Vanishing and exploding gradients on the How to Train Models track. Very small gradients stop learning. Very large ones make weights NaN. Residual connections, normalization, gradient clipping, and a smaller rate are the usual fixes.
This lesson assumes you already worked through Reading the curves.
The idea in practice
Log gradient norm. Clip if it spikes. If it is zero, check the loss and frozen layers.
A concrete check
goal = {
'track': 'How to Train Models',
'lesson': 'Vanishing and exploding gradients',
}
checks = [
'input available at decision time',
'score matches the real decision',
'failure case written down',
]
print(goal['lesson'])
for item in checks:
print('-', item)
Run the sketch locally if you have Python. The printout is a reminder of the checks, not a trained model. Replace the strings with the real inputs from your own example before you treat it as a design.
What usually goes wrong
Adding layers when the gradient is already zero. When this happens, stop adding parameters or tools. Fix the check, the data, or the permission, then run the same example again.
What to write down
- The input you are allowed to use at decision time.
- The output and the score or pass rule.
- One failure you will test on purpose.
- What you will not claim the system can do.
Practice
Say the first log you would read after a NaN loss.
Self-check
- Say Vanishing and exploding gradients in one sentence that mentions an input and an output.
- Name the failure mode in this lesson and the check that would catch it.
Done when: you can explain this lesson without the page open, and you have a written failure case.