Optimism, Pessimism, and Algorithms
date: [2021-09-17 Fri 19:00:00] event: Friday Hacks speaker: Seth Gilbert
Why didn’t we invent bitcoin earlier?
- Lamport is a pessimist
- Satoshi is an optimist
Accountable blockchain protocol
- punishment instead of prevention
- reduce resistant to attack, but log enough info to detect bad behavior
- honest user can obtain evidence that proves bad behavior
Smoothed analysis of algorithms
- Spielman Teng 2004
- Why does the simplex method work so well?
- Main idea:
- Adversary choose worst-case input I
- Add random noise I’ = (I + r)
- Run algorithm on I’
- Works when
- bad cases are rare
- bad cases are unstable
Dynamic networks
- worst case results are bad
- use smoothed analysis
AI that isn’t depressed
- Bloom filter, no false negative, more space fewer false positives
- NN classifier, probably right if it says yes, no output may be a false negative
- Combine a bloom filter and NN classifier, BF -> NN -> BF
- Combine the optimism of machine learning and pessimism of algorithms