Anytime algorithm: Difference between revisions

Revision as of 03:05, 22 December 2007

Latest revision as of 05:18, 6 June 2025

(79 intermediate revisions by 58 users not shown)

Line 1:

{{short description|Algorithm that can return a valid solution to a problem even if interrupted}}

==Other Names==

In [[computer science]], an '''anytime algorithm''' is an [[algorithm]] that can return a valid solution to a [[Computational problem|problem]] even if it is interrupted before it ends. The algorithm is expected to find better and better solutions the longer it keeps running.

An anytime algorithm is also called an interruptible algorithm, however it is different from a contact algorithm because in a contact algorithm the time must be declared in advanced. In a anytime algorithm, a process can just announce that time is up <ref>Hendler</ref>.

Most algorithms run to completion: they provide a single answer after performing some fixed amount of computation. In some cases, however, the user may wish to terminate the algorithm prior to completion. The amount of computation required may be substantial, for example, and computational resources might need to be reallocated. Most algorithms either run to completion or they provide no useful solution information. Anytime algorithms, however, are able to return a partial answer, whose quality depends on the amount of computation they were able to perform. The answer generated by anytime algorithms is an approximation of the correct answer.

==Introduction==

Most programs run for a while and then gives the user the values that they expect within a few seconds. However, sometimes the program can be so complicated that they might take years to figure out completely and the user might have to shut down the computer before the answer is complete. What would happen if someone was to shut down the computer. In most cases, the computer would leave the issue unresolved and not do a thing. However, in the case of anytime algorithms, it would approximate the best answer and return a partial answer to the user. In other words, if the problem was completed, the program would return the 100 percent answer. Should it only be half finished it would return the 50 percent answer.

==Other Names==

== Goal of Anytime Algorithm ==

An anytime algorithm may be also called an "interruptible algorithm". They are different from contract algorithms, which must declare a time in advance; in an anytime algorithm, a process can just announce that it is terminating.<ref name="Hendler">{{cite book |editor-first=James A. |editor-last=Hendler |title=Artificial Intelligence Planning Systems: Proceedings of the First Conference (AIPS 92) |publisher=Elsevier |orig-date=1992 |date=2014 |isbn=978-0-08-049944-4 |url={{GBurl|tgujBQAAQBAJ|pg=PR5}}}}</ref>

The goal of anytime algorithms are to give intelligent systems the ability to make results of better quality in return for turn around time <ref>Zilberstein</ref>. They are also suppose to be flexible in time and resources <ref>Grass</ref>. They are important because artificial intelligence or AI algorithms can take a long time to complete results. This algorithm is designed to complete in a shorter amount of time <ref>Grass</ref>. Also, these are intended to have a better understanding that the system is dependent and restricted to its agents and how they work cooperatively <ref>Grass</ref>. An example the is Newton-Raphson iteration applied to finding the square root of a number b <ref>FOLDOC</ref>. Another example tat uses anytime algorithms is trajectory problems and your aiming for a target <ref>Grass</ref>.

== Goals ==

What makes anytime algorithms unique is their ability to return many possible outcomes for any given output <ref>Zilberstein</ref>. It uses many well defined quality measures to monitor progress in problem solving and distributing computing resources <ref>Zilberstein</ref>. It keeps searching for the best possible answer with the amount of time that it is given <ref>umich</ref> It may not run until completion and may improve the answer if it is allowed to run longer <ref>elook</ref>. This is often used for large decision set problems <ref>Horsch</ref>. This would generally not provide useful information unless it is allowed to finish <ref>Bender</ref>. While this may sound similar to dynamic programming, the difference is that it is fine tuned trough random adjustments, rather than sequential.

The goal of anytime algorithms are to give [[Hybrid intelligent system|intelligent systems]] the ability to make results of better quality in return for turn -around time .<ref name="Zilberstein">{{harvnb|Zilberstein|1996}}</ref>. They are also supposesupposed to be flexible in time and resources .<ref name="Grass">{{cite journal |first=J. |last=Grass |title= Reasoning about computational resource allocation|journal=XRDS: Crossroads, the ACM Magazine for Students |volume=3 |issue=1 |pages=16–20 |date=1996 |doi=10.1145/332148.332154 |s2cid=45448244 |doi-access=free }}</ref>. They are important because [[artificial intelligence]] or AI algorithms can take a long time to complete results. This algorithm is designed to complete in a shorter amount of time .<ref> name="Grass<"/ref>. Also, these are intended to have a better understanding that the system is dependent and restricted to its agents and how they work cooperatively .<ref> name="Grass<"/ref>. An example the is Newton-Raphsonthe [[Newton–Raphson]] iteration applied to finding the square root of a number b .<ref name="FOLDOC">[http://foldoc.org/anytime+algorithm anytime algorithm from Free Online Dictionary of Computing (FOLDOC)]</ref>. Another example tatthat uses anytime algorithms is trajectory problems andwhen youryou're aiming for a target; the object is moving through space while waiting for the algorithm to finish and even an approximate answer can significantly improve its accuracy if given early.<ref> name="Grass<"/ref>.

What makes anytime algorithms unique is their ability to return many possible outcomes for any given input.<ref name="Zilberstein"/> An anytime algorithm uses many well defined quality measures to monitor progress in [[problem solving]] and [[distributed computing]] resources.<ref name="Zilberstein"/> It keeps searching for the best possible answer with the amount of time that it is given.<ref name="umich">{{cite web|title=Anytime algorithms|url=http://ai.eecs.umich.edu/cogarch2/index.html|website=Cognitive architectures|publisher=University of Michigan Artificial Intelligence Laboratory|archiveurl=https://web.archive.org/web/20131213011435/http://ai.eecs.umich.edu/cogarch2/cap/anytime.plan|archivedate=13 December 2013}}</ref> It may not run until completion and may improve the answer if it is allowed to run longer.<ref name="elook">{{cite web|title=Anytime algorithm - Computing Reference|url=http://www.elook.org/computing/anytime-algorithm.htm|website=eLook.org|archiveurl=https://web.archive.org/web/20131212094200/http://www.elook.org/computing/anytime-algorithm.htm|archivedate=12 December 2013}}</ref>

Anytime algorithms are designed to be predictable <ref>Grass</ref>. Another goal is that someone can interrupt the process and the algorithm would give its most accurate result <ref>Grass</ref>. This is why it is called an interruptible algorithm. Another goal of anytime algorithms are to maintain the last result so as they are given more time, they can continue copulating a more accurate result <ref>Grass</ref>.

This is often used for large decision set problems.<ref name="Horsch"/> This would generally not provide useful information unless it is allowed to finish.<ref name="Bender">{{cite book |first=Edward A. |last=Bender |title=Mathematical Methods In Artificial Intelligence |publisher=Wiley |date=1996 |isbn=978-0-8186-7200-2 }}</ref> While this may sound similar to [[dynamic programming]], the difference is that it is fine-tuned through random adjustments, rather than sequential.

Anytime algorithms are designed toso bethat predictableit <ref>Grass</ref>.can Anotherbe goaltold isto thatstop someoneat canany interrupttime theand processwould andreturn the algorithmbest wouldresult giveit itshas mostfound accurate resultso far.<ref> name="Grass<"/ref>. This is why it is called an interruptible algorithm. Another goal ofCertain anytime algorithms are toalso maintain the last result, so asthat if they are given more time, they can continue copulatingfrom awhere morethey accurateleft resultoff to obtain an even better result.<ref> name="Grass<"/ref>.

==Constructing an Anytime Algorithm==

Make an algorithm with a parameter that influences running time. For example, as time increases, this variable also increases. After for a period of time, the search is stopped without having the goal met. This is similar to Jeopardy when the time runs out <<ref>Bender</ref>. The contestants have to represent what they believe is the closest answer, although they may not know it or come even close to figuring out what it could be. This is similar to an hour long test. Although the test questions are not in themselves limiting for time, the test must be completed within the hour. Likewise, the computer has to figure out how much time and resources to spend on each problem <ref>Bender</ref>.

==Working with Decision Treestrees ==

When the decider has to act, there must be some ambiguity. Also, there must be some idea about how to solve this ambiguity. This idea must be translatable to a state to action diagram .<ref name="Horsch">{{harvnb|Horsch|Poole|1998}}</ref>.

== Performance Profileprofile ==

The performance profile estimates the quality of the results based on the input and the amount of time that is allotted to the algorithm .<ref> name="Grass<"/ref>. The better the estimate, the sooner the result would be found .<ref> name="Grass<"/ref>. Some systems have a larger database that gives the probability that the output is the expected output.<ref name="Grass"/> It is important to note that one algorithm can have several performance profiles.<ref name="Teije">Grass{{cite conference |last1=Teije |first1=A.T. |last2=van Harmelen |first2=F. |title=Describing problem solving methods using anytime performance profiles |book-title=Proceedings of the 14th European Conference on Artificial Intelligence |publisher= |___location= |date=2000 |isbn= |pages=181–5 |url=https://core.ac.uk/download/pdf/43408018.pdf }}</ref>. Most of the time performance profiles are constructed using [[mathematical statistics]] using representative cases. For example, in the [[Travelling salesman problem|traveling salesman]] problem, the performance profile was generated using a user-defined special program to generate the necessary statistics .<ref> name="Hendler<"/ref>. In this example, the performance profile is the mapping of time to the expected results .<ref> name="Hendler<"/ref>. This quality can be measured in several ways:

*certainty: where probability of correctness determines quality <ref> name="Hendler<"/ref>

*accuracy: where error bound determines quality <ref> name="Hendler<"/ref>

*specificity: where the amount of particulars determine quality <ref> name="Hendler<"/ref>

==Algorithm prerequisites==

accuracy: where error bound determines quality <ref>Hendler</ref>

Initial behavior: While some algorithms start with immediate guesses, others take a more calculated approach and have a start up period before making any guesses .<ref> name="Teije<"/ref>

*Growth direction: How the quality of the program's changes"output" withor increasingresult, runtimevaries as a function of the amount of time ("run time")<ref> name="Teije<"/ref>

specificity: where the amount of particulars determine quality <ref>Hendler</ref>

*Growth rate: Amount of increase with each step. Does it change constantly, such as in a [[bubble sort]] or does it change unpredictably?

It is important to note that one algorithm can have several performance profiles <ref>Teije</ref>

*End condition: The amount of runtime needed <ref> name="Teije<"/ref>

==What must be determined before an Anytime Algorithm can Start==

Initial behavior: While some algorithms start with immediate guesses, others take a more calculated approach and have a start up period before making any guesses <ref>Teije</ref>

Growth direction: How the quality of the program changes with increasing runtime <ref>Teije</ref>

Growth rate: Amount of increase with each step. Does it change constantly, such as in a bubble sort or does it change unpredictably

End condition: The amount of runtime needed <ref>Teije</ref>

==References==

{{reflist}}

Anytime Algorithm http://foldoc.org/?anytime+algorithm

Anytime Algorithm http://tarono.wordpress.com/2007/03/20/anytime-algorithm

http://ai.eecs.umich.edu/cogarch2/cap/anytime.plan Anytime Algorithm

Bender, Edward A. ''Mathematical Methods In Artificial Intelligence'', IEEE Computer Society Pres, 1996

ELook http://www.elook.org/computing/anytime-algorithm.htm

Grass, Joshua. "Reasoning about Computational Resource Allocation." http://www.acm.org/crossroads/xrds3-1/racra.html

Hendler, James A., ''Artificial Intelligence Planning Systems'', 1992

Horsch, Michael C., Poole, David "An Anytime Algorithm for DecisionMaking under Uncertainty" http://www.cs.ubc.ca/spider/poole/papers/randaccref.pdf

Teije, Annette ten, Harmelen, Frank. "Describing Problem Solving Methods using Anytime Performance Profiles".

== Further reading ==

Zilberstein, Shlomo. "Using Anytime Algorithms in Intelligent Systems". http://anytime.cs.umass.edu/shlomo/papers/aimag96.pdf

{{refbegin}}

*{{cite conference |last1=Boddy |first1=M. |last2=Dean |first2=T. |title=Solving time-dependent planning problems |book-title=Proceedings of the 11th international joint conference on Artificial intelligence |volume=2 |date=1989 |isbn= |pages=979–984 |url=https://dl.acm.org/doi/abs/10.5555/1623891.1623912 |id=Brown University CS-89-03}}

* {{cite journal |last1=Grass |first1=J. |last2=Zilberstein |first2=S. |title=Anytime Algorithm Development Tools |journal=ACM SIGART Bulletin |volume=7 |issue=2 Special Issue on Anytime Algorithms and Deliberation Scheduling |pages= 20–27|date=1996 |doi=10.1145/242587.242592 |s2cid=7670055 |url=https://dl.acm.org/doi/abs/10.1145/242587.242592|url-access=subscription }}

* {{cite conference |arxiv=1301.7384 |last1=Horsch |first1=M.C. |last2=Poole |first2=D. |title=An anytime algorithm for decision making under uncertainty |book-title=Proceedings of the Fourteenth conference on Uncertainty in artificial intelligence |date=1998 |isbn=978-1-55860-555-8 |pages=246–255 |url=http://www.cs.ubc.ca/spider/poole/papers/randaccref.pdf}}

* {{cite tech report |first=E.J. |last=Horvitz |title=Reasoning about inference tradeoffs in a world of bounded resources |publisher=Medical Computer Science Group, Section on Medical Informatics, Stanford University |id=KSL-86-55 |date=March 1986 |url=}}

* {{cite journal |last1=Wallace |first1=R. |last2=Freuder |first2=E. |title=Anytime Algorithms for Constraint Satisfaction and SAT Problems |journal= ACM SIGART Bulletin |volume=7 |issue=2 |pages=7–10 |date=1995 |doi=10.1145/242587.242589 |s2cid=8250394 |doi-access=free }}

* {{cite thesis |first=S. |last=Zilberstein |title=Operational Rationality through Compilation of Anytime Algorithms |publisher=Computer Science Division, University of California at Berkeley |type=PhD |date=1993 |url=https://dl.acm.org/doi/abs/10.5555/193131 |id=UMX GAX94-08166}}

*{{cite journal |first=Shlomo |last=Zilberstein |title=Using Anytime Algorithms in Intelligent Systems |journal=AI Magazine |volume=17 |issue=3 |pages=73–83 |date=1996 |doi= |url=http://rbr.cs.umass.edu/shlomo/papers/Zaimag96.pdf}}

{{refend}}

== Goal of {{DEFAULTSORT:Anytime Algorithm ==}}

==Recommended Reading==

[[Category:Artificial intelligence engineering]]

http://www.acm.org/crossroads/xrds3-1/racra.html

[[Category:Search algorithms]]