Exam Cheat Sheet

Population: This is a subset of all the probable solutions that can solve the given problem.
Chromosomes: A chromosome is one of the solutions in the population.
Gene: This is an element in a chromosome.
Fitness function: This is a function that uses a specific input to produce an improved output.

The genotype is the encoded representation of a solution, typically in the form of a binary string, array, or other structured format. It's the internal format used by genetic operators like crossover and mutation during evolution.

The phenotype is the actual solution as interpreted from the genotype. It is the version of the solution evaluated by the fitness function and used in the real-world problem domain.

The fitness function evaluates how good each solution is. A well-designed fitness function guides the algorithm toward better solutions, while a poor one can lead to misleading or suboptimal results.

One-point crossover: A single crossover point is chosen, and the segments after that point are swapped between the parents.
Two-point crossover: Two crossover points are selected, and the segment between them is exchanged.

In Single-point Crossover, a location is selected randomly on both the parents. The location is the same in both the parents chromosome. The genes to the right of the crossover point is swapped between the two parents.

In Two-point Crossover, two locations are selected randomly in the parents' chromosome. Then the genes in between the two points are swapped between the two parents.

Types include bit-flip (for binary), Scramble Mutation, swap (for permutation), and Gaussian mutation (for real-valued). The choice depends on the chromosome representation.

It randomly selects a group of individuals and chooses the best among them for reproduction. It's simple and effective for maintaining diversity and avoiding premature convergence.

When diversity in the population is low or the search is stuck in local optima, mutation helps by introducing new genetic material, whereas crossover relies on existing traits and may not help in exploration.

Elitism ensures the best individuals from the current generation are carried to the next without modification, helping preserve high-quality solutions.

It's a method where both a candidate solution and its opposite are evaluated. It helps escape local optima by considering alternative search directions.

Immigration refers to moving individuals between separate sub-populations during evolution. It allows high-quality solutions to spread, increases genetic diversity, and helps avoid premature convergence by introducing new genetic material into each sub-population.

Hill Climbing is a local search optimization algorithm that starts with an initial solution and iteratively moves towards better neighboring solutions. It stops when no improvement can be made. It's simple but can get stuck in local optima.

• Simple Hill Climbing: Moves to the first better neighbor found.
• Steepest-Ascent Hill Climbing: Evaluates all neighbors and chooses the best one.
• Stochastic Hill Climbing: Randomly selects a better neighbor.

A local maximum occurs when no neighbors improve the solution, but better solutions exist elsewhere. It can be addressed by random restarts or backtracking to explore different regions of the search space.

A plateau is a flat region where neighbors have the same value, making it hard to decide the next move. It can be addressed by larger steps or random restarts to escape the plateau.

A ridge is a region where the algorithm can't progress due to the slope of the landscape. It can be overcome by larger steps or using simulated annealing to allow some "downhill" moves, helping escape ridges.

SA is inspired by the metallurgical process of annealing, where a material is heated to a high temperature and then slowly cooled to reduce defects and achieve a more stable, low-energy state. Similarly, SA starts with a high "temperature" to explore the search space broadly and gradually "cools" to refine the solution, helping it escape local optima and converge toward a global optimum.

Temperature: Controls the acceptance probability of worse solutions. High temperatures allow more exploration.
Cooling Rate: Determines how quickly the temperature reduces.
Initial Solution: Starting point for the search.

GA is inspired by genetics and uses selection, crossover, and mutation. PSO is inspired by social behavior and uses velocity and position updates. GAs are more explorative; PSO tends to converge faster.

Transfer learning is a method where a model trained on one task (e.g., ImageNet classification) is reused and adapted for another task. This reduces the need for large datasets and allows faster development of models for new applications.

Frozen layers are layers whose weights are fixed during training. They retain the pre-learned features from the source dataset and help maintain general knowledge, especially in the early convolutional layers.

Fine-tuning is the process of continuing training a pre-trained model on a new task or dataset. It usually involves unfreezing some layers and updating them to specialize the model without losing its general capabilities.

Additive fine-tuning involves adding new layers or adapter modules to the pre-trained model while keeping the original weights unchanged. These new components are trained for the target task, preserving the base model's stability.

Trainable layers are updated during fine-tuning. They are typically the last few layers in the network, and they adapt the model to the new task by learning task-specific patterns from the new dataset.

VGG16 uses 3x3 convolutional filters, consistent architecture, and depth (16 layers) to learn complex image features. Its simplicity and uniformity make it widely used, despite its large number of parameters (138 million).

GoogleNet uses Inception modules to process features at multiple scales using 1x1, 3x3, and 5x5 convolutions in parallel. It significantly reduces the number of parameters while maintaining high accuracy on image classification tasks.

An Inception module is a network block that applies multiple convolutions (1x1, 3x3, 5x5) and pooling operations in parallel, then concatenates the outputs. This allows the model to learn both local and global features simultaneously.

1x1 convolutions are used for dimensionality reduction. They reduce the number of input channels before applying larger filters, making the model more efficient and reducing computational cost.