Here are the key points about how neural networks compute:
- Each unit receives weighted inputs from other units plus a bias value
- The weighted inputs and bias are summed to produce the net input
- The net input passes through an activation function to produce the unit's final activation value
- This process is repeated for each unit in the network to compute the overall output
- Networks learn by adjusting the weights and biases to minimize error on training examples
So in summary, neural networks perform computations by propagating inputs through weighted connections, applying activation functions, and adjusting weights/biases during training. The distributed nature of this process enables neural networks to learn complex patterns from large amounts of data.