AI Writing Reader
- 3 Devlogs
- 7 Total hours
This AI will use a neural network to read handwriting (pleasee i need 100% accuracy 🙏). Also I won’t be using any imported libraries except pre-installed ones and struct to read the MNIST dataset.
I have finally finished all the code, but still need to debug ts, bro this is gonna take forever but im almost done. Basically, i have built the backpropogation system for the AI which calculates all the partial derivatives of the equation, multiplies them to get the gradient descent value, then adds it to the weights and biases of the layers. This should theoretically work when i debug this. This is all done without the use of AI and any packages apart from random, which is pretty cool
Now I have finished building the structure for the AI to be built on, like the foundations are now ready for the house to be built. Now I just need to create the AI logic, which shouldnt be too bad (I hope)
I built the core functions needed in the AI, like the normalising functions softmax and sigmoid, the dot multiplication and matrix multiplication and addition functions and used those to overload operators of a Matrix class. I also started to develop the layer class but didnt get far. Most of the time went to researching the formulas for the functions and researching how neural networks work from 3blue1brown. I am mainly following his research from his youtube channel for now then i will add some personal additions later. Its a bit difficult adding comments everywhere since no one really ever looks at my work, i just do it for myself mainly so hopefully i did good at that this time.
