In essence, training a neural network may be comparable to teaching a robot how to play chess.
First, you let it watch lots of chess games, showing it different moves and strategies (data collection stage). The robot makes guesses about the best moves to play, and you tell it how good or bad those moves were (forward pass stage).
Then, it tries to improve its guesses by making small adjustments to its thinking process (back-propagation stage). This process is repeated many times until the robot gets really good at understanding the game.
In the end (inference stage), the robot can play chess on its own and make smart moves.
In most of the cases, building any kind of NN (Neural Network) requires to use either Pytorch, Tensorflow or Keras. But, can we build our own Neural Network?
Let’s find out!
In the above simple description of how a neural network learns, you can find pipeline stages in brackets like: data collection, forward pass, back-propagation, and inference (prediction).
In more technical details, we would need to implement stages like:
- Prepare Data: Collect, preprocess, and format the input data. Ensure the data is clean, normalized, and structured for…
