All popular AI systems that you see today are made up of basic computing units called Neurons or Perceptrons. These connect to form a network called a Neural network that performs various complex computations.
Very well written! Made me think about how often people jump to using tools without understanding what’s happening underneath. How do you approach deciding the number of layers and neurons in a neural network for a new problem, especially when there isn’t a clear reference point or prior work to guide the architecture?
Deciding the number of layers and neurons in a neural network is a trade-off between accuracy and training compute/ inference latency. Most of this comes from trial and error.
Very well written! Made me think about how often people jump to using tools without understanding what’s happening underneath. How do you approach deciding the number of layers and neurons in a neural network for a new problem, especially when there isn’t a clear reference point or prior work to guide the architecture?
Hi Chetan. Thank you!
Deciding the number of layers and neurons in a neural network is a trade-off between accuracy and training compute/ inference latency. Most of this comes from trial and error.
This is the most concise and easy explanation on neural network training.
Though I have some questions.
Thank you! Happy to answer them.