Evolutionary Algorithms Perfecting the Art of "Good Enough"
summary
Claim: evolutionary and heuristic optimizers do not guarantee the true optimum but reliably find pretty great solutions for combinatorial problems where 60 or more stops produce more routes than atoms in the universe. Implementation details include Python proofs of concept with GitHub code for hill climber simulated annealing MC³ genetic algorithms, hill climber is tiny and fast but gets stuck, simulated annealing cools from exploration to exploitation, genetic algorithms use populations and more memory. Worked examples cover knapsack TSP scheduling and ad spending and research use includes stable food web generation.
AI-written summary · about
chapters (13)
- 0:00Evolutionary algorithms intro
- 2:27Heuristic optimizers with randomness
- 8:35Hill climbing basics
- 12:25Simulated annealing explained
- 15:36Genetic algorithms overview
- 20:52Speed power tuning comparison
- 21:55Learning more on optimization
- 23:40Real-world uses
- 24:05Q&A on training sets
- 25:06Start with hill climbers
- 26:57Custom code vs libraries
- 28:06Choosing mutation functions
- 29:11Runtime with convergence plots
related talks
46mTight Genes: Intro to Genetic Algorithms
50mOptimal control, no! Than what? (...)
48mTight Genes: Intro to Genetic Algorithms
57mSolutions That Evolve: Building Self-Improving Systems with Genetic Algorithms
54mAmazing Algorithms for Solving Problems in Software
46mTight Genes: Intro to Genetic Algorithms
34mAlgorithms to live
33mA Hands-On Introduction To Multi-Objective Optimization