Luckily you have a special instrument—let’s call it a slope-finder—that tells you which direction is most downhill from your current position. So you hike a little bit, check the instrument, change course, and hike a bit more.This is the famous analogy for gradient descent, a machine learning technique for (among other things) updating the weights of a nerual network.But it’s also a good analogy for problem solving in general: we move in the most promising direction, periodically updating that direction based on external feedback.So our original question—How much feedback should we absorb?—becomes: How often should we check our slope-finder?The perils of constant feedback
If you go through rest of the write up, you’d realise that ML is actually grappling with issues that the physicists of pre IMRT era had to- not getting trapped in the local minima and finding the global optimal solution. However, is the globally optimal solution really the “right one”? We don’t know because it is not mated to a quantitative output with a feedback loop.
(Spoiler: I am trying to establish this quantitative globally optimal solution through machine learning).