8th September 2003 - Genetic algorithms
Chris Watkins is going to look at some or all of the following issues (time permitting)
- genetic algorithms
- basic population genetics
- stochastic gradient ascent (reduced form of GA)
- batch mode stochastic gradient ascent as approximate EM
- channel capacity of evolution
The core reading is: Chapter 19 of David Mackay's book.
"Why have Sex? Information Aquistion and
Evolution"
Either download/buy the whole book, or get a subset containing Ch19 as pdf,
ps.gz
or 2-up
ps.gz.
Other things to look at:
- "Channel Capacity of Evolution", Chris Watkins 2003
- Using Expectation-Maximization for Reinforcement Learning
- A Survey of Optimization by Building and Using Probabilistic Models
- Simple statistical gradient-following algorithms for connectionist Reinforcement Learning