### Table 6. Use of the multiplier as a percentage of the total time of group doubling. Digit-size Number of multipliers

"... In PAGE 7: ...some cases also the third multiplier, are used very unfrequently (see Table6 , the column corresponding to the 4th multiplier). Hence, for most applications it will be unreasonable to provide this extra hardware unit.... ..."

### Table 1: Compression rates (bits per digit) for the single digit (Digit) and double digit (Pairs) datasets. Boldface marks the best performance on each dataset. m Digits Pairs

1998

"... In PAGE 6: ... The other algorithms mentioned in the table are the mixture of factorial distributions (MF), the completely factorized model (which assumes that every variable is independent of all the others) called \Base rate quot; (BR), the Helmholtz Machine trained by the wake-sleep algo- rithm [4] (HWS), the same Helmholtz Machine where a mean eld approximation was used for training (HMF) and a fully visible and fully connected sigmoid belief net- work (FV). Table1 displays the performances of all the mixture of trees models that we tested. The results are very good: the mixture of trees is the absolute winner for compressing the simple digits and comes in second as a model for pairs of digits.... ..."

Cited by 30

### Table 1: Compression rates (bits per digit) for the single digit (Digit) and double digit (Pairs) datasets. Boldface marks the best performance on each dataset. m Digits Pairs

1998

"... In PAGE 6: ... The other algorithms mentioned in the table are the mixture of factorial distributions (MF), the completely factorized model (which assumes that every variable is independent of all the others) called Base rate (BR), the Helmholtz Machine trained by the wake-sleep algo- rithm [4] (HWS), the same Helmholtz Machine where a mean field approximation was used for training (HMF) and a fully visible and fully connected sigmoid belief net- work (FV). Table1 displays the performances of all the mixture of trees models that we tested. The results are very good: the mixture of trees is the absolute winner for compressing the simple digits and comes in second as a model for pairs of digits.... ..."

Cited by 30

### Table 3.--Summary of programs run in double precision--16 digits

1971

"... In PAGE 8: ... SlOOPR by Lautenschlager employs orthogonal polynomials and uses a modification of Forsythe apos;s metpod [18]. From Table3 we see that the double precision version in 16 digits of S100PR performed best on test problem Y1, however, ranked last of the three programs on test problem Y2. Below are listed the actual coefficients and counts obtained for SlOOPR.... ..."

### Table 2: Average log-likelihood (bits per digit) for the single digit (Digit) and double digit (Pairs) datasets. Results are averaged over 3 runs.

2000

Cited by 68

### Table 2: Average log-likelihood (bits per digit) for the single digit (Digit) and double digit (Pairs) datasets. Results are averaged over 3 runs.

2000

Cited by 68

### Table 2: Average log-likelihood (bits per digit) for the single digit (Digit) and double digit (Pairs) datasets. Results are averaged over 3 runs.

2000

Cited by 68

2000

Cited by 1

### Table 4.2 lists the results from the double-digit recognition tests. The FHMM system showed an average relative improvement in word accuracy of 105% over baseline.

### Table 7.2: Average log-likelihood (bits per digit) for the single digit (Digit) and double digit (Pairs) datasets. Boldface marks the best performance on each dataset. Results are averaged over 3 runs.

1999