I took 20 images from the MNIST dataset and transformed them in three ways: 1) as single-line strings with pixel values (0-256), 2) encoded as Base64, and 3) as JPEG binary (hex). I then fed each format into a model, using 10 examples and 10 test samples for predictions. So far, no model has exceeded 20% accuracy on any format. When will a general model achieve 70% or more on any of those?
1,000@biased
Your goal is to predict the correct label of a MNIST image based on its string unrolled representation.
10 examples with answers
single test
######################################
Your goal is to predict the correct label of a MNIST image based on its base64 representation.
10 examples with answers
single test
######################################
Your goal is to predict the correct label of a MNIST image based on its binary (hex) JPEG representation.
10 examples with answers
single test
@biased Yes, I did in in single shot for all 10 examples, because I didn't want to waste all my queries on that, but it was not better than random chance.
Examples: (1) single-line string
9;0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,31,53,143,143,206,253,210,153,229,105,3,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,44,152,246,252,252,252,253,252,252,252,252,252,128,0,0,0,0,0,0,0,0,0,0,0,0,0,0,19,237,252,252,252,252,252,253,226,153,153,218,252,241,0,0,0,0,0,0,0,0,0,0,0,0,0,0,137,252,252,242,143,10,10,10,8,9,80,138,171,241,0,0,0,0,0,0,0,0,0,0,0,0,0,22,245,252,233,55,0,0,0,0,15,132,252,252,156,84,0,0,0,0,0,0,0,0,0,0,0,0,0,100,252,252,117,0,0,0,0,0,105,252,252,252,126,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,238,252,245,178,74,40,108,184,246,252,252,173,29,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,65,252,252,252,252,252,252,253,252,252,200,34,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,6,66,151,246,239,175,209,253,252,252,10,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,30,27,0,111,253,252,209,6,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,207,255,253,139,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,139,253,206,13,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,220,253,186,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,7,223,253,176,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,78,252,253,77,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,78,252,242,50,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,178,252,220,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,27,227,252,220,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,45,252,252,138,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,25,224,204,63,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
@MikhailDoroshenko
Base64
9;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
@MikhailDoroshenko Binary (hex) JPEG
9;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