AAI_501 Midterm
A* search
avoids expanding paths that are already expensive uses an evaluation function f(n)=g(n)+h(n) -g(n): cost so far to reach n -h(n) estimated cost to goal from n -f(n):estimated total cost of path through n to goal
nondeterministic/stochastic
if the next state of the environment is NOT determined by the current state and action by the agents
Shopping for used AI books on the internet Environment: fully observable versus partially observable deterministic versus non deterministic/stochastic Episodic versus sequential Static versus dynamic Discrete versus continuous Agent: Single agent versus multi agent
partially observable, deterministic, sequential, static, discrete, single agent. (can be multiagent and dynamic if via auction. Can be dynamic if its over a long period of time and prices change)
Exploring the subsurface oceans of Titan Environment: fully observable versus partially observable deterministic versus non deterministic/stochastic Episodic versus sequential Static versus dynamic Discrete versus continuous Agent: Single agent versus multi agent
partially observable, stochastic, sequential, dynamic, continuous, single agent.
goal-based agent
-agent needs a goal in order to make a decision -the behavior of the agent can be easily changed to a different destination by specifying that the destination as goal - more flexible
Main principles of Hidden markov model
-comprises of hidden variables and observables. The state of the process is described by a single discrete random variable. -the observed events have no one-to-one correspondence with states but are linked to states through the probability distribution.
utility based model
-goals are alone not enough to generate high quality behavior -many aciton sequences willl get a taxi to a destination but how about cheaper, quicker, safer? these are defined as utility funtions. -utility fuinction is internalization of performance measure. -when there are conflicts(speed, safety), utility function provides the appropriate tradeoff
Plateau
A flat area of search landscape. It can be flat from which no uphill exits exists or a shoulder from which progress can occur
Local maxima
A peak that is higher than neighboring states but lower than global maxima.
simulated annealing
Allows moves to inferior solutions in order to not get stuck in a poor local optimum. • Dc = F(Snew) - F(Sold) F has to be minimized •Inferior solution accepted with a probability; Generate random number U: (0,1) where u<e^-(deltac/t) •As temperature decrease, probability of worse moves decrease (t is changed with a cooling schedule)
fully observable
An agent's sensors give it access to the complete state of the environment at each point in time. A task environment is effectively fully observable if the sensors detect all aspects that are relevant to the choice of action.
dynamic
If the environment can change while an agent is deliberating
deterministic
If the next state of the environment is completely determined by the current state and the action executed by the agent(s)
Provide a complete problem formulation for the following. Problem formation should include: initial state, goal test, cost function. They can all be a few words or sentence for each. You have a program that outputs the message "illegal input record" when fed a certain file of input records. You know that processing of each record is independent of the other records. You want to discover what record is illegal.
Initial state: considering all input records goal state: considering a single record and it gives illegal input message cost function: number of runs
Provide a complete problem formulation for the following. Problem formation should include: initial state, goal test, cost function. They can all be a few words or sentence for each. Using only four colors, you have to color a planar map in a way such that no two adjacent regions have same color
Initial state: no regions colored goal state: all regions colored and no two adjacent regions have the same color cost function: number of assignments
Your goal is to navigate a robot out of a maze. The robot starts in the center of the maze facing north. You can turn the robot to face north, east, south, or west. You can direct the robot to move forward a certain distance although it will stop after hitting a wall. a) Formulate this problem. This means you will have to describe initial state, goal test, successor function, and cost function. Successor function is a description of the robots successive actions after the initial state. We'll define the coordinate system so that the center of the maze is at (0, 0), and the maze itself is a square from (−1,−1) to (1, 1)
Initial state: robot at coordinate (0,0) facing north goal state: either x>1 or y>1 where (x,y) is the current location successor function: move forward any distance d; change the direction robot is facing cost function: total distance moved
Depth First Search
Visits the child vertices before visiting the sibling vertices A stack is usually implemented
breadth first search
Visits the neighbor vertices before visiting the child vertices A queue is usually implemented
particle filters
a tracking problem. the goal is to keep track of the current location of the system given noisy observations. In order to estimate the state, two sources of information is required. A process model and a measurement model. This algorithm is an iterative process, that is repeated for a certain number of samples. Each sample is weighted by the likelihood it assigns to the new evidence. The probability that a sample is selected is proportional to its weight. It is a Bayesian filter which means estimation is preformed using Bayesian theory. The main idea is updating or pushing beliefs through transitions.
how to find metric in uniform cost search
add the weights of the edges of the graph
what is the frontier
all unexpanded nodes (nodes with no children)
rational agent
chooses whichever action maximizes the expected value of the performance measure given the percept sequence to date
Playing a tennis match Environment: fully observable versus partially observable deterministic versus non deterministic/stochastic Episodic versus sequential Static versus dynamic Discrete versus continuous Agent: Single agent versus multi agent
fully observable, stochastic, episodic, dynamic, continuous, multiagent.
Bidding on an item at an auction Environment: fully observable versus partially observable deterministic versus non deterministic/stochastic Episodic versus sequential Static versus dynamic Discrete versus continuous Agent: Single agent versus multi agent
fully observable, stochastic, sequential, static, discrete, multiagent.
static
if the environment can not change while an agent is deliberating
discrete
if an environment as a finite number of states(chess)
continuous
if an environment as an infinite number of states(driving a car)
single-agent
just one agent (like someone solving a crossword puzzle)
multi-agent
more than one agent (like two people playing chess)
Partially Observable
noisy and inaccurate sensors or because parts of the state are simply missing from the sensor data
What is an admissible heuristic?
one that never overestimates the cost to the goal the estimated cost must always be lower than or equal to the actual cost of reaching the goal state.
simple reflex agent
takes actions based on the current percept ignoring the rest of the percept history.
episodic
the agent's experience is divided into atomic episodes. In each episode the agent receives a percept and then performs a single action. Crucially, the next episode does not depend on the actions taken in previous episodes
sequential
the current decision could affect all future decisions.
Main concepts of first order Markov models
the current state only depends on the previous state. this can be used to simplifier the conditional probability
•Which node will be expanded next in the frontier if uniform cost search (Dijkstra's) is used?
the one with the least cost
How does Dijkstra's algorithm work
•Dijkstra's Algorithm finds the shortest path between a given node (which is called the "source node") and all other nodes in a graph. •This algorithm uses the weights of the edges to find the path that minimizes the total distance (weight) between the source node and all other nodes.
Genetic Algorithms
•Initialize population •Selection •Crossover and mutation •Replacement •Repeat until number of generations or best fit solution
Local Search
•Local search algorithms can solve optimization problems •If the elevation corresponds to an objective function, the goal is to find global maximum or global minimum (gradient descent)
tabu search
•Move to best available neighborhood solution •Maintain tabu list (moves to avoid) maintains a list of solution points that must be avoided. this is called the tabu list which is updated based on memory. certain moves are removed after a time period. the algo allows for exceptions from the list aspiration criteria. it also expands the search area and allows modifying size of list.
A star search
•Uses a combination of two functions •g(n) =cost so far to reach node n •h(n) = heuristic (estimated cost to reach destination from n) •f(n)=g(n)+h(n) ; Expand node with lowest f(n) Greedy search Expands the node with the lowest h(n) - node that appears closest to goal.