Exam 1

d

depth of least cost solution

Agent

perceives its environment through sensors and acts on that environment through actuators PEAS-performance, environment, actuators, sensors Rational agent maximizes performance measure with percept sequence and built in knowledge

g(n)

path cost function cost so far to reach n cost from initial state to n

simple queue

FIFO (BFS)

time complexity

# of nodes expanded/generated how long does it take to find solution?

Space Complexity

# of nodes in memory how much space is used?

Utility-based Agent

(1) If there are many paths (2) Measures which states are preferable; maps states to real number

Goal-based Agent

(1) Info guides actions (looks to the future) (2) Can require reasoning about long sequences of actions (3) Flexible (by changing goals)

Model-based Agent

(1) Maintains internal state that keeps track of the world it can't see now (2) Needs model (to encode knowledge about how world works)

Simple Reflex Agent

(1) Selects actions using current precept (2) If condition then action (3) Easy to implement and uses little memory but only works in fully observable environment and there are infinite loops

consistent heuristic

(monotonic) if for each node n and successor n', h(n)_<c(n,n')+h(n')

Depth Limited Search

A depth first search with a predetermined depth limit. Incomplete if limit is less than depth. Not optimal if limit is greater than depth. time: O(b^l) space: O(bl)

Iterative Deepening Search

A depth limited search that gradually increases the limit l. Generates states multiple times. complete if b is finite optimal if step costs are identical time: O(b^d) space: O(bd) i.e., linear

Breadth First Search (BFS)

A search algorithm that processes all the nodes in a layer, one layer at a time at ever increasing depths complete if b is finite not in infinite spaces or spaces with loops optimal if step costs are identical time and space: O(b^d); exponential FIFO queue goal tests each child immediately finds shortest length solution

Depth First Search

A search that processes all children of a node before sibling nodes; deepest node first not complete (only when m is bounded because fails infinite spaces and spaces with loops) not optimal (because not guaranteed that first node found is best) time: O(b^m) space O(bm) i.e., linear LIFO stack (goes down to the last/deepest)

Uniform Cost Search

Expands the node with the lowest path cost. Complete as long as every step cost is greater than some small positive constant. optimal (because always expands node with lowest path cost time and space: O(b^(1+[c*/E]) Done by storing the frontier as a priority queue. Goal test applied during expansion when removed from frontier not generation. Expands nodes in order of optimality.

environment properties Fully observable, partially observable, single agent, multiagent, deterministic, stochastic, episodic, sequential, static, dynamic, discrete, continuous, known, unknown.

Fully observable (access complete state) VS partially observable Deterministic (next state determined by current state and agent's actions) VS stochastic (in partially observable; deterministic except actions of other agents) Episodic (e.g., coin toss) VS sequential (current decisions affects all future decisions) Static (no time concern) VS dynamic (environment changes while agent is thinking) Known (rules and functions specified) VS unknown

stack

LIFO (DFS)

completeness

algorithm finds a solution if a solution exists, will it be found?

optimality/admissibility

algorithm finds solution with a minimum cost if solution found, is it guaranteed to be optimal?

c*

best goal path cost

f(n)

evaluation function estimated total cost of path through n to the goal g(n)+h(n)=f(N)

Greedy Best First Search

expands to node that has the least estimated cost to goal f(n)=h(n) not complete in tree search because could get caught in dead end's cycles complete in graph search because finite state, can't repeat states, but not optimal same time/space as DFS

A*

f(n)=g(n)+h(n) if h(n)=0, UCS complete and optimal UNDER CERTAIN CONDITIONS if tree search, must be admissible if graph search must be consistent/montonic also complete requires positive step costs that are not arbitrarily small straight line distance always _< actual shortest path expands nodes in order of increasing f-values so the first goal has the least possible f-value optimally efficient i.e., will always expand the fewest nodes high space complexity because must keep all nodes in the explored set

h(n)

heuristic function estimated cost from n to goal

b

max branching factor of search tree

m

max depth of state space (may be infinity)

admissible heuristic

never overestimates true minimum cost from n-state to goal state h(n)_< h*(n)

priority queue

removed according to priority (e.g., cost)

Bidirectional Search

two simultaneous searches, one from the initial state and from the goal state different algorithms may be used but one must keep entire frontier in memory complete if b is finite and both directions use BFS optimal if step costs are identical and both directions use BFS time and space: O(b^(d/2)