View full text Add to dashboard Cite

show abstract

“…Information of the weight matrix W can be inferred by using the method of the singular value decomposition (See, e.g., [26,27]). Suppose that the matrix W W T has eigenvalues λ a Figure 13: Averaged values of the off-diagonal components of W W T (normalized by the diagonal components).…”

Section: Magnetization and Singular Value Decomposition (Svd)mentioning

confidence: 99%

Scale-invariant feature extraction of neural network and renormalization group flow

2018

View full text Add to dashboard Cite

Theoretical understanding of how a deep neural network (DNN) extracts features from input images is still unclear, but it is widely believed that the extraction is performed hierarchically through a process of coarse graining. It reminds us of the basic renormalization group (RG) concept in statistical physics. In order to explore possible relations between DNN and RG, we use the restricted Boltzmann machine (RBM) applied to an Ising model and construct a flow of model parameters (in particular, temperature) generated by the RBM. We show that the unsupervised RBM trained by spin configurations at various temperatures from T=0 to T=6 generates a flow along which the temperature approaches the critical value T_{c}=2.27. This behavior is the opposite of the typical RG flow of the Ising model. By analyzing various properties of the weight matrices of the trained RBM, we discuss why it flows towards T_{c} and how the RBM learns to extract features of spin configurations.

show abstract

“…Reference [14] put forward the idea of block histogram. Reference [15] uses the Canny operator to detect the edge of each frame of the image, then calculates the in-and-out degree of the edge pixels of the image to get the edge change rate, and uses the edge change rate to detect the shot boundary [16]. By detecting the edge features of candidate shot boundaries, the false detection of shot boundaries caused by flash can be avoided and the calculation time can not be greatly increased.…”

Section: Related Workmentioning

confidence: 99%

Detection of Shot Transition in Sports Video Based on Associative Memory Neural Network

2022

Wireless Communications and Mobile Computing

View full text Add to dashboard Cite

Users must quickly and effectively classify, browse, and retrieve videos due to the explosive growth of video data. A variety of shots make up the video data stream. The most important technology in video retrieval is shot detection, which can fundamentally solve many problems, resulting in improved detection effects and even directly affecting video retrieval performance. This paper investigates the shot transition detection algorithm in digital video live broadcasts based on sporting events. To solve the problem of shot transition detection using a single training sample, an AMNN (Associative Memory Neural Network) model with online learning ability is proposed. Experiments on a large football video data set show that this algorithm detects shear and gradual change better than existing algorithms and meets the application requirements of sports video retrieval in most cases.

show abstract

Singular-Value-Decomposition Analysis of Associative Memory in a Neural Network

Cited by 7 publications

References 35 publications

Sports Video Athlete Detection Based on Associative Memory Neural Network

Sports Video Athlete Detection Based on Associative Memory Neural Network

Scale-invariant feature extraction of neural network and renormalization group flow

Detection of Shot Transition in Sports Video Based on Associative Memory Neural Network

Contact Info

Product

Resources

About