EXAM 2 AI study guide

Monte Carlo tree search VS Minimax

1) It can handle large search spaces and problems with incomplete information. 2) It handles non-deterministic problems, where the outcome of a move or action is uncertain. 3) It doesn't require an evaluation function unlike minimax and beta pruning. 4) MCTS is more effective when there are many possible moves as it can focus on the most promising pathway rather then exploring all possible moves like minimax.

Costs of MCMTS VS MINIMAX and Alpha beta pruning

1) MCTS requires a larger number of simulations in order for it to be effective. This is time-consuming and computationally expensive. Contrasting minimax with a lower computational cost. 2) another cost of MCTS is its stochastic nature. It may not always converge to optimal solution. MCTS uses statistical simulations to obtain the value of each move and this introduces noise and inaccuracies in the results. Minimax, on the other hand, is a deterministic algorithm that guarantees to find the optimal solution in a two-player zero-sum game.

Genetic algorithms

A form of the evolutionary algorithm used to find the best solution to the problem by simulating the process of natural selection and genetic variation. The algorithm starts with a population of candidate solutions population evolved over a number of generations with each generation improving on the previous one. 1) Selection: A subset of solutions from a population are selected to form a new generation. 2) Reproduction: The selected solutions are then combined to create new solutions for the next generation. done through crossover involving randomly combining bits from two-parent solutions to create a new offspring solution. The mutation is applied which involves randomly flipping bits in solutions to introduce genetic diversity. 3) Evaluation: The fitness of the new solutions is then evaluated, and the best solutions are selected to form the next population. Genetic algorithms are effective when searching through a large number of possible solutions, as they can efficiently explore the search space and find good solutions. However, they are not guaranteed to find the best solution, as they may get stuck in a local optimum.

Beam search

A heuristic search algorithm is used to find the best solution to a problem in a search space with many possible solutions. It is called "beam search" because it works by maintaining a beam or window of the most promising partial solutions at each step of the search. Beam search can be an effective search algorithm when the search space is large and there are many possible solutions, as it can efficiently identify promising solutions and discard unpromising ones. However, it is not guaranteed to find the best solution, as it may get stuck in a local optimum

HILL CLIMBING

A heuristic, problem-solving strategy in which each step moves you progressively closer to the final goal. Hill climbing basically is starting at some point in a search space and moves in direction of the steepest gradient until a local optimum is reached.

local search algorithms

A type of optimization algorithm that tries to find the best solution to a problem by making small changes to the problem. Small changes keep being made until it cannot improve the solution any further or until it reaches a fixed stopping period. Local search algorithms are often used when the search space is too large to check all possible solutions, and can be effective at finding good solutions quickly but may not find a best possible solution.

Stochastic search algorithms

Are a class of search algorithms that use randomness to make decisions about which paths to explore in a search space. unlike deterministic search algorithms that always follow the same path. stochastic search algorithms randomly select actions or solutions to explore allowing the potential to find better solutions than deterministic algorithms. Types of stochastic search algorithm are the Monte Carlo method, which simulates a large number of random samples to estimate the likelihood of an event or to find the optimal solution to a problem. Or a genetic algorithm can also be used to figure out a solution in a unknown partially observable environment.

Describe Gilovich's account of the representativeness heuristic

Gilovich's account of representativeness heuristic refers to the cognitive bias that people use when making judgments about the likelihood of an event or the category membership of an object. Example(man in lab coat assumed a doctor) Gilovich argues that the representativeness heuristic can lead to errors in judgment because it can cause people to ignore relevant base rate information. Mini case study: Suppose that a hiring manager is reviewing job applications for a software engineering position. One candidate, John, has a beard, wears glasses, and is soft-spoken. The hiring manager may be more likely to assume that John is a good fit for the job based on their mental prototype of a software engineer, which might include characteristics such as being introverted, analytical, and detail-oriented. However, this assumption may overlook other factors that are equally or more important for the job, such as communication skills or ability to work well in a team.

Hill Climbing search

Goal driven informed search technique that focuses on goal and not path to the goalnmmm

statistical regression

Statistical regression is a method used in statistics to model the relationship between two or more variables. It is commonly used to make predictions or estimate the value of one variable based on the value of another variable.. Problem 1: , statistical regression can lead to problems in reasoning when it is misinterpreted or used improperly. One common problem is the confusion of correlation with causation. Just because two variables are correlated does not mean that one causes the other. It is possible that the correlation is simply a coincidence or that there is a third variable that is responsible for both. Problem 2: statistical regression is an overreliance on the model. The model is based on assumptions about the data, and if those assumptions are incorrect, the model may not be accurate. Problem 3: Statistical regression is misleading in the sample size is too small a small sample size does not represent a large population and as a result biased or inaccurate data will lead to inaccurate predictions.

In the context of genetic algorithms, explain mutation and crossover

The mutation is where random changes in some parts of a solution are made to keep diversity and avoid getting stuck in bad solutions. Crossover combines parts of genetic material from both parents to create a new and better one. both of these operations help the algorithm to find the best solution by exploring different possibilities.

Describe how well the standard approach to game playing would apply to games such as tennis, pool, and croquet, which take place in a continuous physical state space.

The standard approach to game playing typically involves modeling the game as a discrete set of states and actions. State transitions are solely determined by the actions of players. Games such as tennis, pool, and croquet take place in continuous physical state space. Players' actions are unpredictable and too many factors are at play. The state of the game is constantly changing as its an unknown environment making it difficult to model the game with a discrete set of states or actions. the agent would need to learn how to predict outcomes of a given action due to the continuous physical state space. Overall the standard approach would not be well suited for games that take place in continuous physical state space.

Explain how an agent might search of a solution in an environment in which the results of actions are not deterministic, or the environment is not completely observable, etc.

When a environment where results of action are not completely observable, the agent needs to use search algorithms that can handle uncertainty and partial observability. First approach is stochastic search algorithms such as Monte Carlo tree search. simulates multiple possible outcomes of an action and selects the most promising one. Monte Carlo tree search works by building a search tree where each node represents a state in the environment. The algorithm then simulates random actions from each node and updates the statistics of each node based on the outcome of the simulation. The selection of the next node to explore is based on a balance between exploration and exploitation, where the algorithm tries to explore unexplored nodes while also exploiting nodes that have a high probability of success. or can utilize a genetic evolution algorithm that produces a population and forms a new generation stronger then the previous solution. In summary, when the results of actions are not deterministic or the environment is not completely observable, the agent can use stochastic search algorithms to handle the uncertainty and find a solution.

Local optima

a solution better then all other solutions in its immediate vicinity, but it may not be the best possible solution for the problem as a whole. Optimization algorithms might face this problem as they get stuck at a local optima and fail to find the best global best solution.

bnOptimization problem

a type of problem where you are trying to find best solution among a set of solutions. The goal is to find a solution that maximizes or minimizes a certain objective, subject to a set of constraints or limitations.

Monte Carlo Simulation

an algorithm that repeatedly samples a problem space randomly in order to obtain a more accurate understanding of the best answers within it. 4 key specs 1)selection selection takes the current state of the tree and selects decisions down that tree to a future state 2)Expansion Moves one step down to expose a new state in the tree provided the state we reach didn't in the game 3) Simulation simulation is the random playout phase that plays a game of completely random decisions from this state until it reaches a terminal state where it wins or loses or a simulation cap. 4) finally, the outcome of the simulation is propagated back up the tree, updating the values of all the nodes that were traversed during the selection process.

SIMULATED ANNEALING

finds the best solution by trying different solutions and gradually improving them. Starting with an initial solution and then making a small change. If change makes the solution better it is kept the algorithm becomes more selective over time and only accepts changes that make the solution better. Simulating annealing is inspired by the process of heating and cooling materials to make them stronger. the algorithm allows for bad solutions early on but becomes more selective as it goes on to find the best solution. Examples of application: 1)The shortest route for a delivery driver. 2)optimizing parameters of a machine learning model. 3) designing a complex system

clustering illusion

tendency to see patterns in random events The clustering illusion is a cognitive bias where people see patterns in meaningless or completely random data. For example (people see shapes or images in clouds or arrangements of stars in the sky) bias creates an overestimation in significance causing people to overlook the role of randomness. Mini-case study of clustering illusion: A stock trader is analyzing stock prices for a particular company over the past month. Trader notices stock price increases every Monday and assumes a trend. Despite this clustering of stock price increases on specific days may also be a result of chance. the trader as a result of thinking this makes poor investment decisions based on illusion assuming pattern will continue in future when in reality, it may not.