Table 9 PVCQTL performance across three proposed approaches. L1, L2, L3 represents each approach ran at different localities.

From: Post-variational classical quantum transfer learning for binary classification

PVCQTL approaches comparisons

Dataset

PVCQTL

Accuracy (%)

Precision (%)

Recall (%)

F1-Score (%)

Run Time

L1

L2

L3

L1

L2

L3

L1

L2

L3

L1

L2

L3

L1

L2

L3

Ants and Bees

(Ep:30, Lr:0.004, Bs:4)

Mod. Obs. Const.

98.04

98.04

97.39

98.11

98.05

97.51

98.04

98.04

97.39

98.04

98.04

97.38

\(\approx 18\)

\(\approx 23\)

\(\approx 24\)

Hybrid

97.39

98.04

98.04

97.41

98.11

98.11

97.39

98.04

98.04

97.38

98.04

98.04

\(\approx 17\)

\(\approx 25\)

\(\approx 31\)

Variational-PV

98.04

98.04

98.69

98.11

98.05

98.72

98.04

98.04

98.69

98.04

98.04

98.69

\(\approx 30\)

\(\approx 59\)

\(\approx 78\)

Cats and Dogs

(Ep:10, Lr:0.004, Bs:8)

Mod. Obs. Const.

99.00

98.80

98.80

99.00

98.80

98.80

99.00

98.80

98.80

99.00

98.80

98.80

\(\approx 15\)

\(\approx 22\)

\(\approx 25\)

Hybrid

98.30

98.40

98.40

98.30

98.40

98.41

98.30

98.40

98.40

98.30

98.40

98.40

\(\approx 23\)

\(\approx 35\)

\(\approx 43\)

Variational-PV

98.50

98.40

98.30

98.51

98.40

98.30

98.50

98.40

98.30

98.50

98.40

98.30

\(\approx 42\)

\(\approx 81\)

\(\approx 107\)

Shirts and Pullover

(Ep:10, Lr:0.004, Bs:8)

Mod. Obs. Const.

96.32

95.7

95.2

96.31

95.75

95.2

96.31

95.7

95.2

96.30

95.7

95.2

\(\approx 15\)

\(\approx 22\)

\(\approx 25\)

Hybrid

96.33

95.42

95.91

96.34

95.43

95.91

96.35

95.41

95.82

96.33

95.42

95.42

\(\approx 23\)

\(\approx 35\)

\(\approx 43\)

Variational-PV

95.41

95.43

95.45

95.43

95.46

95.8

95.40

95.44

95.42

95.40

95.42

95.44

\(\approx 42\)

\(\approx 81\)

\(\approx 107\)

  1. The runtime information represents the combined training and validation time for each dataset. The hyperparameters used in these experiments are listed under each dataset as Epochs (Ep), Learning Rate (Lr), and Batch Size (Bs). Inference time per image was 0.004 secs in all the models.