Week 12- Lecture Notes Topics ‘ dynamic Frog ramnuning - memoizetion avd subproblems - Fivonaeei ] examples - Svrortest paths . questing and DAG views Computational Complexity Dynamite Frog ramming (vp) * Big idea, herd, yet simple. * Powerful algorithmic design technique . Lorge class of seeminglyvexponential problems have @ poly nomial soléHon (“only”) vie dP. Farticularty Foroptimizotion problems (minlmen) - Enomple: Shortest poths. A dynamic Programming i¢ o consoled brute- fore methool. Tt uses recursion end re-use, D6 “controlled. brute- force” DP a “recursion and re-use”Fibonacci Alumbers Filoonacei numbers ave of the Porm 0. 6. 8, -Me, : hes F-5 Goal: Compute Fy Noive Algori ton follows vecursive definition. fib (mn). iP ngo return $2) 2 else veturn 4: fibln) +4ibln-2) => Thode TLnei) + Tla-a) A Olt) » A =e" % 2Tln-2) oll) y gre Exponential BAS! nN / Ga) /\ J Gq) O& aMemoized bd? Algorithm memo = {3 fiw): z iF nm is in memo: return memoln] 3. else: if neQ: fel 4. else fe Holn-D stibln-2) memo LnJs £ ‘. teturn + * $0 Lk) only yecurses Pest time called VK = onlynveememvized cells : Keh2,c0n e memoired cells free ( OliSime) * OW Hime per col Wgworing yeeursion) Polynomial - GOop\ © DP x “recursion Smemoizetion” © memoize. (vemember) and ve-use solutions to Subproblems +net help solve problem — in Fibonacci , subprovlems are fy Fay---1 Fr = ptime + # subproblems » Lime per Subproblem) +Fionecei + # subproblems =n. Hime pey subproblems Ol) a Hime = O(n) Uignoring vecursions )Bottom-up dP Algoritam ' Hbe= Tf 2% for K im Lutyens 3 if K aes gel Bln) 4, else: Pe filo [ee] + fib le-27 | OL. 5 site) + F 6. Teturn fib tw) senactly the seme, computation ab memoized bP (recursion “unrotied ‘) cecumsion va ein general : tepelogical sort of sub problem dependency bag. . proctically jester: no secur sion : anely sis more olovious * Con save space: last afin > Ot)Shortest Paths + Recursive formulation Stun) emnin {uotv) + 5L6ud | tow) €€ J + Memoized DP algorithm : toked infinite Hme if cycles. (necessary to handle negative cycles) + Works for directed aeyelic glse) aphs iv ose) | ~(effectively v°5/ 5ts.09 topological gost + Gellman 1 Sls.) ford rolled into single recursion) $v) $lss) + Suloprotolem dependency Mould ve slo a Oeyclic « -more subproblems temove eyelie dependence SulSveNgvortest Sv path using & & edges ~ tecurrente? Selb) + min {Sur S0(S¥) = 0 for SHV | bose _ Sy (53) + 0 for amy " / aeeaee - goal » BU) = Sieg lv) iP no negative cyple e -—Mewoi ze ~time : # Subproblems - [¥ime per sulpzoblems) wiwt olv) 6 olv’) - actually 8 Uindeqree lw) jor Syl) =D>tme OVE indegree(a)) = OLVE) [4.u) avotuad | Ou) € J BEUMAN FoRS |Guessing How to design yecurrence ewart shortest Sov peth es whet is the last edge in path? den't Know s guess tt is (uv) path is shortest su poth + edge (uv) eee eee by opHmel substructure scott 18s isu) + uatuev) Ser Onother sulproblem, © to Find best quess, try all Cl choices) ond use best. o*key: small (pelymomial ) # possible quesses per Sub problem — typieally this dominates time] subproblem. * DP zx tecdRStOWeb memoi zation +quessing DAG view ~\ike replicating POs . graph to vepresent Hme ~ converting shortest He 2 peths in graph to shortest pots in BAG 7 "Dee shortest paths in some DAGSuen ary + DP & careful orute fore = quessing 4 vecursion + memoi cation = dividing into teasonolle y# suloproblems whose Solution relate - acyoliely - usually via questing ports of solution ctimes de suloproblems x [time per suproblem) —_ treating recursive calls os oli) Cusuelly wainly guessing ) - eosentially On amortization - Count each suloproblem only once ; ofter Pivst timey/costs OL) vie memoizetion “DP & Shortest pathy in some DAG.5 aby steps to Dynomie Geograrmming define subproblems count 4 subproblems queee (pert of solytion) count + choices telate suloproblem selutions compute time per subproblem feturse + memoize problems Hme + lime per suloprobele) o X® suoprovlems uild DP table betom-up cheek subproblems acyclic |+epologicel order. Solve original problem: => extva time = 0 Subproblem oR by wunking oubproblem solutions. Enomples Fitoona cet Shortest paths Sulopro blems on &, (ov) for VEN, 0 KE IVI Jordhe KE vi 2 ming sv poth using £ Kedges #subproblems ” v* guess viothing edge Into Lif ony) ¥ choice® L indleq reel) +1 Teeurrence Fe Feat Fea Bld) = vain V5 re Us,u) + wl? He per Sulproblem Bw Coe Fl i. ols indegree(™) logical ee for Keli 9 for Keo, Wie) for vEV fotel Hme etm) oe ™ . +OlV) unless efficient ebout original ee ee prdolem Fn Sy to9) dor vev entra ime 6) pv)C omp utational Compleaity definitions: Ps {problems solvable in (nt) time J (polynomial) EXP: { problems soluable in oo") time } (eaponentia) ak . { problems solvable in Pinite tiene 3 “recursive” lee SS ? ‘onal difficulty ee exe we & uncomputalole /) P g EFS R undeadeble Examples: negative - weight cycles detection EP hess EERE wut ¢P whe wins fram given load canFiguration | Tetris ¢ EXP but don't Know whether € P survive given pieces Prom given tooard. Given a computer program does it ever halt Lstop) ? * uncomputable ( ¢R) + no algoritim solver it (correctly in Binite me on all inputs) * decision problem: answer is YES or NOMost Decision foblems are Uncomputable + program x binary string x nonnegative integer EA) s decision problem = @ function Prom binary strings (= non- veg. intergers) to [YESin, No lor} oz inbnite sequewe of bits = real number ER WI not nearly enough 78g yams fos all problems * ea0n program solves only one problem * => almost all problems cannot be solved NP NP: { Decision problems selveble in pelynowriel Hime via & \ucky alqoxttam} “The lucky algorithm con make . ” * e \ueky gueses, @lwoys right without trying oll options * nondeterministic model + a\qori thn makes gueses ond then says YES or NO * guesses queranteed to lead to YES outcome iP possible Example: Tetris ENP * nondetermini she algoritowm t quess each move, did 1 survive 1 + prooP oP Yes: Wot what moves to make (rules of Tetris are cosy )AP NPs {decision problems with solubens Hat con be checked” iw polynomial, Hime} => when answer is YES, it con be proved, and polyromial -time algorithm con check proof. —_— computation 7. diffi culty ne Ve ae ee, Ungomputable / undedidatole R P#AP —— Ikie a big Genyecbure (worth 41,000,000) + zconnot enginee? luck . £ generating (preots of) solutions con be herder then checking themHadness and completeness Claim: Tf PNP, then Tetris ENP P Pro? * Tetris is AP- hard = “es hard as” every problem ent Injoct . Tetris is AIP complete = NP i [NP-herd) NP- Ward a : Ex?- hard i _ Chess AP : ss Exe uncomputolde/ ba we undecidalole * Chess is CAP-tomplete = EXPN ExP-herd- ExP- hard is as hard as every protlem in GaP. TP NP # EXP, then Chess ¢ Ext\ne. Whether NP EXP ib also an open problem but less” Jomous /™ impo:Reductions Convert the problem into @ problem that i> already known Wow to solve | instead of solving Prom serateh) «© most Common algoritnun design technique + unweighted -shortest path > wei ghied (set weights) + min product path » shortest path [take 1095) * longest path + shortes path (megotive celea * shortest order four shortest path (x copies of the graph) + Cheapest leaky-tank path > Shortest path Pee 4 q (gre we teduchon) All of the aloove ave One-calli¥eductions: A problem ANB problem -+ Gsolubion — A solution Multcall seductionie: soove A using Free calls +> B, “in this "sense,every algorithwn reduces problem > modelo? computation ”N¢- Complete Problems NP- Complete problems ave ell interreducible usin Polynomial Hime teduchons (some difficulty) We can use reductions to prove NP-hardness Tetris. Ertamples of NP- Complete Peovolems « Knapsack ¢ 3- partition : given w integers, divide them into triples of equal sum? ° Travelling Selesmen Problem + ~ shortest path qhat visits all vertices fo oven, reyple is KL weight gal Pecision version) \ongest common “subsequence of Kk strings Mine sweeper, Soduku and most puzzles SAT : given @ Boolean jormula (end,or, not), is it ever true? ¢ Shortest paths amidst obstacles in SB * 3 coloring a given graph «Find largest clique in@ given q7eph-