18 IMITATION LEARNING

Size: px
Start display at page:

Download "18 IMITATION LEARNING"

Transcription

1 18 IMITATION LEARNING Programming is a skill bes acquired by pracice and example raher han from books. Alan Turing So far, we have largely considered machine learning problems in which he goal of he learning algorihm is o make a single predicion. In many real world problems, however, an algorihm mus make a sequence of decisions, wih he world possibly changing during ha sequence. Such problems are ofen called sequenial decision making problems. A sraighforward example which will be he running example for his chaper is ha of selfdriving cars. We wan o rain a machine learning algorihm o drive a car. Bu driving a car is no a single predicion: i s a sequence of predicions over ime. And as he machine is driving he car, he world around i is changing, ofen based on is own behavior. This creaes complicaed feedback loops, and one of he major challenges we will face is how o deal wih hese feedback loops. To make his discussion more concree, le s consider he case of a self-driving car. And le s consider a very simplisic car, in which he only decision ha has o be made is how o seer, and ha s beween one of hree opions: {lef, righ, none}. In he imiaion learning seing, we assume ha we have access o an exper or oracle ha already knows how o drive. We wan o wach he exper driving, and learn o imiae heir behavior. Hence: imiaion learning (someimes called learning by demonsraion or programming by example, in he sense ha programs are learned, and no implemened). A each poin in ime = 1... T, he car recieves sensor informaion x (for insance, a camera phoo ahead of he car, or radar readings). I hen has o ake an acion, a ; in he case of he car, his is one of he hree available seering acions. The car hen suffers some loss l ; his migh be zero in he case ha i s driving well, or large in he case ha i crashes. The world hen changes, moves o ime sep + 1, sensor readings x +1 are observed, acion a +1 is aken, loss l +1 is suffered, and he process coninues. The goal is o learn a funcion f ha maps from sensor readings x o acions. Because of close connecions o he field of reinforcemen learning, his funcion is ypically called a policy. The measure of Learning Objecives: Be able o formulae imiaion learning problems. Undersand he failure cases of simple classificaion approaches o imiaion learning. Implemen soluions o hose problems based on eiher classificaion or daase aggregaion. Relae srucured predicion and imiaion learning. Dependencies:

2 imiaion learning 213 success of a policy is: if we were o run his policy, how much oal loss would be suffered. In paricular, suppose ha he rajecory (denoed τ) of observaion/acion/reward riples encounered by your policy is: τ = x 1, a }{{} 1, l 1, x 2, a 2, l }{{} 2,..., x T, a T, l }{{} T (18.1) = f (x 1 ) = f (x 2 ) = f (x T ) The losses l depend implicily on he sae of he world and he acions of he policy. The goal of f is o minimize he expeced loss E τ f =1 T l, where he expecaion is aken over all randomness in he world, and he sequence of acions aken is according o f Imiaion Learning by Classificaion We will begin wih a sraighforward, bu brile, approach o imiaion learning. We assume access o a se of raining rajecories aken by an exper. For example, consider a self-driving car, like ha in Figure A single rajecory τ consiss of a sequence of observaions (wha is seen from he car s sensors) and a sequence of acions (wha acion did he expec ake a ha poin in ime). The idea in imiaion learning by classificaion is o learn a classifier ha aemps o mimic he exper s acion based on he observaions a ha ime. In paricular, we have τ 1, τ 2,..., τ N. Each of he N rajecories comprises a sequence of T-many observaion/acion/loss riples, where he acion is he acion aken by he exper. T, he lengh of he rajecory is ypically called he ime horizon (or jus horizon). For insance, we may ask an exper human driver o drive N = 20 differen roues, and record he observaions and acions ha driver saw and ook during hose roues. These are our raining rajecories. We assume for simpliciy ha each of hese rajecories is of fixed lengh T, hough his is mosly for noaional convenience. The mos sraighforward hing we can do is conver his exper daa ino a big muliclass classificaion problem. We ake our favorie muliclass classificaion algorihm, and use i o learn a mapping from x o a. The daa on which i is rained is he se of all observaion/acion pairs visied during any of he N rajecories. In oal, his would be NT examples. This approach is summarized in Algorihm 18.1 for raining and Algorihm 18.1 for predicion. How well does his approach work? The firs quesion is: how good is he exper? If we learn o mimic an exper, bu he exper is no good, we lose. In general, i also seems unrealisic o believe his algorihm should be able o improve on he exper. Similarly, if our muliclass classificaion algorihm A is crummy, we also lose. So par of he quesion how well does 1 I s compleely okay for f o look a more han jus x when making predicions; for insance, i migh wan o look a x 1, or a 1 and a 2. As long as i only references informaion from he pas, his is fine. For noaional simpliciy, we will assume ha all of his relevan informaion is summarized in x. exper Figure 18.1: A single exper rajecory in a self-driving car.

3 214 a course in machine learning Algorihm 43 SupervisedImiaionTrain(A, τ 1, τ 2,..., τ N ) 1: D (x, a) : n, (x, a, l) τ n // collec all observaion/acion pairs 2: reurn A(D) // rain muliclass classifier on D Algorihm 44 SupervisedImiaionTes( f ) 1: for = 1... T do 2: x curren observaion 3: a f (x ) // ask policy o choose an acion 4: ake acion a 5: l observe insananeous loss 6: end for 7: reurn T =1 l // reurn oal loss his work is he more basic quesion of: wha are we even rying o measure? There is a nice heorem 2 ha gives an upper bound on he loss 2 Ross e al suffered by he SupervisedIL algorihm (Algorihm 18.1) as a funcion of (a) he qualiy of he exper, and (b) he error rae of he learned classifier. To be clear, we need o disinguish beween he loss of he policy when run for T seps o form a full rajecory, and he error rae of he learned classifier, which is jus i s average muliclass classificaion error. The heorem saes, roughly, ha he loss of he learned policy is a mos he loss of he exper plus T 2 imes he error rae of he classifier. Theorem 18 (Loss of SupervisedIL). Suppose ha one runs Algorihm 18.1 using a muliclass classifier ha opimizes he 0-1 loss (or an upperbound hereof). Le ɛ be he error rae of he underlying classifier (in expecaion) and assume ha all insananeous losses are in he range [0, l (max). Le f be he learned policy; hen: E τ f l }{{} loss of learned policy E τ exper l } {{ } loss of exper +l (max) T 2 ɛ (18.2) Inuiively, his bound on he loss is abou a facor of T away from wha we migh hope for. In paricular, he muliclass classifier makes errors on an ɛ fracion of i s acions, measured by zero/one loss. In he wors case, his will lead o a loss of l (max) ɛ for a single sep. Summing all hese errors over he enire rajecory would lead o a loss on he order of l (max) Tɛ, which is a facor T beer han his heorem provides. A naural quesion (addressed in he nex secion) is wheher his is analysis is igh. A relaed quesion (addressed in he secion afer ha) is wheher we can do beer. Before geing here, hough, i s worh highlighing ha an exra facor of T is really

4 imiaion learning 215 bad. I can cause even very small muliclass error raes o blow up; in paricular, if ɛ 1/T, we lose, and T can be in he hundreds or more Failure Analysis The bigges single issue wih he supervised learning approach o imiaion learning is ha i canno learn o recover from failures. Tha is: i has only been rained based on exper rajecories. This means ha he only raining daa i has seen is ha of an exper driver. If i ever veers from ha sae disribuion, i may have no idea how o recover. As a concree example, perhaps he exper driver never ever ges hemselves ino a sae where hey are direcly facing a wall. Moreover, he exper driver probably ends o drive forward more han backward. If he imperfec learner manages o make a few errors and ge suck nex o a wall, i s likely o resor o he general drive forward rule and say here forever. This is he problem of compounding error; and yes, i does happen in pracice. I urns ou ha i s possible o consruc an imiaion learning problem on which he T 2 compounding error is unavoidable. Consider he following somewha arificial problem. A ime = 1 you re shown a picure of eiher a zero or a one. You have wo possible acions: press a buon marked zero or press a buon marked one. The correc hing o do a = 1 is o press he buon ha corresponds o he image you ve been shown. Pressing he correc buon leads o l 1 = 0; he incorrec leads o l 1 = 1. Now, a ime = 2 you are shown anoher image, again of a zero or one. The correc hing o do in his ime sep is he xor of (a) he number wrien on he picure you see righ now, and (b) he correc answer from he previous ime sep. This holds in general for > 1. There are wo imporan hings abou his consrucion. The firs is ha he exper can easily ge zero loss. The second is ha once he learned policy makes a single misake, his can cause i o make all fuure decisions incorrecly. (A leas unil i luckily makes anoher misake o ge i back on rack.) Based on his consrucion, you can show he following heorem 3. 3 Kääriäinen 2006 Theorem 19 (Lower Bound for SupervisedIL). There exis imiaion learning problems on which Algorihm 18.1 is able o achieve small classificaion error ɛ [0, 1/T under an opimal exper, bu for which he es loss is lower bounded as: E τ f l }{{} loss of learned policy T ɛ [ 1 (1 2ɛ) T+1 (18.3) which is bounded by T 2 ɛ and, for small ɛ, grows like T 2 ɛ.

5 216 a course in machine learning Up o consans, his gives maching upper and lower bounds for he loss of a policy learned by supervised imiaion learning ha is prey far (a facor of T) from wha we migh hope for Daase Aggregaion Supervised imiaion learning fails because once i ges off he exper pah, hings can go really badly. Ideally, we migh wan o rain our policy o deal wih any possible siuaion i could encouner. Unforunaely, his is unrealisic: we canno hope o be able o rain on every possible configuraion of he world; and if we could, we wouldn really need o learn anyway, we could jus memorize. So we wan o rain f on a subse of world configuraions, bu using configuraions visied by he exper fails because f canno learn o recover from is own errors. Somehow wha we d like o do is rain f o do well on he configuraions ha i, iself, encouners! This is a classic chicken-and-egg problem. We wan a policy f ha does well in a bunch of world configuraions. Wha se of configuraions? The configuraions ha f encouners! A very classic approach o solving chicken-and-egg problems is ieraion. Sar wih some policy f. Run f and see wha configuraions is visis. Train a new f o do well here. Repea. This is exacly wha he Daase Aggregaion algorihm ( Dagger ) does. Coninuing wih he self-driving car analogy, we firs le a human exper drive a car for a while, and learn an iniial policy f 0 by running sandard supervised imiaion learning (Algorihm 18.1) on he rajecories visied by he human. We hen do somehing unusual. We pu he human exper in he car, and record heir acions, bu he car behaves no according o he exper s behavior, bu according o f 0. Tha is, f 0 is in conrol of he car, and he exper is rying o seer, bu he car is ignoring hem 4 and simply recording heir acions as raining daa. This is shown in Figure Based on rajecories generaed by f 0 bu acions given by he exper, we generae a new daase ha conains informaion abou how o recover from he errors of f 0. We now will rain a new policy, f 1. Because we don wan f 1 o forge wha f 0 already knows, f 1 is rained on he union of he iniial exper-only rajecories ogeher wih he new rajecories generaed by f 0. We repea his process a number of imes MaxIer, yielding Algorihm This algorihm reurns he lis of all policies generaed during is run. A very pracical quesion is: which one should you use? There are essenially wo choices. The firs choice is jus o use he final policy learned. The problem wih his approach is ha Dagger can be somewha unsable in pracice, and policies do no monoonically 4 This is possibly errifying for he exper! f 0 exper Figure 18.2: In DAgger, he rajecory (red) is generaed according o he previously learned policy, f 0, bu he gold sandard acions are given by he exper.

6 imiaion learning 217 Algorihm 45 DaggerTrain(A, MaxIer, N, exper) 1: τ (0) n n=1 N run he exper N many imes 2: D 0 (x, a) : n, (x, a, l) τ (0) n // collec all pairs (same as supervised) 3: f 0 A(D 0 ) // rain iniial policy (muliclass classifier) on D 0 4: for i = 1... MaxIer do 5: τ (i) n n=1 N run policy f i 1 N-many imes // rajecories by f i 1 6: D i (x, exper(x)) : n, (x, a, l) τ (i) n // collec daa se ( ij=0 ) 7: f i A D j 8: end for // observaions x visied by f i 1 // bu acions according o he exper // rain policy f i on union of all daa so far 9: reurn f 0, f 1,..., f MaxIer // reurn collecion of all learned policies improve. A safer alernaive (as we ll see by heory below) is o es all of hem on some held-ou developmen asks, and pick he one ha does bes here. This requires a bi more compuaion, bu is a much beer approach in general. One major difference in requiremens beween Dagger (Algorihm 18.3) and SupervisedIL (Algorihm 18.1) is he requiremen of ineracion wih he exper. In SupervisedIL, you only need access o a bunch of rajecories aken by he exper, passively. In Dagger, you need access o hem exper hemselves, so you can ask quesions like if you saw configuraion x, wha would you do? This pus much more demand on he exper. Anoher quesion ha arises is: wha should N, he number of rajecories generaed in each round, be? In pracice, he iniial N should probably be reasonably large, so ha he iniial policy f 0 is prey good. The number of rajecories generaed by ieraion subsequenly can be much smaller, perhaps even jus one. Inuiively, Dagger should be less sensiive o compounding error han SupervisedIL, precisely because i ges rained on observaions ha i is likely o see a es ime. This is formalized in he following heorem: Theorem 20 (Loss of Dagger). Suppose ha one runs Algorihm 18.3 using a muliclass classifier ha opimizes he 0-1 loss (or an upperbound hereof). Le ɛ be he error rae of he underlying classifier (in expecaion) and assume ha all insananeous losses are in he range [0, l (max). Le f be he learned policy; hen: ( ) l l l +l (max) Tɛ + (max) T log T O E τ f }{{} loss of learned policy E τ exper } {{ } loss of exper MaxIer Furhermore, if he loss funcion is srongly convex in f, and MaxIer is (18.4)

7 218 a course in machine learning Õ(T/ɛ), hen: E τ f l }{{} loss of learned policy E τ exper l } {{ } loss of exper +l (max) Tɛ + O(ɛ) (18.5) Boh of hese resuls show ha, assuming MaxIer is large enough, he loss of he learned policy f (here, aken o be he bes on of all he MaxIer policies learned) grows like Tɛ, which is wha we hope for. Noe ha he final erm in he firs bound ges small so long as MaxIer is a leas T log T Expensive Algorihms as Expers Because of he srong requiremen on he exper in Dagger (i.e., ha you need o be able o query i many imes during raining), one of he mos subsanial use cases for Dagger is o learn o (quickly) imiae oherwise slow algorihms. Here are wo prooypical examples: 1. Game playing. When a game (like chess or minecraf) can be run in simulaion, you can ofen explicily compue a semi-opimal exper behavior wih brue-force search. Bu his search migh be oo compuaionally expensive o play in real ime, so you can use i during raining ime, learning a fas policy ha aemps o mimic he expensive search. This learned policy can hen be applied a es ime. 2. Discree opimizers. Many discree opimizaion problems can be compuaionally expensive o run in real ime; for insance, even shores pah search on a large graph can be oo slow for real ime use. We can compue shores pahs offline as raining daa and hen use imiaion learning o ry o build shores pah opimizers ha will run sufficienly efficienly in real ime. Consider he game playing example, and for concreeness, suppose you are rying o learn o play soliaire (his is an easier example because i s a single player game). When running DaggerTrain (Algorihm 18.3 o learn a chess-playing policy, he algorihm will repeaedly ask for exper(x), where x is he curren sae of he game. Wha should his funcion reurn? Ideally, i should reurn he/an acion a such ha, if a is aken, and hen he res of he game is played opimally, he player wins. Compuing his exacly is going o be very difficul for anyhing excep he simples games, so we need o resor o an approxiamion.

8 imiaion learning 219 Algorihm 46 DephLimiedDFS(x, h, MaxDeph) 1: if x is a erminal sae or MaxDeph 0 hen 2: reurn (, h(x)) // if we canno search deeper 3: else // reurn no acion ( ) and he curren heurisic score 4: BesAcion, BesScore, // keep rack of bes acion & is score 5: for all acions a from x do 6: (_, score) DephLimiedDFS(x a, h, MaxDeph 1) // ge score // for acion a, deph reduced by one by appending a o x 7: if score > BesScore hen 8: BesAcion, BesScore a, score // updae racked bes acion & score 9: end if 10: end for 11: end if 12: reurn (BesAcion, BesScore) // reurn bes found acion and is score A common sraegy is o run a deph-limied deph firs search, saring a sae x, and erminaing afer a mos hree of four moves (see Figure 18.3). This will generae a search ree. Unless you are very near he end of he game, none of he leaves of his ree will correspond o he end of he game. So you ll need some heurisic, h, for evaluaing saes ha are non-erminals. You can propagae his heurisic score up o he roo, and choose he acion ha looks bes wih his deph four search. This is no necessarily going o be he opimal acion, and here s a speed/accuracy rade-off for searching deeper, bu his is ypically effecive. This approach summarized in Algorihm Srucured Predicion via Imiaion Learning Figure 18.3: imi:dldfs: Deph limied deph-firs search A final case where an exper can ofen be compued algorihmically arises when one solves srucured predicion (see Chaper 17) via imiaion learning. I is cleares how his can work in he case of sequence labeling. Recall here ha prediced oupus should be sequences of labels. The running example from he earlier chaper was: x = monsers ea asy bunnies (18.6) y = noun verb adj noun (18.7) One can easily cas he predicion of y as a sequenial decision making problem, by reaing he producion of y in a lef-o-righ manner. In his case, we have a ime horizon T = 4. We wan o learn a policy f ha firs predics noun hen verb hen adj hen noun on his inpu.

9 220 a course in machine learning Le s suppose ha he inpu o f consiss of feaures exraced boh from he inpu (x) and he curren prediced oupu prefix ŷ, denoed φ(x, ŷ). For insance, φ(x, ŷ) migh represen a similar se of feaures o hose use in Chaper 17. I is perhaps easies o hink of f as jus a classifier: given some feaures of he inpu senence x ( monsers ea asy bunnies ), and some feaures abou previous predicions in he oupu prefix (so far, produced noun verb ), he goal of f is o predic he ag for he nex word ( asy ) in his conex. An imporan quesion is: wha is he exper in his case? Inuiively, he exper should provide he correc nex label, bu wha does his mean? Tha depends on he loss funcion being opimized. Under Hamming loss (sum zero/one loss over each individual predicion), he exper is sraighforward. When he exper is asked o produce an acion for he hird word, he exper s response is always adj (or whaever happens o be he correc label for he hird word in he senence i is currenly raining on). More generally, he exper ges o look a x, y and a prefix ŷ of he oupu. Noe, imporanly, ha he prefix migh be wrong! In paricular, afer he firs ieraion of Dagger, he prefix will be prediced by he learned policy, which may make misakes! The exper also has some srucured loss funcion l ha i is rying o minimize. Like in he previous secion, he exper s goal is o choose he acion ha minimizes he long-erm loss according o l on his example. To be more formal, we need a bi of noaion. Le bes(l, y, ŷ) denoe he loss (according o l and he ground ruh y) of he bes reachable oupu saring a ŷ. For insance, if y is noun verb adj noun and ŷ is noun noun, and he loss is Hamming loss, hen he bes achievable oupu (predicing lef-o-righ) is noun noun adj noun which has a loss of 1. Thus, bes for his siuaion is 1. Given ha noion of bes, he exper is easy o define: exper(l, y, ŷ) = argmin bes(l, y, ŷ a) (18.8) a Namely, i is he acion ha leads o he bes possible compleion afer aking ha acion. So in he example above, he exper acion is adj. For some problems and some loss funcions, compuing he exper is easy. In paricular, for sequence labeling under Hamming loss, i s rivial. In he case ha you can compue he exper exacly, i is ofen called an oracle. 5 For some oher problems, exacly compuing an oracle is compuaionally expensive or inracable. In hose cases, one can ofen resor o deph limied deph-firs-search (Algorihm 18.4) o compue an approximae oracle as an exper. To be very concree, a ypical implemenaion of Dagger applied o sequence labeling would go as follows. Each srucured raining example (a pair of senence and ag-sequence) gives rise o one rajec- 5 Some lieraure calls i a dynamic oracle, hough he exra word is unnecessary.

10 imiaion learning 221 ory. A raining ime, a predic ag seqence is generaed lef-o-righ, saring wih he empy sequence. A any given ime sep, you are aemping o predic he label of he h word in he inpu. You define a feaure vecor φ(x, ŷ), which will ypically consis of: (a) he h word, (b) lef and righ neighbors of he h word, (c) he las few predicions in ŷ, and (d) anyhing else you can hink of. In paricular, he feaures are no limied o Markov syle feaures, because we re no longer rying o do dynamic programming. The exper label for he h word is jus he corresponding label in he ground ruh y. Given all his, one can run Dagger (Algorihm 18.4) exacly as specified. Moving o srucured predicion problems oher han sequence labeling problems is beyond he scope of his book. The general framework is o cas your srucured predicion problem as a sequenial decision making problem. Once you ve done ha, you need o decide on feaures (his is he easy par) and an exper (his is ofen he harder par). However, once you ve done so, here are generic libraries for compiling your specificaion down o code Furher Reading TODO furher reading

1 Review of Zero-Sum Games

1 Review of Zero-Sum Games COS 5: heoreical Machine Learning Lecurer: Rob Schapire Lecure #23 Scribe: Eugene Brevdo April 30, 2008 Review of Zero-Sum Games Las ime we inroduced a mahemaical model for wo player zero-sum games. Any

More information

Lecture 2-1 Kinematics in One Dimension Displacement, Velocity and Acceleration Everything in the world is moving. Nothing stays still.

Lecture 2-1 Kinematics in One Dimension Displacement, Velocity and Acceleration Everything in the world is moving. Nothing stays still. Lecure - Kinemaics in One Dimension Displacemen, Velociy and Acceleraion Everyhing in he world is moving. Nohing says sill. Moion occurs a all scales of he universe, saring from he moion of elecrons in

More information

Chapter 7: Solving Trig Equations

Chapter 7: Solving Trig Equations Haberman MTH Secion I: The Trigonomeric Funcions Chaper 7: Solving Trig Equaions Le s sar by solving a couple of equaions ha involve he sine funcion EXAMPLE a: Solve he equaion sin( ) The inverse funcions

More information

Longest Common Prefixes

Longest Common Prefixes Longes Common Prefixes The sandard ordering for srings is he lexicographical order. I is induced by an order over he alphabe. We will use he same symbols (,

More information

Ensamble methods: Bagging and Boosting

Ensamble methods: Bagging and Boosting Lecure 21 Ensamble mehods: Bagging and Boosing Milos Hauskrech milos@cs.pi.edu 5329 Senno Square Ensemble mehods Mixure of expers Muliple base models (classifiers, regressors), each covers a differen par

More information

Ensamble methods: Boosting

Ensamble methods: Boosting Lecure 21 Ensamble mehods: Boosing Milos Hauskrech milos@cs.pi.edu 5329 Senno Square Schedule Final exam: April 18: 1:00-2:15pm, in-class Term projecs April 23 & April 25: a 1:00-2:30pm in CS seminar room

More information

Learning Objectives: Practice designing and simulating digital circuits including flip flops Experience state machine design procedure

Learning Objectives: Practice designing and simulating digital circuits including flip flops Experience state machine design procedure Lab 4: Synchronous Sae Machine Design Summary: Design and implemen synchronous sae machine circuis and es hem wih simulaions in Cadence Viruoso. Learning Objecives: Pracice designing and simulaing digial

More information

Seminar 4: Hotelling 2

Seminar 4: Hotelling 2 Seminar 4: Hoelling 2 November 3, 211 1 Exercise Par 1 Iso-elasic demand A non renewable resource of a known sock S can be exraced a zero cos. Demand for he resource is of he form: D(p ) = p ε ε > A a

More information

Vehicle Arrival Models : Headway

Vehicle Arrival Models : Headway Chaper 12 Vehicle Arrival Models : Headway 12.1 Inroducion Modelling arrival of vehicle a secion of road is an imporan sep in raffic flow modelling. I has imporan applicaion in raffic flow simulaion where

More information

= ( ) ) or a system of differential equations with continuous parametrization (T = R

= ( ) ) or a system of differential equations with continuous parametrization (T = R XIII. DIFFERENCE AND DIFFERENTIAL EQUATIONS Ofen funcions, or a sysem of funcion, are paramerized in erms of some variable, usually denoed as and inerpreed as ime. The variable is wrien as a funcion of

More information

Lecture 33: November 29

Lecture 33: November 29 36-705: Inermediae Saisics Fall 2017 Lecurer: Siva Balakrishnan Lecure 33: November 29 Today we will coninue discussing he boosrap, and hen ry o undersand why i works in a simple case. In he las lecure

More information

( ) a system of differential equations with continuous parametrization ( T = R + These look like, respectively:

( ) a system of differential equations with continuous parametrization ( T = R + These look like, respectively: XIII. DIFFERENCE AND DIFFERENTIAL EQUATIONS Ofen funcions, or a sysem of funcion, are paramerized in erms of some variable, usually denoed as and inerpreed as ime. The variable is wrien as a funcion of

More information

Diebold, Chapter 7. Francis X. Diebold, Elements of Forecasting, 4th Edition (Mason, Ohio: Cengage Learning, 2006). Chapter 7. Characterizing Cycles

Diebold, Chapter 7. Francis X. Diebold, Elements of Forecasting, 4th Edition (Mason, Ohio: Cengage Learning, 2006). Chapter 7. Characterizing Cycles Diebold, Chaper 7 Francis X. Diebold, Elemens of Forecasing, 4h Ediion (Mason, Ohio: Cengage Learning, 006). Chaper 7. Characerizing Cycles Afer compleing his reading you should be able o: Define covariance

More information

KINEMATICS IN ONE DIMENSION

KINEMATICS IN ONE DIMENSION KINEMATICS IN ONE DIMENSION PREVIEW Kinemaics is he sudy of how hings move how far (disance and displacemen), how fas (speed and velociy), and how fas ha how fas changes (acceleraion). We say ha an objec

More information

T L. t=1. Proof of Lemma 1. Using the marginal cost accounting in Equation(4) and standard arguments. t )+Π RB. t )+K 1(Q RB

T L. t=1. Proof of Lemma 1. Using the marginal cost accounting in Equation(4) and standard arguments. t )+Π RB. t )+K 1(Q RB Elecronic Companion EC.1. Proofs of Technical Lemmas and Theorems LEMMA 1. Le C(RB) be he oal cos incurred by he RB policy. Then we have, T L E[C(RB)] 3 E[Z RB ]. (EC.1) Proof of Lemma 1. Using he marginal

More information

Christos Papadimitriou & Luca Trevisan November 22, 2016

Christos Papadimitriou & Luca Trevisan November 22, 2016 U.C. Bereley CS170: Algorihms Handou LN-11-22 Chrisos Papadimiriou & Luca Trevisan November 22, 2016 Sreaming algorihms In his lecure and he nex one we sudy memory-efficien algorihms ha process a sream

More information

Learning a Class from Examples. Training set X. Class C 1. Class C of a family car. Output: Input representation: x 1 : price, x 2 : engine power

Learning a Class from Examples. Training set X. Class C 1. Class C of a family car. Output: Input representation: x 1 : price, x 2 : engine power Alpaydin Chaper, Michell Chaper 7 Alpaydin slides are in urquoise. Ehem Alpaydin, copyrigh: The MIT Press, 010. alpaydin@boun.edu.r hp://www.cmpe.boun.edu.r/ ehem/imle All oher slides are based on Michell.

More information

Learning a Class from Examples. Training set X. Class C 1. Class C of a family car. Output: Input representation: x 1 : price, x 2 : engine power

Learning a Class from Examples. Training set X. Class C 1. Class C of a family car. Output: Input representation: x 1 : price, x 2 : engine power Alpaydin Chaper, Michell Chaper 7 Alpaydin slides are in urquoise. Ehem Alpaydin, copyrigh: The MIT Press, 010. alpaydin@boun.edu.r hp://www.cmpe.boun.edu.r/ ehem/imle All oher slides are based on Michell.

More information

Final Spring 2007

Final Spring 2007 .615 Final Spring 7 Overview The purpose of he final exam is o calculae he MHD β limi in a high-bea oroidal okamak agains he dangerous n = 1 exernal ballooning-kink mode. Effecively, his corresponds o

More information

Biol. 356 Lab 8. Mortality, Recruitment, and Migration Rates

Biol. 356 Lab 8. Mortality, Recruitment, and Migration Rates Biol. 356 Lab 8. Moraliy, Recruimen, and Migraion Raes (modified from Cox, 00, General Ecology Lab Manual, McGraw Hill) Las week we esimaed populaion size hrough several mehods. One assumpion of all hese

More information

Online Convex Optimization Example And Follow-The-Leader

Online Convex Optimization Example And Follow-The-Leader CSE599s, Spring 2014, Online Learning Lecure 2-04/03/2014 Online Convex Opimizaion Example And Follow-The-Leader Lecurer: Brendan McMahan Scribe: Sephen Joe Jonany 1 Review of Online Convex Opimizaion

More information

Lecture 4 Notes (Little s Theorem)

Lecture 4 Notes (Little s Theorem) Lecure 4 Noes (Lile s Theorem) This lecure concerns one of he mos imporan (and simples) heorems in Queuing Theory, Lile s Theorem. More informaion can be found in he course book, Bersekas & Gallagher,

More information

16 Max-Flow Algorithms

16 Max-Flow Algorithms A process canno be undersood by sopping i. Undersanding mus move wih he flow of he process, mus join i and flow wih i. The Firs Law of Mena, in Frank Herber s Dune (196) There s a difference beween knowing

More information

Phys1112: DC and RC circuits

Phys1112: DC and RC circuits Name: Group Members: Dae: TA s Name: Phys1112: DC and RC circuis Objecives: 1. To undersand curren and volage characerisics of a DC RC discharging circui. 2. To undersand he effec of he RC ime consan.

More information

MATH 4330/5330, Fourier Analysis Section 6, Proof of Fourier s Theorem for Pointwise Convergence

MATH 4330/5330, Fourier Analysis Section 6, Proof of Fourier s Theorem for Pointwise Convergence MATH 433/533, Fourier Analysis Secion 6, Proof of Fourier s Theorem for Poinwise Convergence Firs, some commens abou inegraing periodic funcions. If g is a periodic funcion, g(x + ) g(x) for all real x,

More information

Fishing limits and the Logistic Equation. 1

Fishing limits and the Logistic Equation. 1 Fishing limis and he Logisic Equaion. 1 1. The Logisic Equaion. The logisic equaion is an equaion governing populaion growh for populaions in an environmen wih a limied amoun of resources (for insance,

More information

Notes for Lecture 17-18

Notes for Lecture 17-18 U.C. Berkeley CS278: Compuaional Complexiy Handou N7-8 Professor Luca Trevisan April 3-8, 2008 Noes for Lecure 7-8 In hese wo lecures we prove he firs half of he PCP Theorem, he Amplificaion Lemma, up

More information

Deep Learning: Theory, Techniques & Applications - Recurrent Neural Networks -

Deep Learning: Theory, Techniques & Applications - Recurrent Neural Networks - Deep Learning: Theory, Techniques & Applicaions - Recurren Neural Neworks - Prof. Maeo Maeucci maeo.maeucci@polimi.i Deparmen of Elecronics, Informaion and Bioengineering Arificial Inelligence and Roboics

More information

Inventory Analysis and Management. Multi-Period Stochastic Models: Optimality of (s, S) Policy for K-Convex Objective Functions

Inventory Analysis and Management. Multi-Period Stochastic Models: Optimality of (s, S) Policy for K-Convex Objective Functions Muli-Period Sochasic Models: Opimali of (s, S) Polic for -Convex Objecive Funcions Consider a seing similar o he N-sage newsvendor problem excep ha now here is a fixed re-ordering cos (> 0) for each (re-)order.

More information

Reading from Young & Freedman: For this topic, read sections 25.4 & 25.5, the introduction to chapter 26 and sections 26.1 to 26.2 & 26.4.

Reading from Young & Freedman: For this topic, read sections 25.4 & 25.5, the introduction to chapter 26 and sections 26.1 to 26.2 & 26.4. PHY1 Elecriciy Topic 7 (Lecures 1 & 11) Elecric Circuis n his opic, we will cover: 1) Elecromoive Force (EMF) ) Series and parallel resisor combinaions 3) Kirchhoff s rules for circuis 4) Time dependence

More information

Two Coupled Oscillators / Normal Modes

Two Coupled Oscillators / Normal Modes Lecure 3 Phys 3750 Two Coupled Oscillaors / Normal Modes Overview and Moivaion: Today we ake a small, bu significan, sep owards wave moion. We will no ye observe waves, bu his sep is imporan in is own

More information

23.2. Representing Periodic Functions by Fourier Series. Introduction. Prerequisites. Learning Outcomes

23.2. Representing Periodic Functions by Fourier Series. Introduction. Prerequisites. Learning Outcomes Represening Periodic Funcions by Fourier Series 3. Inroducion In his Secion we show how a periodic funcion can be expressed as a series of sines and cosines. We begin by obaining some sandard inegrals

More information

Finish reading Chapter 2 of Spivak, rereading earlier sections as necessary. handout and fill in some missing details!

Finish reading Chapter 2 of Spivak, rereading earlier sections as necessary. handout and fill in some missing details! MAT 257, Handou 6: Ocober 7-2, 20. I. Assignmen. Finish reading Chaper 2 of Spiva, rereading earlier secions as necessary. handou and fill in some missing deails! II. Higher derivaives. Also, read his

More information

Math 2142 Exam 1 Review Problems. x 2 + f (0) 3! for the 3rd Taylor polynomial at x = 0. To calculate the various quantities:

Math 2142 Exam 1 Review Problems. x 2 + f (0) 3! for the 3rd Taylor polynomial at x = 0. To calculate the various quantities: Mah 4 Eam Review Problems Problem. Calculae he 3rd Taylor polynomial for arcsin a =. Soluion. Le f() = arcsin. For his problem, we use he formula f() + f () + f ()! + f () 3! for he 3rd Taylor polynomial

More information

Laplace transfom: t-translation rule , Haynes Miller and Jeremy Orloff

Laplace transfom: t-translation rule , Haynes Miller and Jeremy Orloff Laplace ransfom: -ranslaion rule 8.03, Haynes Miller and Jeremy Orloff Inroducory example Consider he sysem ẋ + 3x = f(, where f is he inpu and x he response. We know is uni impulse response is 0 for

More information

3.1 More on model selection

3.1 More on model selection 3. More on Model selecion 3. Comparing models AIC, BIC, Adjused R squared. 3. Over Fiing problem. 3.3 Sample spliing. 3. More on model selecion crieria Ofen afer model fiing you are lef wih a handful of

More information

RL Lecture 7: Eligibility Traces. R. S. Sutton and A. G. Barto: Reinforcement Learning: An Introduction 1

RL Lecture 7: Eligibility Traces. R. S. Sutton and A. G. Barto: Reinforcement Learning: An Introduction 1 RL Lecure 7: Eligibiliy Traces R. S. Suon and A. G. Baro: Reinforcemen Learning: An Inroducion 1 N-sep TD Predicion Idea: Look farher ino he fuure when you do TD backup (1, 2, 3,, n seps) R. S. Suon and

More information

) were both constant and we brought them from under the integral.

) were both constant and we brought them from under the integral. YIELD-PER-RECRUIT (coninued The yield-per-recrui model applies o a cohor, bu we saw in he Age Disribuions lecure ha he properies of a cohor do no apply in general o a collecion of cohors, which is wha

More information

3.1.3 INTRODUCTION TO DYNAMIC OPTIMIZATION: DISCRETE TIME PROBLEMS. A. The Hamiltonian and First-Order Conditions in a Finite Time Horizon

3.1.3 INTRODUCTION TO DYNAMIC OPTIMIZATION: DISCRETE TIME PROBLEMS. A. The Hamiltonian and First-Order Conditions in a Finite Time Horizon 3..3 INRODUCION O DYNAMIC OPIMIZAION: DISCREE IME PROBLEMS A. he Hamilonian and Firs-Order Condiions in a Finie ime Horizon Define a new funcion, he Hamilonian funcion, H. H he change in he oal value of

More information

Simulation-Solving Dynamic Models ABE 5646 Week 2, Spring 2010

Simulation-Solving Dynamic Models ABE 5646 Week 2, Spring 2010 Simulaion-Solving Dynamic Models ABE 5646 Week 2, Spring 2010 Week Descripion Reading Maerial 2 Compuer Simulaion of Dynamic Models Finie Difference, coninuous saes, discree ime Simple Mehods Euler Trapezoid

More information

5.1 - Logarithms and Their Properties

5.1 - Logarithms and Their Properties Chaper 5 Logarihmic Funcions 5.1 - Logarihms and Their Properies Suppose ha a populaion grows according o he formula P 10, where P is he colony size a ime, in hours. When will he populaion be 2500? We

More information

More Digital Logic. t p output. Low-to-high and high-to-low transitions could have different t p. V in (t)

More Digital Logic. t p output. Low-to-high and high-to-low transitions could have different t p. V in (t) EECS 4 Spring 23 Lecure 2 EECS 4 Spring 23 Lecure 2 More igial Logic Gae delay and signal propagaion Clocked circui elemens (flip-flop) Wriing a word o memory Simplifying digial circuis: Karnaugh maps

More information

Unit Root Time Series. Univariate random walk

Unit Root Time Series. Univariate random walk Uni Roo ime Series Univariae random walk Consider he regression y y where ~ iid N 0, he leas squares esimae of is: ˆ yy y y yy Now wha if = If y y hen le y 0 =0 so ha y j j If ~ iid N 0, hen y ~ N 0, he

More information

V The Fourier Transform

V The Fourier Transform V he Fourier ransform Lecure noes by Assaf al 1. Moivaion Imagine playing hree noes on he piano, recording hem (soring hem as a.wav or.mp3 file), and hen ploing he resuling waveform on he compuer: 100Hz

More information

INTRODUCTION TO MACHINE LEARNING 3RD EDITION

INTRODUCTION TO MACHINE LEARNING 3RD EDITION ETHEM ALPAYDIN The MIT Press, 2014 Lecure Slides for INTRODUCTION TO MACHINE LEARNING 3RD EDITION alpaydin@boun.edu.r hp://www.cmpe.boun.edu.r/~ehem/i2ml3e CHAPTER 2: SUPERVISED LEARNING Learning a Class

More information

Echocardiography Project and Finite Fourier Series

Echocardiography Project and Finite Fourier Series Echocardiography Projec and Finie Fourier Series 1 U M An echocardiagram is a plo of how a porion of he hear moves as he funcion of ime over he one or more hearbea cycles If he hearbea repeas iself every

More information

Chapter 2. First Order Scalar Equations

Chapter 2. First Order Scalar Equations Chaper. Firs Order Scalar Equaions We sar our sudy of differenial equaions in he same way he pioneers in his field did. We show paricular echniques o solve paricular ypes of firs order differenial equaions.

More information

Designing Information Devices and Systems I Spring 2019 Lecture Notes Note 17

Designing Information Devices and Systems I Spring 2019 Lecture Notes Note 17 EES 16A Designing Informaion Devices and Sysems I Spring 019 Lecure Noes Noe 17 17.1 apaciive ouchscreen In he las noe, we saw ha a capacior consiss of wo pieces on conducive maerial separaed by a nonconducive

More information

3.6 Derivatives as Rates of Change

3.6 Derivatives as Rates of Change 3.6 Derivaives as Raes of Change Problem 1 John is walking along a sraigh pah. His posiion a he ime >0 is given by s = f(). He sars a =0from his house (f(0) = 0) and he graph of f is given below. (a) Describe

More information

2.7. Some common engineering functions. Introduction. Prerequisites. Learning Outcomes

2.7. Some common engineering functions. Introduction. Prerequisites. Learning Outcomes Some common engineering funcions 2.7 Inroducion This secion provides a caalogue of some common funcions ofen used in Science and Engineering. These include polynomials, raional funcions, he modulus funcion

More information

SOLUTIONS TO ECE 3084

SOLUTIONS TO ECE 3084 SOLUTIONS TO ECE 384 PROBLEM 2.. For each sysem below, specify wheher or no i is: (i) memoryless; (ii) causal; (iii) inverible; (iv) linear; (v) ime invarian; Explain your reasoning. If he propery is no

More information

Solutions from Chapter 9.1 and 9.2

Solutions from Chapter 9.1 and 9.2 Soluions from Chaper 9 and 92 Secion 9 Problem # This basically boils down o an exercise in he chain rule from calculus We are looking for soluions of he form: u( x) = f( k x c) where k x R 3 and k is

More information

FITTING EQUATIONS TO DATA

FITTING EQUATIONS TO DATA TANTON S TAKE ON FITTING EQUATIONS TO DATA CURRICULUM TIDBITS FOR THE MATHEMATICS CLASSROOM MAY 013 Sandard algebra courses have sudens fi linear and eponenial funcions o wo daa poins, and quadraic funcions

More information

Explaining Total Factor Productivity. Ulrich Kohli University of Geneva December 2015

Explaining Total Factor Productivity. Ulrich Kohli University of Geneva December 2015 Explaining Toal Facor Produciviy Ulrich Kohli Universiy of Geneva December 2015 Needed: A Theory of Toal Facor Produciviy Edward C. Presco (1998) 2 1. Inroducion Toal Facor Produciviy (TFP) has become

More information

!!"#"$%&#'()!"#&'(*%)+,&',-)./0)1-*23)

!!#$%&#'()!#&'(*%)+,&',-)./0)1-*23) "#"$%&#'()"#&'(*%)+,&',-)./)1-*) #$%&'()*+,&',-.%,/)*+,-&1*#$)()5*6$+$%*,7&*-'-&1*(,-&*6&,7.$%$+*&%'(*8$&',-,%'-&1*(,-&*6&,79*(&,%: ;..,*&1$&$.$%&'()*1$$.,'&',-9*(&,%)?%*,('&5

More information

10. State Space Methods

10. State Space Methods . Sae Space Mehods. Inroducion Sae space modelling was briefly inroduced in chaper. Here more coverage is provided of sae space mehods before some of heir uses in conrol sysem design are covered in he

More information

BU Macro BU Macro Fall 2008, Lecture 4

BU Macro BU Macro Fall 2008, Lecture 4 Dynamic Programming BU Macro 2008 Lecure 4 1 Ouline 1. Cerainy opimizaion problem used o illusrae: a. Resricions on exogenous variables b. Value funcion c. Policy funcion d. The Bellman equaion and an

More information

Sequential Importance Resampling (SIR) Particle Filter

Sequential Importance Resampling (SIR) Particle Filter Paricle Filers++ Pieer Abbeel UC Berkeley EECS Many slides adaped from Thrun, Burgard and Fox, Probabilisic Roboics 1. Algorihm paricle_filer( S -1, u, z ): 2. Sequenial Imporance Resampling (SIR) Paricle

More information

Section 7.4 Modeling Changing Amplitude and Midline

Section 7.4 Modeling Changing Amplitude and Midline 488 Chaper 7 Secion 7.4 Modeling Changing Ampliude and Midline While sinusoidal funcions can model a variey of behaviors, i is ofen necessary o combine sinusoidal funcions wih linear and exponenial curves

More information

STATE-SPACE MODELLING. A mass balance across the tank gives:

STATE-SPACE MODELLING. A mass balance across the tank gives: B. Lennox and N.F. Thornhill, 9, Sae Space Modelling, IChemE Process Managemen and Conrol Subjec Group Newsleer STE-SPACE MODELLING Inroducion: Over he pas decade or so here has been an ever increasing

More information

Solutions Problem Set 3 Macro II (14.452)

Solutions Problem Set 3 Macro II (14.452) Soluions Problem Se 3 Macro II (14.452) Francisco A. Gallego 04/27/2005 1 Q heory of invesmen in coninuous ime and no uncerainy Consider he in nie horizon model of a rm facing adjusmen coss o invesmen.

More information

Lecture Notes 2. The Hilbert Space Approach to Time Series

Lecture Notes 2. The Hilbert Space Approach to Time Series Time Series Seven N. Durlauf Universiy of Wisconsin. Basic ideas Lecure Noes. The Hilber Space Approach o Time Series The Hilber space framework provides a very powerful language for discussing he relaionship

More information

SZG Macro 2011 Lecture 3: Dynamic Programming. SZG macro 2011 lecture 3 1

SZG Macro 2011 Lecture 3: Dynamic Programming. SZG macro 2011 lecture 3 1 SZG Macro 2011 Lecure 3: Dynamic Programming SZG macro 2011 lecure 3 1 Background Our previous discussion of opimal consumpion over ime and of opimal capial accumulaion sugges sudying he general decision

More information

Predator - Prey Model Trajectories and the nonlinear conservation law

Predator - Prey Model Trajectories and the nonlinear conservation law Predaor - Prey Model Trajecories and he nonlinear conservaion law James K. Peerson Deparmen of Biological Sciences and Deparmen of Mahemaical Sciences Clemson Universiy Ocober 28, 213 Ouline Drawing Trajecories

More information

13.3 Term structure models

13.3 Term structure models 13.3 Term srucure models 13.3.1 Expecaions hypohesis model - Simples "model" a) shor rae b) expecaions o ge oher prices Resul: y () = 1 h +1 δ = φ( δ)+ε +1 f () = E (y +1) (1) =δ + φ( δ) f (3) = E (y +)

More information

Morning Time: 1 hour 30 minutes Additional materials (enclosed):

Morning Time: 1 hour 30 minutes Additional materials (enclosed): ADVANCED GCE 78/0 MATHEMATICS (MEI) Differenial Equaions THURSDAY JANUARY 008 Morning Time: hour 30 minues Addiional maerials (enclosed): None Addiional maerials (required): Answer Bookle (8 pages) Graph

More information

Written HW 9 Sol. CS 188 Fall Introduction to Artificial Intelligence

Written HW 9 Sol. CS 188 Fall Introduction to Artificial Intelligence CS 188 Fall 2018 Inroducion o Arificial Inelligence Wrien HW 9 Sol. Self-assessmen due: Tuesday 11/13/2018 a 11:59pm (submi via Gradescope) For he self assessmen, fill in he self assessmen boxes in your

More information

Two Popular Bayesian Estimators: Particle and Kalman Filters. McGill COMP 765 Sept 14 th, 2017

Two Popular Bayesian Estimators: Particle and Kalman Filters. McGill COMP 765 Sept 14 th, 2017 Two Popular Bayesian Esimaors: Paricle and Kalman Filers McGill COMP 765 Sep 14 h, 2017 1 1 1, dx x Bel x u x P x z P Recall: Bayes Filers,,,,,,, 1 1 1 1 u z u x P u z u x z P Bayes z = observaion u =

More information

SPH3U: Projectiles. Recorder: Manager: Speaker:

SPH3U: Projectiles. Recorder: Manager: Speaker: SPH3U: Projeciles Now i s ime o use our new skills o analyze he moion of a golf ball ha was ossed hrough he air. Le s find ou wha is special abou he moion of a projecile. Recorder: Manager: Speaker: 0

More information

ACE 562 Fall Lecture 5: The Simple Linear Regression Model: Sampling Properties of the Least Squares Estimators. by Professor Scott H.

ACE 562 Fall Lecture 5: The Simple Linear Regression Model: Sampling Properties of the Least Squares Estimators. by Professor Scott H. ACE 56 Fall 005 Lecure 5: he Simple Linear Regression Model: Sampling Properies of he Leas Squares Esimaors by Professor Sco H. Irwin Required Reading: Griffihs, Hill and Judge. "Inference in he Simple

More information

Section 5: Chain Rule

Section 5: Chain Rule Chaper The Derivaive Applie Calculus 11 Secion 5: Chain Rule There is one more ype of complicae funcion ha we will wan o know how o iffereniae: composiion. The Chain Rule will le us fin he erivaive of

More information

t is a basis for the solution space to this system, then the matrix having these solutions as columns, t x 1 t, x 2 t,... x n t x 2 t...

t is a basis for the solution space to this system, then the matrix having these solutions as columns, t x 1 t, x 2 t,... x n t x 2 t... Mah 228- Fri Mar 24 5.6 Marix exponenials and linear sysems: The analogy beween firs order sysems of linear differenial equaions (Chaper 5) and scalar linear differenial equaions (Chaper ) is much sronger

More information

Lecture 2 October ε-approximation of 2-player zero-sum games

Lecture 2 October ε-approximation of 2-player zero-sum games Opimizaion II Winer 009/10 Lecurer: Khaled Elbassioni Lecure Ocober 19 1 ε-approximaion of -player zero-sum games In his lecure we give a randomized ficiious play algorihm for obaining an approximae soluion

More information

Self assessment due: Monday 4/29/2019 at 11:59pm (submit via Gradescope)

Self assessment due: Monday 4/29/2019 at 11:59pm (submit via Gradescope) CS 188 Spring 2019 Inroducion o Arificial Inelligence Wrien HW 10 Due: Monday 4/22/2019 a 11:59pm (submi via Gradescope). Leave self assessmen boxes blank for his due dae. Self assessmen due: Monday 4/29/2019

More information

Asymptotic Equipartition Property - Seminar 3, part 1

Asymptotic Equipartition Property - Seminar 3, part 1 Asympoic Equipariion Propery - Seminar 3, par 1 Ocober 22, 2013 Problem 1 (Calculaion of ypical se) To clarify he noion of a ypical se A (n) ε and he smalles se of high probabiliy B (n), we will calculae

More information

Physics 235 Chapter 2. Chapter 2 Newtonian Mechanics Single Particle

Physics 235 Chapter 2. Chapter 2 Newtonian Mechanics Single Particle Chaper 2 Newonian Mechanics Single Paricle In his Chaper we will review wha Newon s laws of mechanics ell us abou he moion of a single paricle. Newon s laws are only valid in suiable reference frames,

More information

PENALIZED LEAST SQUARES AND PENALIZED LIKELIHOOD

PENALIZED LEAST SQUARES AND PENALIZED LIKELIHOOD PENALIZED LEAST SQUARES AND PENALIZED LIKELIHOOD HAN XIAO 1. Penalized Leas Squares Lasso solves he following opimizaion problem, ˆβ lasso = arg max β R p+1 1 N y i β 0 N x ij β j β j (1.1) for some 0.

More information

( ) is the stretch factor, and x the

( ) is the stretch factor, and x the (Lecures 7-8) Liddle, Chaper 5 Simple cosmological models (i) Hubble s Law revisied Self-similar srech of he universe All universe models have his characerisic v r ; v = Hr since only his conserves homogeneiy

More information

Maintenance Models. Prof. Robert C. Leachman IEOR 130, Methods of Manufacturing Improvement Spring, 2011

Maintenance Models. Prof. Robert C. Leachman IEOR 130, Methods of Manufacturing Improvement Spring, 2011 Mainenance Models Prof Rober C Leachman IEOR 3, Mehods of Manufacuring Improvemen Spring, Inroducion The mainenance of complex equipmen ofen accouns for a large porion of he coss associaed wih ha equipmen

More information

Notes on Kalman Filtering

Notes on Kalman Filtering Noes on Kalman Filering Brian Borchers and Rick Aser November 7, Inroducion Daa Assimilaion is he problem of merging model predicions wih acual measuremens of a sysem o produce an opimal esimae of he curren

More information

1. An introduction to dynamic optimization -- Optimal Control and Dynamic Programming AGEC

1. An introduction to dynamic optimization -- Optimal Control and Dynamic Programming AGEC This documen was generaed a :45 PM 8/8/04 Copyrigh 04 Richard T. Woodward. An inroducion o dynamic opimizaion -- Opimal Conrol and Dynamic Programming AGEC 637-04 I. Overview of opimizaion Opimizaion is

More information

Dynamic Programming 11/8/2009. Weighted Interval Scheduling. Weighted Interval Scheduling. Unweighted Interval Scheduling: Review

Dynamic Programming 11/8/2009. Weighted Interval Scheduling. Weighted Interval Scheduling. Unweighted Interval Scheduling: Review //9 Algorihms Dynamic Programming - Weighed Ineral Scheduling Dynamic Programming Weighed ineral scheduling problem. Insance A se of n jobs. Job j sars a s j, finishes a f j, and has weigh or alue j. Two

More information

5. Stochastic processes (1)

5. Stochastic processes (1) Lec05.pp S-38.45 - Inroducion o Teleraffic Theory Spring 2005 Conens Basic conceps Poisson process 2 Sochasic processes () Consider some quaniy in a eleraffic (or any) sysem I ypically evolves in ime randomly

More information

5.2. The Natural Logarithm. Solution

5.2. The Natural Logarithm. Solution 5.2 The Naural Logarihm The number e is an irraional number, similar in naure o π. Is non-erminaing, non-repeaing value is e 2.718 281 828 59. Like π, e also occurs frequenly in naural phenomena. In fac,

More information

15. Bicycle Wheel. Graph of height y (cm) above the axle against time t (s) over a 6-second interval. 15 bike wheel

15. Bicycle Wheel. Graph of height y (cm) above the axle against time t (s) over a 6-second interval. 15 bike wheel 15. Biccle Wheel The graph We moun a biccle wheel so ha i is free o roae in a verical plane. In fac, wha works easil is o pu an exension on one of he axles, and ge a suden o sand on one side and hold he

More information

Linear Time-invariant systems, Convolution, and Cross-correlation

Linear Time-invariant systems, Convolution, and Cross-correlation Linear Time-invarian sysems, Convoluion, and Cross-correlaion (1) Linear Time-invarian (LTI) sysem A sysem akes in an inpu funcion and reurns an oupu funcion. x() T y() Inpu Sysem Oupu y() = T[x()] An

More information

Technical Report Doc ID: TR March-2013 (Last revision: 23-February-2016) On formulating quadratic functions in optimization models.

Technical Report Doc ID: TR March-2013 (Last revision: 23-February-2016) On formulating quadratic functions in optimization models. Technical Repor Doc ID: TR--203 06-March-203 (Las revision: 23-Februar-206) On formulaing quadraic funcions in opimizaion models. Auhor: Erling D. Andersen Convex quadraic consrains quie frequenl appear

More information

23.5. Half-Range Series. Introduction. Prerequisites. Learning Outcomes

23.5. Half-Range Series. Introduction. Prerequisites. Learning Outcomes Half-Range Series 2.5 Inroducion In his Secion we address he following problem: Can we find a Fourier series expansion of a funcion defined over a finie inerval? Of course we recognise ha such a funcion

More information

Random Walk with Anti-Correlated Steps

Random Walk with Anti-Correlated Steps Random Walk wih Ani-Correlaed Seps John Noga Dirk Wagner 2 Absrac We conjecure he expeced value of random walks wih ani-correlaed seps o be exacly. We suppor his conjecure wih 2 plausibiliy argumens and

More information

0.1 MAXIMUM LIKELIHOOD ESTIMATION EXPLAINED

0.1 MAXIMUM LIKELIHOOD ESTIMATION EXPLAINED 0.1 MAXIMUM LIKELIHOOD ESTIMATIO EXPLAIED Maximum likelihood esimaion is a bes-fi saisical mehod for he esimaion of he values of he parameers of a sysem, based on a se of observaions of a random variable

More information

This document was generated at 7:34 PM, 07/27/09 Copyright 2009 Richard T. Woodward

This document was generated at 7:34 PM, 07/27/09 Copyright 2009 Richard T. Woodward his documen was generaed a 7:34 PM, 07/27/09 Copyrigh 2009 Richard. Woodward 15. Bang-bang and mos rapid approach problems AGEC 637 - Summer 2009 here are some problems for which he opimal pah does no

More information

Chapter 15: Phenomena. Chapter 15 Chemical Kinetics. Reaction Rates. Reaction Rates R P. Reaction Rates. Rate Laws

Chapter 15: Phenomena. Chapter 15 Chemical Kinetics. Reaction Rates. Reaction Rates R P. Reaction Rates. Rate Laws Chaper 5: Phenomena Phenomena: The reacion (aq) + B(aq) C(aq) was sudied a wo differen emperaures (98 K and 35 K). For each emperaure he reacion was sared by puing differen concenraions of he 3 species

More information

A First Course on Kinetics and Reaction Engineering. Class 19 on Unit 18

A First Course on Kinetics and Reaction Engineering. Class 19 on Unit 18 A Firs ourse on Kineics and Reacion Engineering lass 19 on Uni 18 Par I - hemical Reacions Par II - hemical Reacion Kineics Where We re Going Par III - hemical Reacion Engineering A. Ideal Reacors B. Perfecly

More information

EE363 homework 1 solutions

EE363 homework 1 solutions EE363 Prof. S. Boyd EE363 homework 1 soluions 1. LQR for a riple accumulaor. We consider he sysem x +1 = Ax + Bu, y = Cx, wih 1 1 A = 1 1, B =, C = [ 1 ]. 1 1 This sysem has ransfer funcion H(z) = (z 1)

More information

IMPLICIT AND INVERSE FUNCTION THEOREMS PAUL SCHRIMPF 1 OCTOBER 25, 2013

IMPLICIT AND INVERSE FUNCTION THEOREMS PAUL SCHRIMPF 1 OCTOBER 25, 2013 IMPLICI AND INVERSE FUNCION HEOREMS PAUL SCHRIMPF 1 OCOBER 25, 213 UNIVERSIY OF BRIISH COLUMBIA ECONOMICS 526 We have exensively sudied how o solve sysems of linear equaions. We know how o check wheher

More information

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation Course Noes for EE7C Spring 018: Convex Opimizaion and Approximaion Insrucor: Moriz Hard Email: hard+ee7c@berkeley.edu Graduae Insrucor: Max Simchowiz Email: msimchow+ee7c@berkeley.edu Ocober 15, 018 3

More information

Section 3.5 Nonhomogeneous Equations; Method of Undetermined Coefficients

Section 3.5 Nonhomogeneous Equations; Method of Undetermined Coefficients Secion 3.5 Nonhomogeneous Equaions; Mehod of Undeermined Coefficiens Key Terms/Ideas: Linear Differenial operaor Nonlinear operaor Second order homogeneous DE Second order nonhomogeneous DE Soluion o homogeneous

More information

Some Basic Information about M-S-D Systems

Some Basic Information about M-S-D Systems Some Basic Informaion abou M-S-D Sysems 1 Inroducion We wan o give some summary of he facs concerning unforced (homogeneous) and forced (non-homogeneous) models for linear oscillaors governed by second-order,

More information

E β t log (C t ) + M t M t 1. = Y t + B t 1 P t. B t 0 (3) v t = P tc t M t Question 1. Find the FOC s for an optimum in the agent s problem.

E β t log (C t ) + M t M t 1. = Y t + B t 1 P t. B t 0 (3) v t = P tc t M t Question 1. Find the FOC s for an optimum in the agent s problem. Noes, M. Krause.. Problem Se 9: Exercise on FTPL Same model as in paper and lecure, only ha one-period govenmen bonds are replaced by consols, which are bonds ha pay one dollar forever. I has curren marke

More information

Planning in POMDPs. Dominik Schoenberger Abstract

Planning in POMDPs. Dominik Schoenberger Abstract Planning in POMDPs Dominik Schoenberger d.schoenberger@sud.u-darmsad.de Absrac This documen briefly explains wha a Parially Observable Markov Decision Process is. Furhermore i inroduces he differen approaches

More information