Jump to content

recreating early ai experiments - character recognition


Recommended Posts

  • Moderators

Recreating one of the early ai experiments : recognising handwritten numbers

 

The MNIST dataset features some 60,000 handwritten characters (digits), captured in 28x28 pixel frames

 

In this experiment only the first 200 of these 60,000 digits are used

Training based on this small trainingset only takes about 30 seconds

 

The testset contains some 20 digits randomly taken from the dataset

These digits have not been used in the training

 

Given these constraints it all performs amazingly well

 

Give it more trainingdata and increase training time and it only becomes better

 

Projectcode

 

 

Capture.JPG

 

 

Link to post
Share on other sites
  • Moderators

The images are in a 28x28 matrix

this translates to a neural network of 28*28=784 input neurons (1 for each pixel), 100 or so hidden neurons and 10 output neurons (digits 0-9)

 

so each picture translates to an input format of "array[784] of float", filled with rgb values i.e. [0,0,0,0,255,143, ... 0,0,0]

 

and a target output format of "array[10] of float" i.e. [0.01, 0.01, 0.99, 0.01 .... 0.01] where the output value in this example denotes a '2'

 

note :

although rgb values are integers, I used floats in the input format as all rgb values are recalculated to a value between 0.01 and 0.99

(TrainingSet.TrainingRecords[f].&inputs[q]/255*0.99)+0.01)

Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...