petr-kovalev/punched-cards-fashion-mnist

Object recognition by random binary data lookup for Fashion MNIST

'Punched cards' for Fashion-MNIST

Object recognition by sparse random binary data lookup. Based on this article

Performing single-shot Fashion-MNIST objects recognition by lookup over the most representative sparse input bit sets of the training data (out of 28⋅28⋅8 = 6272 bits per training sample)

Bit vector set similarity evaluation using the maximum spanning tree is described in this article

The same algorithm applied to the QMNIST dataset is here

The same algorithm applied to the Oracle-MNIST dataset is here

Program output:

Punched card bit length: 8

Average single-shot correct recognitions on fine-tune iteration: 13816, 13525

Top punched card per input:
Training results: 26439 correct recognitions of 60000
Test results: 4318 correct recognitions of 10000

Top 39 (5%) punched cards per input:
Training results: 37506 correct recognitions of 60000
Test results: 6161 correct recognitions of 10000

All punched cards:
Training results: 42378 correct recognitions of 60000
Test results: 6965 correct recognitions of 10000

Punched card bit length: 16

Average single-shot correct recognitions on fine-tune iteration: 17690, 17925, 18054, 18110, 18125, 18124

Top punched card per input:
Training results: 27040 correct recognitions of 60000
Test results: 4466 correct recognitions of 10000

Top 19 (5%) punched cards per input:
Training results: 38868 correct recognitions of 60000
Test results: 6367 correct recognitions of 10000

All punched cards:
Training results: 43809 correct recognitions of 60000
Test results: 7195 correct recognitions of 10000

Punched card bit length: 32

Average single-shot correct recognitions on fine-tune iteration: 22180, 22714, 23060, 23266, 23385, 23461, 23500, 23515, 23519, 23514

Top punched card per input:
Training results: 28434 correct recognitions of 60000
Test results: 4665 correct recognitions of 10000

Top 9 (5%) punched cards per input:
Training results: 40266 correct recognitions of 60000
Test results: 6628 correct recognitions of 10000

All punched cards:
Training results: 44518 correct recognitions of 60000
Test results: 7279 correct recognitions of 10000

Punched card bit length: 64

Average single-shot correct recognitions on fine-tune iteration: 27204, 27628, 27911, 28100, 28242, 28339, 28406, 28450, 28483, 28501, 28508, 28507

Top punched card per input:
Training results: 31868 correct recognitions of 60000
Test results: 5135 correct recognitions of 10000

Top 4 (5%) punched cards per input:
Training results: 39346 correct recognitions of 60000
Test results: 6427 correct recognitions of 10000

All punched cards:
Training results: 44765 correct recognitions of 60000
Test results: 7312 correct recognitions of 10000

Punched card bit length: 128

Average single-shot correct recognitions on fine-tune iteration: 31871, 32308, 32555, 32691, 32785, 32845, 32887, 32921, 32942, 32955, 32964, 32974, 32976, 32970

Top punched card per input:
Training results: 35077 correct recognitions of 60000
Test results: 5714 correct recognitions of 10000

Top 2 (5%) punched cards per input:
Training results: 39231 correct recognitions of 60000
Test results: 6445 correct recognitions of 10000

All punched cards:
Training results: 44752 correct recognitions of 60000
Test results: 7308 correct recognitions of 10000

Punched card bit length: 256

Average single-shot correct recognitions on fine-tune iteration: 35171, 35528, 35875, 36067, 36199, 36276, 36330, 36375, 36407, 36425, 36437, 36441, 36454, 36460, 36465, 36468, 36464

Top punched card per input:
Training results: 39199 correct recognitions of 60000
Test results: 6429 correct recognitions of 10000

Top 1 (5%) punched cards per input:
Training results: 39199 correct recognitions of 60000
Test results: 6429 correct recognitions of 10000

All punched cards:
Training results: 44519 correct recognitions of 60000
Test results: 7247 correct recognitions of 10000

Press "Enter" to exit the program...