Brian Christian. Tim Griffiths – Optimal stopping. Without info and defined candidate, see 37% and chose next good one. 37% chance to get best. We tend to stop early. Look then leap.
Explore/exploit. Sorting theory Caching theory. Computers don’t just crunch. They estimate. Guess. Chance. Trade tine with accuracy. Mistakes by humans say more about difficulty of problems we face rather than thinking. Challenges. Time constraint, incomplete info, uncertainty, changing conditions.
Real logic includes:Don’t always consider all options (stopping )Don’t always go for best outcome every time. Make a mess, flip a coin sometimes. Forget don’t forget. Don’t always consider
Meryl flood prisoners dilemma and traveling salesman. 37% of secretary problem. Chapter 8Not all bell curvePower law. Scale free. Spreads over many scales. Many things below mean, a few way above – income, movie revenue. Preferential attachment produces power. Celebrating sites bring people it. Baize rules. Predict. What ever age. Double.
Overfit to met incentives. They work. We built what CEO says to measure. Optimize the wrong thing. Wrote training of collecting casings. Number of cases, go after easy only. Win per stage
Cross validation. Testing. How results generalize to other data.
Baize rules.
If you don’t know much about a complex situation. – stock market. Best to roll with itChapter 4Less can be more. Useful restraint. Use reptile jaw bones for ear. Don’t over adjust. (Off band wagon) early stopping. Too many data points can overfit. Accurate unstetdanig bit over adjust. Early stopping. When to think less. What u can measure and what really matters. High uncertainty and little data, go for it. Broad brush. Fat sharpey, bit pen.
Least recently used. Throw out oldCache. Clothes by bed. Most recently accessed. Version. Of LRU
Forgetting curve. Mind tuned to remember recent things including words and ideas. Cached and drift away over time.Cognitive decline is in part result of knowing so much.
Chapter 7 through put vs interruption. Stay on single task as long as possible and still stay responsive. Interruptions diminish productivity
Chapter 8 Baize rule. Reasoning in uncertainty. La place. Interference backwards. If no nothing, win # of wins +1 / # of attempts +2Diff if we have info such as lifespan of humans. Things always go on just a bit longer in erling distribution. Power, normal, erling – endless. Equally likely. Roulette
Chapter 9 relax. Seating Combinations of wedding. Far too high to calculatePerfect solutions untraceable. No keep working, not give up. Just relax. Relax constraints then solve and try adding back in. Traveling salesman problem and let him visit One city twice. Discreet optimization. Police distribution Lagrangian. Impose penalty on variance.
Chapter 10 randomness When to use chance. Sampling better in some situation cause it provides an answer. Monte Carlo method. Finding big prime. Show probability that a number is prime. Random test of the number. Anealing way substance change state as heats or cools. Rate big impact on final shape. (Temp really velocity)
Chapter 11 connection. Requires confirmation. Phone call and response. Packet switching instead of circuit switching (always connected). Packets go on and off. Each packet finds it own way. More paths, more stable. Opposite of phones. Ax packets send and revicier have numbers. Ready for 101. If102 shows, buffers and says ready fir 101. If 101 arrives, say ready for 103. Skype not use ax, we do it fir them.
Exponential backoff. Dropped connection. More time between retries. Not give up and not waste too many resources. Justice. Begin with one day in jail and increase each time. Gridlock ramp up till first drop. Radical drop no then increase slow each success. Additive increase, multiplicative decrease. Cut back by half. Language. Interruption important. Back channel. Role of listeners. Poor feedback destroys tale.
Buffer load. Cue. Recursion. I know that you know that I know.