笔记摘要

Algorithms to live by:
1. Optimal stopping problem: 37% for apartment hunting: grab the first after that threshold.
2. People don't need therapist, they need an algorithm.
3. Algorithm is not confined to mathematics. It stems from the Stone Age.
4. Science is a way of thinking rather than a body of knowledge.
5. Make decisions in dealing with uncertainty, time constraint, partial information and a rapidly changing world.
6. Chapter1: the secretary problem. When to stop looking. Look - 37% - Leap. Find the best after leap stage.
7. For full information situation: never compromise below 50% percentile.
8. The stopping threshold depends on the cost of searching (or waiting cost).
9. Similar problem: dating, buying / selling house, parking, hiring - all optimal stopping problem.
10. Explore - exploit trade off (gather/use information). The multi armed band problem.
11. Seize the day or seize the life - the interval makes the strategy.
12. The particular is the gateway to universal.
13. Calculate Giton's index (a number for the ratio of exploration-exploitation)
14. Jeff Bezos: minimal regret decision
15. Choose the arm with highest CI bar
16. Searching - Sorting: only sort in the context of future searching, find the optimal balance of two.
17. Caching: forget it. It's as important of forgetting as remembering (William James).
18. LRU: least recently used item for caching eviction
19. Some hypotheses are more likely than others (prior).
20. Mine people's expectations (our priors are pretty accurate).
21. Overfitting: over thinking. Ocma's razor. Penalize complexity. If you couldn't explain it in simple ways you don't understand it fully. The added the complexity should contribute significantly better.
Portfolio model: mean variance portfolio optimization. Only when you have confidence in the estimate.
Regularization: deliberately thinking and doing less
22. Traveling salesman problem: constraint relaxation. Continuous relaxation. Erlang relaxation: exponential Erlang heuristic.
23. Randomness: sampling. Some problems can only be solved using random sampling: polynomial identity problem. Also, is it free or equal that makes the society more justice?
24. The uniqueness of CS: articulate the complexity of a problem. Always looking for trade off: time, space, error probably (certainty)
25. Mathematicians are trained to work against their instincts.
26. Poor listeners destroy the tale. The feedback from listeners determine how well a story is told. Feedback channel.
27. Successful deals come from anticipating the anticipation of others
28. Mechanism Design: the opposite of game theory. Change rules could fundamentally change people's behaviors. Like shop open in Sunday or people taking vacation.
29. Spinning vs Blocking: waiting in restaurant problem. Former, unnecessary vigilance.
30. Design for cognitive kindness.
31. Algorithm is only good for easy problems. When situations get complicated, there is no way to find optimal solution. Human thing is superior then.