Question 1

How does a genetic algorithm work?

Accepted Answer

A genetic algorithm (GA) maintains a population of candidate solutions (chromosomes). Each generation: (1) evaluate fitness of every individual; (2) select parents using tournament, roulette-wheel, or rank selection  fitter individuals are more likely to reproduce; (3) produce offspring by crossover (combining genetic material from two parents); (4) apply mutation (small random changes) at a low rate; (5) optionally retain elite individuals unchanged. Over many generations, genes correlated with high fitness spread through the population, and the average fitness rises.

Question 2

What is Holland's Schema Theorem?

Accepted Answer

Holland's Schema Theorem shows that short, low-order schemata (building blocks  bit-string templates with some positions fixed) that have above-average fitness receive exponentially increasing representation in successive generations. Despite evaluating only N individuals per generation, a GA implicitly processes O(N) schemata simultaneously  this 'implicit parallelism' explains why GAs can search enormous spaces efficiently. The theorem motivates the building-block hypothesis: good solutions are assembled from short, fit sub-patterns.

Question 3

What is the difference between crossover and mutation in a GA?

Accepted Answer

Crossover (recombination) is the primary search operator: it combines two parent chromosomes to create offspring, mixing genetic material. One-point crossover splits both parents at a random locus and swaps the tails; uniform crossover independently decides each gene from which parent it comes. Mutation is a background operator that introduces small random changes (flipping a bit, or adding Gaussian noise for real-valued GAs). Mutation prevents premature convergence by maintaining diversity, but too-high mutation rate destroys good solutions  typically p_m = 1/L where L is chromosome length.

Question 4

Why does a GA struggle with deceptive functions?

Accepted Answer

A deceptive fitness function deliberately misleads the building-block assembly process: short sub-patterns that individually look fit actually combine to give poor fitness, while the true global optimum requires sub-patterns that look individually poor. Classic deceptive functions ('deceptive traps') have low-order schemata pointing toward a local optimum away from the global one. Standard GAs without diversity-preserving mechanisms (niching, fitness sharing, island models) often converge to the deceptive attractor. This is why no-free-lunch results are important: GAs aren't universally best.

🧪 Genetic Algorithm Visualizer: Evolution, TSP, and Optimization

What did you find?