Informed Search and Applications

1
Informed Search and Applications

• Reading Recommended paper
• Focused Crawling a new approach to
  topic-specific Web resource discover,
• Soumen Chakrabarti, Martin van den Berg, Byron
  Dom
• http//www.cs.berkeley.edu/soumen/doc/www1999f/pd
  f/www1999f.pdf
• Next class chapter 5

2
Homework Notes

• What is a bigram frequency?
• Bigram letter pair (a,b)
• Bigram list has probabilities for each letter
  pair. Ratio of frequency of that pair compared to
  all possible combinations
• How many random initial states have solutions?
  Not very many. Just let your program run until
  you get between 10-20 and base your averages on
  that.

3
Local Search Algorithms

• Operate using a single current state
• Move only to neighbors of the state
• Paths followed by search are not retained
• Iterative improvement
• Keep a single current state and try to improve it

4
Advantages to local search

• Use very little memory usually a constant
  amount
• Can often find reasonable solutions in large or
  infinite state spaces (e.g., continuous)
• Unsuitable for systematic search
• Useful for pure optimatization problems
• Find the best state according to an objective
  function
• Traveling salesman

5
(No Transcript)
6
(No Transcript)
7
Steepest Ascent
8
(No Transcript)
9
Problems for hill climbing

• When the higher the heuristic function the
  better maxima (objective fns) when the lower
  the function the better minima (cost fns)
• Local maxima A local maximum is a peak that is
  higher than each of its neighboring states, but
  lower than the global maximum
• Ridges a sequence of local maxima
• Plateaux an area of the state space landscape
  where the evaluation function is flat

10
Some solutions

• Stochastic hill-climbing
• Chose at random from among the uphill moves
• First-choice hill climbing
• Generates successors randomly until one is
  generated that is better than current state
• Random-restart hill climbing
• Keep restarting from randomly generated initial
  states, stopping when goal is found
• Simulated annealing
• Generate a random move. Accept if improvement.
  Otherwise accept with continually decreasing
  probability.
• Local beam search
• Keep track of k states rather than just 1

11
Online Search

• Agent operates by interleaving computation and
  action
• No time for thinking
• The agent only knows
• Actions (s)
• The step-cost function c(s,a,s)
• Goal-test (s)
• Cannot access the successors of a state without
  trying all actions

12
Assumptions

• Agent recognizes a state it has seen before
• Actions are deterministic
• Admissable heuristics
• Competitive ratio Compare cost that agent
  actually travels with cost of the actual shortest
  path

13
What properties of search are desirable?

• Will A work?
• Expand nodes in a local order
• Depth first
• Variant of greedy search
• Difference from offline search
• Agent must physically backtrack
• Record states to which agent can backtrack and
  has not yet explored

14
Online DFS - setup

• Inputs s, a percept that identifies the current
  state
• Static
• result, a table indexed by action and state,
  initially empty
• unexplored a table that lists, for each visited
  state, the actions not yet tried
• unbacktracked a table that lists, for each
  visited state, the backtracks not yet tried
• s,a the previous state and action, initially null

15
Online DFS the algorithm

• If Goal-test(s) then return stop
• If s is a new state then unexploreds ?
  actions(s)
• If s is not null then do
• resulta,s?s,resultreverse-a,ss
• Add s to the front of unbacktrackeds
• If unexploreds is empty
• If unbacktrackeds is empty t hen return stop
• Else a? action b such that resultb,spop(unback
  trackeds)
• Else a?pop(unexploreds)
• s? s
• Return a

16
Back to Online Search

• Would hill-climbing be appropriate?

17
Learning Real Time A (LRTA)

• Augment hill-climbing with memory
• Store current best estimate of cost from node to
  goal H(s)
• Initially, H(s) h(s)
• Update H(s) through experience
• Estimated cost to reach the goal through neighbor
  s
• H(s) c(s,a,s) H(s)

18
Real-World Application for AI Search

• The Problem
• World Wide Web is a vast resource
• Google reports indexing 3.3 billion web pages as
  of 1/2004
• About 600GB of text changes every month
• A search engine requires continuous crawls to
  index all pages
• Inktomi
• Cluster of 100s of Sun Sparc stations
• 75GB RAM each
• 1 TB disk
• Crawls greater than 10 million pages/day

19
Alternative focused crawl

• Start with a user specified topic hierarchy (like
  Yahoo, but user specific)
• Used for training the crawler to distinguish
  relevant pages from irrelevant pages
• Classifier a learned function that predicts
  relevance
• Resource discovery Starting from a node in the
  hierarchy the crawler branches out to find other
  relevant pages
• Must determine which outgoing links are good ones
  
• Simultaneously, the crawler runs a topic
  distillation algorithm to identify hubs
• Pages with large numbers of links to relevant
  documents

20
Why is this an AI Search Problem?

• What is the search space?
• What is the goal test?
• What is the heuristic function?

21
System Architecture

• Classifier makes relevance judgments on pages
  crawled to decide on link expansion
• Distiller determines a measure of centrality of
  crawled pages to determine visit priorities
• Crawler dynamically reconfigurable priority
  controls governed by classifier and distiller

22
Distillation

• A good strategy for the crawler is to identify
  hubs pages that are almost exclusively
  collection of links to authoritative resources
  that are relevant to the topic
• Authorities a web page with many incoming links,
  particularly from high-prestige, relevant pages

23
Identifying hubs and authorities

• Each node has two scores, iteratively determined
• a(v) number of incoming edges from relevant
  nodes
• h(v) number of outgoing edge to relevant nodes
• Weight these scores by the relevance scores of
  the pages they point to (a probability between 0
  and 1)