Share your repls and programming experiences

← Back to all posts
Demonstration Of Effects Of Transfer Learning On Simple Networks

After a bit of tweaks from geeksforgeeks, I have created a system that can transfer learn, and add, subtract, and multiply (although it does not fully generalize). Here we see an AI that was trained on n x 10 without background knowledge, and next we see an AI that was trained on n x 5 and then n x 10, the second AI is more precise than the first and has a fixed value, while the first AI can have varying answers from run to run, sometimes being precise, other times being kinda off..
(The most off I've seen for the first AI is about 4142 as the answer, I've seen barely any fluctuation from the second one that used transfer learning.)