Posts

Sorted by New

Wiki Contributions

Comments

Out-of-sample error equals in-sample error plus a penalty for model complexity

Y.S. Abu-Mostafa, in explaining the VC inequality of PAC learning.

This phenomenon--that it takes less much less will-power to carry out a predefined plan than to continually make decisions while acting--is well-known to athletes, especially runners. I think most people who train would agree that the hardest part of a workout is getting started, and it takes surprisingly little will-power to finish. There have been some interesting studies showing that an experienced runner maintains a constant pace on a long run--he has an nonconscious sense of the pace he will be able to maintain, which is largely independent of his current motivational state. Training books never say "run as far as you can at this pace"--this would be mentally and physically exhausting. Runners are often urged to set a pace and distance for themselves before starting the run.

From my experience there's also a long-term feedback process that affects how hard it is to start the activity. If I set myself a workout plan that is too strenuous and leaves me exhausted after each workout, then it becomes harder and harder to start, and after a while, I'm not putting my running shoes on.

So if we can generalize these lessons to other activities that normally require will-power, they would be 1.) set a plan (on whatever timescale is sensible) and follow it so you're not exerting will-power constantly, and 2.) set yourself a pace so that if you follow your plan, you don't feel exhausted afterwards--that way it will be possible to sustain the plan in the long term. Regarding (2), people have suggested enforcing a schedule like "45 minutes work, 15 minutes break", though I haven't tried this myself.

I think that by far most-powerful willpower hack of all is making oneself accountable to other people, but that's a topic for another day.