Think Together · Start with a Short Story
A picture of a number enters the robot, passes through several transformations, and finally produces a result like “probably a 3.”
Lay Out the Process
- Pixel representationNumerical input
- First layerCalculate an intermediate representation
- Later layersCombine it further
- Output layerClass scores
A forward pass sends input through a network structure to produce intermediate results and, finally, an output.
Each layer uses its own parameters to process the previous layer's result. For the same trained model, a new input follows the same structure but may produce different numbers.
During training, error information is also used to calculate how the parameters should change. That is a different direction of computation.
Now You Try
If only some pixels in the picture change, could the output change too?
Ready? Open this to see the explanation
It could, because the intermediate calculations depend on the input. How much and in what direction depends on the model.
One Idea to Take Away
A forward pass sends input through a network structure to produce intermediate results and, finally, an output.
Try covering the diagram above and explaining each step to someone in your family. If one step is hard to explain, return to that card and look again.