### TABLE II MSE Values from Computer Simulations of DVQ on Video Images with Codebooks of Varying Size Codebook Size

1995

Cited by 7

### Table 1: Summary of RMS retrieval errors for both neural net and regression methods, for three channel sets. TIGR error is error retrieving odd numbered pro les. All retrievals are with AIRS instrument noise added.

1995

"... In PAGE 9: ...04K. Table1 and Figures 2, 3, and 4 give a summary of RMS error for both neural nets and regression, for several channel sets. Training (or regression) is performed on even-numbered TIGR pro les.... In PAGE 13: ... If ^ Tb is Tb with added noise, then let ^ Tb0 = BT 1 ^ Tb, let C be the least squares solution to C ^ T 0 b = T 0, and D = B2CBT 1 . Table1 and Figures 2, 3 and 4 summarize RMS testing error for the regression method, and compare regression results with neural nets. As with the neural nets, the eigenvector bases are determined from and the regression is performed on even-numbered TIGR pro les, while the error shown is for retrievals of the odd-numbered TIGR pro les.... In PAGE 21: ... In addition the adaptive learning rate variation of backprop that we used has several parameters: learning rate increment, decrement, and error threshold. Parameters for the adaptive learning algo- rithm, as used to train the 728-input net (run 410) described in Figure 2 and Table1 are as follows. Parameter Run 410 Useful Range momentum 0.... ..."

Cited by 1

### Table 4: Coefficients of three channel models

1995

Cited by 3

### Table 1 shows the results of an EEG recording with 3 channels. The dynamic range and the quantization was the same for all three channels. But the amplifier noise of channel #3 was more than twice as large as for channel #1 and #2. Accordingly, the SNR is also smaller.

"... In PAGE 3: ...l. 1999). In case of sleep EEG, the typical entropy difference between EEG and quantization is 8 to 11bits per sample (Penzel, 2001). Table1 : Overview of characteristic values ... ..."

### Table III shows the average computational complexities per 8 8 block for decoding the test images, Lenna and boat, by the additive vector decoder in the experiments of Section V- A. In the calculations of the complexities, we did not count the blocks that do not contain any nonzero quantized AC coefficients, since these DC blocks can be trivially decoded. Comparing with the inverse DCT decoder, the vector decoder could decode an image with fewer multiplications but slightly more additions. The numbers of additions in Table III are based on code vector dimension 14

1998

Cited by 2

### Table 3.1. General properties for three channel types.

### Table 3.1. General properties for three channel types.

1999

### Table 3.1. General properties for three channel types.

### Table 4.2: ANOVA statistics for the three channel types

### Table 3.1. General properties for three channel types.

1999