Exam Glossary
Canonical terms from the question bank — use these exact words in your answers.
Genetic Algorithms
- Population
- Subset of all probable solutions to the problem.
- Chromosome
- One candidate solution within the population.
- Gene
- A single element within a chromosome.
- Genotype
- Encoded internal representation (binary string, array) used by operators.
Not the same as phenotype — genotype is internal, phenotype is evaluated.
- Phenotype
- Decoded real-world solution evaluated by the fitness function.
- Fitness function
- Scores how good a solution is; guides evolution toward better answers.
- Elitism
- Carrying best individuals unchanged into the next generation.
- Immigration
- Moving individuals between sub-populations to spread good genes and increase diversity.
- Premature convergence
- Population loses diversity and settles on a suboptimal solution too early.
Operators
- One-point crossover
- Swap segments after a single cut point.
- Two-point crossover
- Swap the segment between two cut points.
- Tournament selection
- Pick the best from a random subgroup.
- Bit-flip mutation
- Flip bits in binary chromosomes.
- Gaussian mutation
- Add random noise to real-valued genes.
- Opposition-based learning
- Evaluate both a solution and its opposite.
Local Search
- Hill Climbing
- Iteratively move to better neighbors until stuck.
- Local maximum
- Peak surrounded by worse neighbors but not globally best.
- Plateau
- Flat region — all neighbors equal fitness.
- Ridge
- Diagonal slope where single-step moves can't improve.
- Simulated Annealing
- Probabilistically accept worse moves early, tighten over time.
Deep Learning
- Transfer learning
- Reuse pre-trained model on a new task.
- Frozen layers
- Weights fixed — not updated during training.
- Trainable layers
- Weights updated during fine-tuning (usually last layers).
- Fine-tuning
- Continue training pre-trained model on new data.
- Additive fine-tuning
- Add new layers/adapters; keep original weights frozen.
- Inception module
- Parallel 1×1, 3×3, 5×5 convs + pooling, concatenated.
- 1×1 convolution
- Dimensionality reduction before expensive filters.