A neural network is a function: it takes input and produces output. In between, however, it uses layers of hidden neurons to perform calculations that affect the end output. Dr. Yasaman Bahri is a computer scientist at Google Brain who is working hard to understand how neural networks learn to perform tasks! To visualize how different connections, layers, and neurons can affect the training process, check out this Neural Network Playground!
Start by choosing a dataset on the left top corner and hit "RUN" to see how the connections between nodes evolve throughout training (represented by the dashed line). Also watch how the outputs of each individual node changes (by hovering over the the squares in the middle hidden layer), and how these tweaks affect the final output (color-coded in big square on the right most).
Depending on the dashed lines, the final output can be very different, even though there is only one simple activation function (a hyperbolic tangent function in this screenshot)!