Table 1 TTN prediction time.

From: Quantum-inspired machine learning on high-energy physics data

 

Model M16 (incl. all 16 features)

Model B8 (best 8 features determined by QuIPS)

χ

Prediction time

Accuracy

Free parameters

Prediction time

Accuracy

Free parameters

200

345 μs

70.27% (63.45%)

51,501

–

–

–

100

178 μs

70.34% (63.47%)

25,968

–

–

–

50

105 μs

70.26% (63.47%)

13,214

–

–

–

20

62 μs

70.31% (63.46%)

5576

–

–

–

16

–

–

–

19 μs

69.10% (62.78%)

264

10

40 μs

70.36% (63.44%)

1311

19 μs

69.01% (62.78%)

171

5

37 μs

69.84% (62.01%)

303

19 μs

69.05% (62.76%)

95

  1. Prediction time, accuracy with (and without) applied cuts Δ and number of free parameters of the TTN for different bond-dimension χ when we reduce the TTN model with QIANO, both for the complete 16 (left) and the QuIPS reduced 8 features (right). For the model M16 with all 16 features (left), we trained the TTN with χ = 200 and truncate from there while for the reduced model B8 (right), the original bond-dimension was χ = 16 (being the maximum χ in this subspace).
Back to article page