You are on page 1of 17
a Doais 221002010) — Assiqnment — | GE Bhanowi YQulition TdeUligunt Ageibs se entities that poterive -lhive envisonmerd: rele duchiions. ee oe peurrtion , and ke action, to achioue Sprcific goals. Thue ert, may exhibit arctonomous behavior, heosning copablities | vadayeta bili . . Example in Healthcare! An dntetligent din dualthcare could be ‘diagnostic aystin that onalyrys sptliont “supmpitoms , medion) history, and Aut cusutt to pruide orerete iagonesss. This sypitin Pein selivant snfetmation, employs saroning do make ixformd decisions ond sucomnunds oppropiah duwatmurt ‘eyitions - Uniformed Search Algerithma: Uninforrud eaiich algorithms explore the seoxch apace woithout scomicening additional information about the problim othr than the vonnectiuity of the abodes. framplu, include Boucdth iret Search and! Depth -trst Seach. Tr erred beach i drformedt search algerithens , alo Known sax tewiitic seanch algorithins, ux demain. specific Kriowledge drewiistics to quide the search’ prows mote efficertly A® CA~ stax) dis 1 tommon example thed considers deth-the west do suach wo node wand an sstimate af ‘the vumaining cost tothe goal. DY dupsrmiad Machine ung'- Th supvudad duotning, the algort See ee eek inqutl howls covwnedld ia output dabel. the goal dy do dost the mopping "sbetwern inpulk and” - outputs , making poudicacions ot wlan ications on nw, unsen data- Uasu vind Machine feounin + Unsupervised tusining inveliss training on wnlabeld date-the walgorithin discomes attoins., sulofionshaps < ot sbutlwus in the dala without explict aquidamee. Common taxks includes. cluttering similar date pein ot seducing “thi, dimunsionlty 4 al deta: : 4 : . 4) Junction Approximations tunclion approximation en mackie Ler Mette to Ale proces af arlimating van winknowrs “ ‘ern ie ae mathematical function baud on va sel of input oulpue " ee a geal dx do shaun vo model that tan genvialigg well do me pudication on niw, uthien data ‘poink,. Applisaction:. Prucxion Hoxyming ‘Examples. Mand Data Selene au uu to wna lie 40ul conditio, rwrathor patton, and op dual do otlincye caucigartion , fodilization sand haveing sehhancing vop wyield and susrowiee ~affieiency , : ' Lou t- Application: baal Research. si ‘example t- A) systems con anal al datobaes t, preuick Lawton it, vulewant es wand gal docurunt., iding in Segal suxeaich vand decision’ ~making @) touvard chainin: i is Rele.- 4n jowsord el a othe seasoning Ppous stack “Hy teneusn socks wond wpe, to destivee ies cavialiboras « 6 Pesqusses trom the tnikial data ’ te suach goal. Backwand Chaining t. Rok: Th backward hoining , Yu wunsoning with oF qoo! and’ werk, backward to hebben ne “ashich Jac er surly ‘can dead to the achievement Of that goal: h jack procs stack + deruverd chain! skards with data ond moun tewand 2 goal, whi backward el aining shorty with og geal ond thorke backward do find the sobuert data. rtowward chaining Js often und when thoes ned Jor continuous infeunee , swhearuas dborkurand chalning de suitable for gout Saat clianing dently cand thondle mining or incooreel data, ( Aliosn ond incomsistineiss inthe dataset techniques include Prpudlation dor miming value , samouing cutlies, and odldrersing sdatta formed issues: byQData trommformation + Conuoct and ndumalixe dota do make it duitabl qo machine Leaning algsiittms. This may involve sealing featious encoding cotigorical variables, & duandfowming dota. distubutions Datta Reduction: Reduee the dimnionality of the dataset to anhonce computational elficieney and suduee neie Techniquss und include footw geleclion , dimensionally suduclion and aggregation @ Probability Appreximotily couuct Cenc) Liartning:- PAC Lertaing vis a theotdtical frame that focuses on the ability of a machine cleoserin ciithm to down a Vypethanis hat wu “probably approrimatly evict” band on a limited amount of trainin date. the geal is net ion but sather achieving a 1 thigh probability o| correc Concept, “Probability coroect”s “The hypothesis is oxcuratt On new, wren data . “Appruorimattely Corvurk”» the tuppotheris ivaldg and teow windeelying potion in the dete. Sieg eA, “ta —r PAC ing ackne, AB unend 3 iwi din swat uate dal ‘edlainily and wandomns rt provide a balan tlw the practical consiqialions of Umited SHMUEL + urtth ‘probability qh oo the need or aceumacy & the data & computational a chal nu t- . \ “sporantiad Gomplaity:-Backivacking may dina do om exponaaid No - of possibilitics vmaking at impractical fot curtain comple * Cop's | a Opeimigation Gipfiaully: % may shuggle with finding optimal tolutions, os al explous the search apace ehoustively without Considering hewestics. 2 i Memehy Usage 1 Backlyacking can consume 4.9 nil icant rremey , Aspecially sin ceeses whee a _ aeatch A pace nuds to be Drarwwural - : Benefits: Completeness: Garkbracking guanartus finding a solition df om xis, ag dh scomnplete Ltn for csp's. 4 Flexi ily - ge con trandle oo uvicle of constraint pes, ad sunt and bi vonstuaind, ing at vod . tase of iplenuntation :- Backtracking is concerituially simmp le and sony to implmint,, providing oO straightfousard approach bb soluing CoP's. 10) 44 Aimlaing : Aduartagess- — Simplicity y Hill climbing 4 ightlewsard ond say- to - icity i 4 a suaig ld {orusare tasy —rroeal optima: 4 works wsell in curtain senoiies whic tocal weptima wou voreeptable ot the goal dy do find seasonably geod eelution qpsekly. aed tek J Global optima: $y t struck vin local optima ond, dail to the gical epinud, b " thack of Backlracking : TM oul Lacktrack whieh ean dine ot ability to vxplore athendtive paths. sae, PMS Seanehy Ydanlages Aobal Splines Beals Ael eawely sh, weipinble of nding qlebol ating fe scondideing nullipte plolle simetlancordley, ' Mexibility BN sallow the incorporation of duvnritien do quidh the Seewreh baud er idomaiin speeci ie Kabila’ Quawobacks: Complesity: yplenwating rand: andecitariding verte dtvel sennch aN de pote comple i comparud to Hill clive bing: Nemoty auager- may seeqyiciste aignificarr| mermoty,espertall un Scenarios ath va dea deunch apace « / is Supenived heevening re Chareeclotstees labeled Qelay Trairiing data induces input-output pairs. Goal:-Predicl the outhul jor uw , angen impute . Examples Classificeclion rau segresion castes « Unww roxvdine heortniing Lhasecleristics :. Unlabeled ectar traning data Jacks seplicit output, abels. Goals Discover pattiuns sulectionshins , of aburctuvus inthe date - Beomples:. sclustoding 5 dinuniionality suduction + trfotaement hesuung Beat Tyrteaclions Agent Levins though inaction with an ent eniistonrnendt Goal heasun a poliey do maximine wumuldlive stewards. trom pls + Game playing ' stobolic tontuol OH CLs to eal duatning sulison dobeld data, while vuruupoudseal duanning choles weil 2 unlabeled deta —yReinfoicement dhorniing muolin an it irturoching awith on aavinonrnt and duaxning thuough fudkock. ay Challenges: Computational Loreplery + Approrimad infornee mithods fra be. computtationally aepenuive , spuially fer henge Boynion Nelworye “Accuracy ug. Spal Jrrade-off :- Balancing the track -off{ ddween 9 aceunacy cand computartional speed ds whallen ing : Handling Continuous Uariables:-Approvimett infounes drebniques m atrugale with Rayuian Melwotks containing continusus arial, mane Networks: Jot chime - gerin! dato, handling dumporal ependenctes odds sonipelxtly do approdincele inforance a + on Qesiston~ Making :- hulprtimal Qudaiond: Traduca in wappormat infucnee Lind de suboptimal ceciwton- maleing fone Rew wice Consumption + Compudettio expensive yrthods sequise sigrificant sources, impacting cal tine ducision pit Unesctairidy din “rele: Appuoimations intiodiuee euctatniby , influuncing the confidence in decision outcomes: ‘ Pouignmunt 9. Ay A canbe es Bpeton ‘Probl cosP) inmolues finding a soluibiny, a set of was bles, cach avith sel fnedd domains, uhill sotitye aqusified constrain’ . “the goo! va do wdwign value do vanioble ‘his sau thal muh all eotbainb. | vin Kalewance of Backtrac king - Backbacling din CSP spac kivacking is a syskmotic search valgoriclb mn auved-[ot solving cep! Mt oxplotes the search space sl sovalgning wali do variables 2 dhacktracks when it encourtes” wa violabion of conduoints « —rorktvacking is particnlady wwlwant im core wap to ih abil do efficiently Seplote pomibl casignmants & backtrack volun | rucewon, tiraatte Ly huoding do a vol solution + Ye) antdtigade A fe anu entities that porecive Whale enuixonmnt, rake ducisbrs sbelud on thet poraption, and dnki actions to chien specie goals. there ts vane axhibit sulenemous dehovier, sleosunins vapa bilities ond adaptabilit ‘ gxample: an Healthcanst: Arn sve in tuattheare would be o ubual fualth ouistard that faoniter, o potiedls tualth date - machine hesning xefuts tothe procs a ye} function apport mation in n of i ing won wnknewn funetion A Sra. wed. snnut output pabe The palig to iowa modi! that can 2 elie veel to wake wollte on rut , wraein data. point. Novy Proloobility theory in context of Al involues the mothumatical dearuverk modiling ond yuasoring sabout uncocainity « Ot quouidis 0 aa ety ond vou wnevetainily dn pre duision maling Cbs SA ae duuaatin search altho that efficiently find. oy optimal poth in a garth space by comb i a cont a te Foy ftom the stad node & a. huusivbic: eaimale of the vamataing YF cost to 44 aval: y Shp by hs. P stip Approach: : ; 'y Snthigadion rvatalig. son apn dist walt oh a yshln Ytualudlion: tualuct nods baud on a vost june "ee wohwu gin ds dle cert mn the doves hin? 44 tid! 4. trdimat. dp the goal , and 4¢n) as Hu total wtimaa’ i Stxpand: epand Ha node with the Jowat +(n) value cbt 2 Updak coils. Update the cont and tuurirtio walua jor reignboring * rods vind: % Pupuati- Contin the prow voll the goal viods da suaoclud® ot t efen Sut is emit. / . shine “er : — Cons a. guid inet Lun with o| AY evaluat: pila daaalaen axttea ered ste tt plistitinés| pata thot avi Leth co —elfectine ancl dead towards 2 the goal. 2) typlandion ;- ATs a teristic search algorithm that ‘cierttly dinds h down i optimal atts ino graph. Combirus the cot of the ‘thenct vode and wo twwirtic edimate of the sumaining cod cb eae : Vinitiabigation:- tritialize an openlist vi : pert dot with the atodt node. Atvaluobion:- tualucte nodu Lara on Abe tod funchion ln) safnbehen shou gin) ds the cot dntom the Hat, Wen) is the hewiabis, wAimat to tu goal, & ln) ds the dolal ecimatd cod % tepand> Sepand the node with the lowed 4in) value. 4) Updatt cons t- Update the cout & hewisbic volun from ruighboung 2Y seuat + Conlinis the proces unclil the geal node wh steaclud 0% the — oren lint & empty. & the retticed hearing +. (red data fo sbeaining aims to poudied ovlput sbaved or, it feotwus rand inuolues selassification ott sagrasion clash, « ale ‘ suporwdel Tented 1. Dees umlabeled data fer thawing aims do discover pattirns ot saledtionsbips avithin the tote, and includes clustering cond cinenvionality vecuction dass: : Runfotcemed heaning - rola san agent dasnin i q thou Nh intaction wilh an enuisonm, “eeciuing deealback in the form fl suwords , and aims to clein 0 policy fos ceeision making . ’ : Diifkewuneess. Aupeutisel tauring oulis on labeled date, while sursupoucud lrasininng dak with unlabelid date . tet Ruinfoteement hawning inwolus an cf irtluacting wth on omsenmunt & duosining Aheough Qudback. dd date clearing + Sdunti|y cond hanclle mining outline, and inconttencies in the dataset - ; — Techniques include imputation for viwiing vals jsemoviing outlion, Ing data Jormat issues. ; ] oy Data Trans crmoition:. 1 > Gonuecl & normalize data to make ut guitable for machine deotning algercther, . aeES Fr Trvclus stoling deatious , encoding categorical sandal dronsfedring a ishituling, UPN waniable, and syQatta oe . 7 Reduce the imensionality al -the wlatasel to enhance computational elficiney cand secuce wa P ‘ —yTechwiguts Snclude qurtin. cabion, dimensionality suductinn angpaqation: . 98) sorusad chain’ yy” ®) Forward chainin “Anil actes seasoning ie ynown fost & Peoys Sunt ee ~ Suitable a genesios white corctinusus monitoting & inf Ate seeqyu em in export syitins siagnotis cand supstirs cotth ‘ Sree Chaining : / . rTritsts stensoning ‘ion a gor! & works backward -to find feck \ BC seurley Suitable Jor goal- driun ganoaies, probline ~soluing ond planning. ~ Common ul im Seenorios who the jocus dis om vachieuen suing “ped: squcll io objeitines. Diwtorctio Metso: cele tard wilh data’ vond mows, owed sa. gels awhile backawand den chaining starts with a goal & works backusard do find seluant dota: = chair an vis suitable Jor tinfowne, awhile backward chi Se os maall seasenin Penfouinte § eunarios Lt forward chaining a Cortinusus moniterin inference we cue ial, New injetmation i fennel dad oh tn Use Rackword Ne sloous us On ndliating ai soli toluin 4 problem, , Goale ow well defined, 2 he aapstern need, to rhe hac kusarel to suach thm: 3B) finery t Seotivuss, Modes and Ares brdikien ou sepeutad we nodes & sulabionsbrips bower otitis au supousentiid Ontey « pebub utes y Mods con have othuibulis & xcs ton howe labels tp _ dugalbe the natn of seletionships . es: emer duclial Abuction: ‘ t pe nebworks often exhibit a hiewarehio! Aucune , allowing ee seprountation ef complex sielotionships. usbuatioe Bxompler Comidin 0 semantic edith suparerting Ao wled valor animal, / Node Lon, thrhart, Herbivore sCovtnivehe Ayes a, Seacts” . / . / fas gemartic ver nelusetk \tauolly ste 4 the slat! : bles ick tix sifting hh ein deena Ay Aiqnificonees : . Dealing swith Uncertainty» Pobabiliuti . ood veenteuy fy Pra to di a rie sutcomes - ‘ Decsion- Making, Under tnewrdainity Tn Al opplicstbions, decisions e\taon nud, to be made in situations whee eubtorus ase cwetain- Putiatiae trample: Consider a mudical diagnosis Corctinuous Uaniabl- Handling Baysion Networks with ori, + Metable srcbeocluces complexity « a : 8 Dumamie B ion Networks: ter Hime - gescies data. , dealin ye _ wth lempotal dependencies adds on additional layer of 1 7” Complexity. ae ‘Dedinion - Making \ _ Suborti Decisions. Trace racist ‘ § ference dad "n aubertimna) daciuton make Preston . . Reseweer Consumption + Computa tin eK ‘ thods : Pere met Ce Sarde susewiens,, tl cual- Hine, dusistodd ; Uneesttalnitiy ain Rolls Kimmotlons tibstodliree uncertain! : ‘fleneing the conju wl suclcomes . 4 Assianmunt— 2. Unifotenel beach picts: Alo known as" blind earch walgorthms - Ty Exploe He search space usithout lusing any additional knowl about the probum. 1 wre , e Trfcroed Search’ Algetithmns Alo Known as Yhewtishic seach salgotittims. ! de — Utilize domain -snecilic Knew! ot henwristics to quicle the search mete a cat ge j . >® Consbuaint Jattislaction Poblims (ese) > Truelue finding 2 solution do a set of variables each with dsfined domains, ashile satisfying spetifiad contrasts. —7 Goal u 40 auign yaluss to variables in a wey that mec Ul ecorutiaint . . Tmportance of Backtrockin rAyskmatic Search velit Exploration : : — Completencss Flexibility 4° Yc) Provide a Faaltumatiead feanuwork for modeling wand asoning wobout uncectoinity i / oy Ka components” Melw probability dlistitbidions .B Edt ‘ cand Yeondiltonal probability. Bayesian infucence oy dupweutised Machine Kasning 1 a ° _ uses dabeld traiving data, usher woch input has’ a a dodllg outhut Jeuel- oe Oe ve do dian “the mapping lo inputs & outpads to make qrudictions en su junseen data. Unsupowised Machine Leduning: Lo ves unlabeled training data’, whore eoch ¢ hos a te the algeiithm discomes patties within tte data. > Goal ds to dind hidden sbuctiows ot groupings sin the data Se Ny Toward Chaining: SAL avely woth *Suilatle Vad gy opel spelen R uliannoyliz 4 yf Prachevrared! chaining + \ ~YAlole with a qokl & aucrle backwared lo {ind fort. \ ~YAutlable Jos geal: diver senator, problem sobung ond plannisy, Known jack won Proqusus foworl a qool, for covdinucus montbting #R infouines Zener, , 06d shen the Joes ds ov sathiewinng opuell in Oh fortores I Advanta mets Mewar! chainin ub good for dunamic’ olla enuivonmuerily, —¥ Backward thainin, ds olteeti 1 uwell~ deified al Suentid cake | * i “ n “ . ¥ BB) mY se a duibiutio: gemrich salgptitton el efietently find. epbimal paths in a anh , , Combinus athe cost ol the path from tn stad nook 2 4 huwiatic, artimade of the viemaing cork to the goal. , BLP - by step Approach: YTnitialination t Trtialize & open did with atost node. . a tualuction 1 fualuak nodu boxed on dhe cost junction € . sine glndahtn) niFapsey Srexpand:-tepane| the node with aha owed tn) wabtee . 4) Updel costs on the cot & Mewuibio malig for nuighlooting S)Repecd + contin unl the apo! rodt us suaishid om the ofan dit aby emily: : ~ teomplrs Comics o guid: baud pathfinding problim with obstacles A* eval poths aed on actual veost A, oe huwuytic « tt qmovitises path, thet wu both cert > effective A dab foward the goad - pean Me hay frcloalislity freddy corvuct (pac) : ital Jromavedk focusing en: a machine eng siping tabiltly fo awn a tet thot as “probably a covuel” dad on limid tyoining data. Balancs blw the wed for accwiaey & praca! convent, Concept: ih “Probability cerect” + the Vypothesis ds wotewial with high Probability en nw, unien data. = “Approriqnatly covuct™. The ong patton tn te data: Troplications » ee she wnenihatniy. & sinners im sual wertd dato Wapotheris Js clo enough 46 the tru 2) Or Ai ilicanee+ wy ond unestainity by arenes probabilitin, te diffeunt putcoms: — Exsential fer decision - making w under ‘wdoin conditions. piludy otive | txonipl' ° Medical sdinyroits systems Atnaviot- Diagnesing ma pobient, Wns. Probabilistic Com! marks Aw probabilit os to diflseunt discos bard on e wt “A Pole: the siyitim ealeulabs the tthalihood of tock bseare quer He shsoued een “ue 4 a mow ped EEN Agricul: Poucstion faxmin €, Applications ar anclynyes sel ‘conditions y weather patton ae Gop lualth data ° Example dptimiging inv ction 1 feck tigation zg Sfistiustiny eL. ‘ frentoud Hop wie ™ ‘ak Spliattion A I poweud Systins onaly al dottabares . ye teamples Paniling with subvert cos dow , Preadint & Ugal documerch for efficterdt at pica Medicine: bin! Disco f Application :..av ys molculax stuetivus ter hug clevelo trample + Awiting in stool detection & eased Llp By By MUL climbing - abun th: vicki + tay ‘to eal & implrunt. > bowl on Imo Can fi wl optma icientt Limitations ah a. —* Global optima Moy at stuck iw local iin ming Mae qual opttinuam . . Bet - Host Search. Aevenathy Alobal otras Capable af {i lokel opting oo theiblty Allows ei i" hase jer dSmaine aruer|ic tnowlida® . Limitations .- Compl i tye Mote complex a Ao Hull clinels Moy ~ Us - sequive ai ‘cant memoly , Ti lotag™ femelle spaces: m ely Aypate chaning :- ces gaudy \ hondle musing oi incovurl data , outlive R apconsi slenciss « Nechniquuss tentuu siatting ytneescing categorical vali 2 transfotmin data slandarixation N data transformation i Toeass Conver & notmaline data ler eed iden learning stasis . Jechniqus Feative sealing , encoding tal eqosienl vawiabls & Avensfotming delo dibdbubiors. de) Gupeuaed Yecvuning V => Labeled Training Dotor- tach input has a wowsponding pudtut label: —¥ Gool> heawn the mapping ble tnpacl, & output feu peudiction + Drsupenused leaning 1. — Unlabelid roning Detar Ko explicit output Labels. =) Goal > Discouse pattuuns, vulationghips ot ébuctus in “lee dota - Reinforcement \ansuniing + y Agent - Enuionnunt Titaction.. Leasining Ahorough anteater with on envivoruvunt . ; > Goals awn o policy slo maximine currulotine, vexuands. ium, suposed dranning vu kiss on labeled data, whil unsupeuised sng duals with” unlabeled dota: . . YR nforermant hovering wolus on aqunt twelorareting with an envivionmurt & Leownshg Atorough specdback op, ba

You might also like