The best expert versus the smartest algorithm

Size: px
Start display at page:

Download "The best expert versus the smartest algorithm"

Transcription

1 Theoretical Computer Science The best expert versus the smartest algorithm Peter Chen a, Guoli Ding b; a Department of Computer Science, Louisiana State University, Baton Rouge, LA, 70803, USA b Department of Mathematics, Louisiana State University, Baton Rouge, LA, 70803, USA Abstract In this paper, we consider the problem of online prediction using expert advice. Under dierent assumptions, we give tight lower bounds on the gap between the best expert and any online algorithm that solves the problem. c 004 Elsevier B.V. All rights reserved. Keywords: Online algorithm; Online prediction; Expert advice 1. Introduction The problem of online prediction using expert advice is for a predictor to predict, along with n other experts, a sequence 1 ; ;:::; {0; 1}. Here, the only assumptions we make are the following. Assumption 1.1. Before predicting each j, the predictor knows the predictions of the experts on this term. Also, right after predicting each j, the predictor is given the true value of this term. Notice that we do not assume anything on possible patterns of either or the sequences of the predictions of the experts. The goal of the predictor is to score as close to the best expert as possible. We point out that this is dierent from the goal that tries to predict as accurate as possible, which is a problem studied in [1]. Next, we make the problem more precise. Suppose x x 1 ;x ;:::;x is a sequence of predictions, made by the predictor or by the experts. It is worth mentioning that, Corresponding author. address: gding1@lsu.edu G. Ding /$ - see front matter c 004 Elsevier B.V. All rights reserved. doi: /j.tcs

2 36 P. Chen, G. Ding / TheoreticalComputer Science sometimes, terms of x might be allowed to take any value in the interval [0; 1]. The loss of x is dened to be Lx; x j j : j1 Let i be the sequence of predictions made by expert i and let Then { i :16 i 6 n}: L ; min{l i ;:16 i 6 n} is the loss of the best expert. For any strategy A of the predictor, 1 let A ; bethe sequence of predictions generated according to A. We measure the performance of A by the worst gap between the losses of the predictor and the best expert, i.e., by G A n; sup L A ; ; L ; : ; Clearly, the goal of the predictor is to minimize G A n; over all strategies A. In this paper, we analyze upper and lower bounds of G A n;. To get a lower bound, let us make the following assumption, which is more in favor of the predictor, and thus will make the result stronger: Assumption 1.. The predictions of the predictor can be any real number in the interval [0; 1], while the predictions of the experts can only be 0 or 1. In addition, before predicting 1, the predictor knows not only, but also, the entire prediction sequence of each expert. In other words, other than the actual value of each j, the predictor knows everything else before predicting 1. Under Assumption 1., it is proved in [] that, for all strategies A, lim inf n lim inf G A n; ln n 1: 1.1 We should point out that, because of the order the two limits are taken, it is assumed, implicitly, in the above inequality that is signicantly larger than n. We will see later that the situation is quite dierent if is smaller than n. For upper bounds, let us make a dierent assumption, which is less in favor of the predictor, and thus will make the result stronger. 1 Here we assume that, when predicting two sequences and,if and turn out to be the same, and the predictions of the experts are also the same, then the strategy should generate two identical sequences of predictions. For a strategy that generates two dierent sequences of predictions, due to randomization or similar reasons, we will consider it as several strategies under our term and this change does not aect our discussions.

3 P. Chen, G. Ding / TheoreticalComputer Science Assumption 1.3. The predictions of the predictor and the experts can be any real number in [0; 1]. The predictor also knows before predicting 1. Under Assumption 1.3, an online algorithm a strategy for the predictor A is given in [] for which lim inf n lim inf G A n; ln n 6 1: 1. In fact, what has been proved is that, for all positive integers n and, the algorithm A satises lnn +1 G A n; 6 + log n +1 : 1.3 Notice that 1.3 is much stronger than 1. since it upper bounds G A n; for all n and. Having such a bound is important because very often, in various applications, n and are not arbitrarily large. In this paper, we improve lower bound 1.1 inthe same way. Theorem 1.1. Under Assumption 1., for any algorithm A, any integer n, and any [0; 1], if n : 8 ln n + 8 lnn +1n 1 ; 1.4 then 1 ln n G A n; 1 1 n 1 n ; 1.5 where 1 lnn ln n +8 : +1 n 1 4 First, as easily shown below, 1.1 is a consequence of Theorem 1.1, and thus our theorem is indeed an improvement of 1.1 in the sense that our result implies 1.1 yet it is not implied by 1.1. Corollary 1.1. Inequality 1.1 holds for all online prediction algorithms A. Proof. By setting lnnn 1, it is straightforward to verify that lim inf n lim inf G A n; ln n

4 364 P. Chen, G. Ding / TheoreticalComputer Science lim inf n lim inf n 1 ; lim inf 1 1 ln n 1 n 1 n 1 1 n n holds for all 0; 1, and so 1.1 follows. Further remarks on Theorem 1.1. a If n 1, it is easy to see that the algorithm that copies the only expert will perform exactly the same as the best expert and thus G A n; 0, for all. Because of this, we can say that the assumption n in the theorem does not miss any interesting cases. b Inequality 1.5 still holds if we set 0. We introduce this extra parameter because we need it in proving Corollary 1.1. c For any 0, it is clear that nn 0, as n. Therefore, the requirement n is more or less the same as n. c One may wonder if the requirement n can be dropped completely. For instance, one may ask if there could exist a constant c 0, a function dn; with lim n dn; 0, and such that ln n G A n; c + dn; 1.6 holds for all n,, and A. Unfortunately, the answer is negative. Consider the strategy A that predicts 1 all the time. Then L A; ;, for all and. It follows that G A n; sup L A ; ; L ; ; inf ; L ; 6 : 1.7 Clearly, this inequality contradicts 1.6, for every 0, when n is suciently large. This contradiction indicates that a condition similar to n is required to prove any lower bound of the form 1.6. So far we have discussed the situation when is bigger than n. When is smaller than n, we have seen from remark d that lower bounds 1.1 and 1.5 no longer hold. Moreover, as indicated by our next result, that upper bounds 1. and 1.3 are not very close to the truth either. Theorem 1.. For all n,, and A, under Assumption 1., we have G A n; 1 1 n n :

5 P. Chen, G. Ding / TheoreticalComputer Science By combining this result with 1.7 we obviously have the following. Corollary 1.. For all, under Assumption 1., lim n inf A G An; : This result suggests that, if n is signicantly larger than, then the predictor cannot catch up with the best expert. The only thing the predictor can do is to predict 1 all the time so that it would not be left too far behind the best expert. Remaining questions. We have seen the behavior of G A n; when n and are far apart. But the situation is not very clear when n and are very close. For instance, it would be very interesting to know the asymptotic behavior of inf A G A ;, or the exactly value of inf A G A n; n. We closed this section by giving an outline of the rest of the paper. In Section, we present all necessary mathematical formulas. Then we prove Theorem 1. in Section 3. The proof of Theorem 1.1 will be given in the last section.. Preliminaries In this section, we present all necessary mathematical tools needed for the rest of the paper. Our rst lemma is well known and can be found in many Calculus books. Lemma.1 Stirling formula. For each positive integer m, m! m m m e m ; e where m satises 0 m 1 1m : The next lemma consists of some basic facts about normal distribution, which can also be found in most probability books. Lemma.. Let z 1 z e x dx. Then i 0 1 ; ii z1 z, for all z 0; and iii z 1 z+1 e z, for all z 0. Lemma.3. If n 1, then xx n +1 x n is decreasing over the interval [ 0; 1 ].

6 366 P. Chen, G. Ding / TheoreticalComputer Science Proof. Since n 1 and x 1 0;, it is clear that x nx n 1 1 x n 1 0 and thus the lemma follows. Lemma.4. For every x in [ 1 0; ], we have i ln1 + x x x + x3 3 x4 4, and ii ln1 x x x x3 3 x4. Proof. Notice that both inequalities are equalities when x 0. In addition, ln1 + x x + x x3 3 + x x 1+x x + x 3 x4 1+x 0 holds for all x 0, and thus i follows. Similarly, ln1 x+x + x + x3 3 + x4 1 1 x +1+x + x +x 3 x3 1 x 1 x holds for all x 0; 1, and thus ii follows. 0 Lemma.5. Let x and y be nonnegative numbers. Then e x y e y x. Proof. Notice that the inequality is an equality when x 0. Moreover, when taking derivative with respect to x we have e x y e y + x e x y +1 0: Therefore, the result follows. Lemma.6. Let fx be a function dened on the interval [c; c + ]. i If f x 0 on c; c +, then c+ c fx fc + 1 dx 0; ii If f x 0 on c; c +, then c+1 c fx fc dx + c+ c+1 Proof. If f x 0, then fx is convex and thus 1 fx 1+ 1 fx x1 + x f fx fc + dx 0:

7 P. Chen, G. Ding / TheoreticalComputer Science holds for all x 1 ;x in [c; c + ]. Consequently, c+ c fx fc + 1 dx c+1 c c+1 c c+1 c 0; fx fc + 1 dx + fx fc + 1 dx + c+ c+1 c+1 fx+fc + x fc + 1 dx c fx fc + 1 dx fc + x fc + 1 dx as required. On the other hand, if f x 0, then fx is concave and thus, for all x 1 ;x [c; c + ] and all [0; 1], we have and fx 1 +1 x fx 1 +1 fx f1 x 1 + x 1 fx 1 +fx : Let x x 1 +1 x. Then, adding the last two inequalities gives us fx+fx 1 + x x fx 1 +fx : Now, it is clear that c+1 c fx fc dx + c+1 c 0; c+ fx fc dx + and thus the lemma is proved. fx fc + dx c+1 c+1 c fc + x fc + dx Lemma.7. The inequality z 4 + z max{50z ; 48z 4 +1} holds for all z 0. Proof. It is easy to see that, for z 0, 48z z6 1 ; max{50z ; 48z 4 +1} 50z 1 6 z 6 ; 48z z ; where :908 and :157 are the only two positive solutions to the equation 50z 48z Therefore, to prove the Lemma, we only need to prove the following three claims.

8 368 P. Chen, G. Ding / TheoreticalComputer Science Claim 1. z z 4 +1 z, for z 1;. Claim. z z 4 +1 z, for z 0; 0:3. Claim 3. z z z, for z 0;. Since z is an increasing function and 0:075, it follows that: 48z 4 +1 z 6 48z :050:8z 4 +0:1 z 4 + and thus Claim 1 is proved. To prove Claim, let Then fz z4 + 48z 4 +1 z: 1+8z f 4 15z 3 e z z 1+8z 4 e z and f z z 4 1+8z z3 40z 4 e z 1+8z 4 3 e z : For each z 0; 0:3, it is easy to see that 3 40z 4 0, and so we have f z60. Consequently, f z is a decreasing function, which implies that fz min{f0;f0:3} f00 for all z 0; 0:3, which proves Claim. We prove Claim 3 by considering three intervals, 0; 0:73, [0:73; 0:8], and 0:8;. If z 0:8;, we deduce from 0:8 0:0548 0:056 that z z z z 4 + :8z z 1:4 +0:4 0; and thus the desired inequality holds. Next, let gz z4 + 50z : Then gz is decreasing over [0:73; 0:8], since g z 0:04 z 4 z 3. Recall that z is an increasing function. Therefore, for z [0:73; 0:8], gz g0:80:0753 0:071 0:73 z; which is what we want. It remains to prove z z z, over 0; 0:73. In the remainder of this proof, we assume that z belongs to 0; 0:73. Let h 1 z z +z 50 z and h z z 5lnz:

9 P. Chen, G. Ding / TheoreticalComputer Science Then h 1z z 4z e z ; h 1 z+1z 4 400z e z and h z 5 4z : z Clearly, h z 0, which implies that h z is a decreasing function. Consequently, h z h 0:73 :63935 :5876 ln which implies z ln 400z5 1 : Therefore, h 1 z 1z 4 400z e z 1 z 4 e z ; e z 400z5 1 0; which implies that h 1 z is an increasing function. Notice that h 1 0:5 6:8 0 and h 1 0:6 :1 0. It follows that h 1 z 0 has a unique solution, say z 0, which means that h 1 z achieves its minimum at z 0. To complete our proof, we only need to show that h 1 z 0 0. For this, we apply Taylor s expansion theorem to h 1 z atz 0:56. We have h 1 z h 1 0:56 + h 10:56z 0: h 1 z 0:56 ; where is a number between z and 0:56. Since h 1 0:56 0:35 0, we conclude that z 0 is between 0:56 and 0:6. Now, from h 1 z 0 and h 10:56 0:13 we deduce that h 1 z 0 h 1 0:56 + h 10:56z 0 0:56 0:116 0:40:6 0:560:1 0; which competes our proof of the Lemma. The last is another technical lemma, which we will use in the last section. Lemma.8. Let and be positive numbers and let gx e x 1 1:16 + x: 1 x { Then gx60 if 06x6 min 1 48 ; 5 }.

10 370 P. Chen, G. Ding / TheoreticalComputer Science Proof. Notice that g0 0 and e x g x 1 1 x 3 x+ 1:16 6 e148 0: : Thus, the result follows. 1: Proving Theorem 1. It is clear from the denition of G A n; that, in order to prove a lower bound, we only need to nd one sequence and one set of sequences, for every A, such that L A ; ; L ; is greater than or equal to the lower bound. Since there are too many choices for A, it is dicult or may be impossible to choose and. The way to solve this problem is to use probabilistic method. Recall that, under Assumption 1., the predictions of the experts must be 0 or 1. Suppose each expert makes prediction by tossing, independently, a fair coin. In addition, suppose the outcome is also determined by tossing a fair coin. In the following, we establish a lower bound on the expected gap between the losses of any A and the best expert. Then we prove Theorem 1. using this bound. For each i 1; ;:::;n and j 1; ;:::;, let X i; j be the random variable such that X i; j 0, if the prediction of expert i on j is correct, and X i; j 1, if this prediction is wrong. Then it is not dicult to verify that X {X i;j :16 i 6 n and 1 6 j 6 } is a set of mutually independent random variables with Px i;j 0PX i;j 1 1 : For each i 1; ;:::;n; let S i X i; 1 + X i; + + X i;, which is the number of mistakes made by expert i. It follows that the number of mistakes made by the best expert is S min{s 1 ;S ;:::;S n }. Lemma 3.1. For all j 0; 1;:::;, let p j p j j. Then n E[S] p j : i1 ji Proof. If x is a nonnegative integer not exceeding, it is clear that n PS x PS 1 x; S x; ; S n x n PS i x p j i1 jx

11 and so, Therefore, P. Chen, G. Ding / TheoreticalComputer Science PS x PS x PS x +1 n E[S] x p j x0 x x1 x1 jx jx jx n p j 1 x0 x1 jx+1 x p j n jx+1 p j n n p j jx n n x p j x 1 p j n p j x1 jx and the lemma is proved. jx jx+1 p j n: Lemma 3.. Under Assumption 1., for all n,, and A, we have G A n; Gn;, where n Gn; p j : 3.1 i1 ji Proof. For each j 1; ;:::;, let Y j be the outcome of j. Here, we assume that Y {Y j : j 1; ;:::;} is a set of mutually independent random variables with PY j 0 PY j 1 1. For any algorithm A, let AY; X 1 ; ;:::;. Notice that, for each j, E[ j Y j ] j + 1 j 1 : Therefore, E[L A Y; X ; Y]. Also notice that E[LX ; Y] ES, so, from Lemma 3.1, we deduce that E[L A Y; X ; Y LX ; Y] Gn;. Consequently, for some special values of Y and X, say and, we must have L A ; ; L ; Gn;, and thus the lemma follows. For each i 1; ;:::;+ 1, let P i P i i 1 p j : 3. j0

12 37 P. Chen, G. Ding / TheoreticalComputer Science Lemma 3.3. For all n and, we have where Gn; n; i1 1 P i n + P n i n; ; { 0 if is even; 1 n if is odd: Proof. From 3.1 it is clear that n Gn; p j i1 i1 ji n 1 i 1 p j j0 i +1 n p j n; ji i +1 n p j n; : 3.3 Notice that, by setting i +1 i, j j, and using p k p k, we have n n p j i +1 ji i 1 i 1 i1 p j j+1 i n i 1 p j j 0 i 1 j0 p j n and thus the lemma follows from 3.3 and 3.. Proof of Theorem 1.. In is even, by Lemmas 3., 3.3 and.3, we have G A n; i1 1 P 1 n + P n 1 ji 1 1 n n 1 + and the result follows. If is odd, a little extra eort is needed. When 1,itis straightforward to verify from 3.1 that Gn; 1 1 n n n 1 and thus the theorem holds in this case. When 3, again, using Lemmas 3., 3.3 and.3, we have G A n; 1 1 P 1 n + P1 n n; i1

13 P. Chen, G. Ding / TheoreticalComputer Science The theorem is proved. 1 1 n n n 1 1 n + 1 n : n 1 n n + n 1 8 n 1 4. Proving Theorem 1.1 We prove the theorem by proving a sequence of lemmas. Our starting point is Lemma 3.. In the rest of the paper, let z be a positive number and let 1 a z : 4.1 Lemma 4.1. For every positive integer, we have Gn; a 1 P a+1 ; where is the function dened in Lemma.3. Proof. It follows from Lemmas 3.3 and.3 that Gn; P i n; i1 a P i i1 ia+1 P i n; a ap a+1 n; : 4. The lemma clearly holds if is even. When is odd, using the fact that n; 1, we deduce from 4. and Lemma.3 that Gn; 1 a a P a a 1 P a P a+1 a 1 P a+1 : The lemma is proved.

14 374 P. Chen, G. Ding / TheoreticalComputer Science Next, we need to approximate each p j and then P a+1. Lemma 4.. Let j6 be a nonnegative integer and let h j. If 4h6, then h exp 1 p j 6 exp 1 + h4 3 3 ; 1 h Proof. By Stirling s formula, we have! p j j! j! 6 e e 11 j j e j j j e j 1 j j e 11 : 4.3 j j j j First, it is clear that j j h + h 1 : h Then, when 4h6, we deduce from Lemma.4 that j ln j j ln j 1 h j jln 6 j h h 8h h4 4 h h + h 8h3 + h h + h h h j 3 h h h + 8h h4 4 4h 4 6 h + h and thus the lemma follows from Lemma 4.3. Suppose z 6j6 and 16z. Then p j e j ; 4

15 P. Chen, G. Ding / TheoreticalComputer Science where 8z exp z Proof. Let h j. Then we have 4h64z 16z 6 and so Lemma 4. applies. Notice that h j, and exp h exp h 1 z +1: 1 + z 4 Thus, the result follows from Lemma 4.. Lemma 4.4. If 16z, then P a z +3 3 : Proof. Let x 1 e x and let b a 1. We rst consider the case when is even. Since p j 1 j0 and p j p j, for every j 0; 1;:::;, it follows that j0 p j + 1 p 1 : By 3. and Lemma 4.3, we have P a+1 a j0 p j 1 1 p p j ja b j1 ja+1 j j : 4.6

16 376 P. Chen, G. Ding / TheoreticalComputer Science Observe that, by Lemma. and 4.1, 1 0 b j1 j b +1 + z+ b+1 0 b+1 0 xdx 0 b j xdx 0 b j: 4.7 Let k be the largest even integer not exceeding. Then b+1 A : 0 k 1 j0 j k xdx 0 b j j1 j+1 x j dx + k k+3 k+ x k dx + j+ j+1 k+ k+1 x k + dx + b 1 jk+1 j1 j1 x j + dx x k + dx j+3 j+1 x j + dx: 4.8 It is straightforward to verify that x 0 over 0;, and x 0 over ;. Thus, we conclude from Lemma.6 and the fact that x is decreasing on 0; that A k+1 k k+ k+1 k+3 k+ k +1 k dx k + k + dx k +3 k + dx +0 k +1 k+k +3 k + 1 e 4k 4k+1 e 4k + e k+ 4k+5 e k+ : 4.9 By Lemma.5, it follows that A 1 4k +1 4k +5 and thus the lemma follows from k

17 P. Chen, G. Ding / TheoreticalComputer Science The situation for odd is similar. First, P a+1 a p j j0 1 1 ja+1 p j ja+1 j b j 1 j b j 1 j1 b 1 + z+ xdx b Let k + 1 be the largest odd integer not exceeding. Then B : b k xdx b j 1 j1 x 1 dx j+ x j + 1 dx + j0 j+1 k+ k+1 k+4 k+3 x k + 1 dx + 0 k+3 x k + 3 dx + b 1 j1 j+3 j 1 : 4.10 j+ k+ i+ jk+ x j + 3 dx x k + 3 dx k + k +1+k +4 k + 3 4k +5 3 j x j + 1 dx +3 3 : 4.11 Therefore, by 4.10 and 4.11, the lemma also holds for odd t. The proof is complete. Lemma 4.5. Suppose max{5; z 4 + z }. Then P a+1 z z4 +.

18 378 P. Chen, G. Ding / TheoreticalComputer Science Proof. By Lemma.7, we have 16z. Also observe that z z : 3 Thus the lower bound in Lemma 4.4 can be simplied as P a+1 z +3 3 : 4.1 By taking 8z 4 +11, 4z, and x 1, we deduce from Lemmas.8 and.7 that 8z :16 +z 1 : Therefore, and thus the lemma follows from :16 3 z4 +1:16z : z4 + ; Proof of Theorem 1.1. We rst prove that the assumption n implies max {5; z 4 + z}, if we take 1 lnn z : Notice that 8ln + 8 ln+1 11:6 and thus 5 holds. Furthermore, by Lemma., we have z 4 + z 6 z4 + z + 1e z 1 lnn + 1 lnn +1n 1 6 n: Therefore, the assumptions in Lemma 4.5 are satised. Consequently, by Lemma. again, we have P a+1 z z4 + e z z +1 z4 +

19 P. Chen, G. Ding / TheoreticalComputer Science lnn + lnn +1 n 1 : 4.13 In addition, from 1.4 it is clear that 0. Thus, we deduce from Lemma 4.1 and Lemma.3 that Gn; The proof is complete. a 1 z 1 1 : Appendix To illustrate our results, we plot the following curves for n and 6100: The dotted curve is V n; inf A G A n;, which is the function that we are trying to approximate. The top curve is lnn +1 + log n +1 ; an upper bound of V n;, which is obtained in []. The bottom curve is our lower bound given in Theorem 1.1 and the fourth curve is Gn;, which, by our Lemma 3., is another lower bound of V n; Fig Fig. 1. An illustration of our results.

20 380 P. Chen, G. Ding / TheoreticalComputer Science Acknowledgements The authors are grateful to their friend Steve Seiden, who introduced this problem to them. They wish to acknowledge that they have beneted from discussions with Steve. The authors would also like to thank the two anonymous referees for their valuable suggestions. This Research is partially supported by NSF Grant DMS , AFOSR Grant F , and NSF Grant ITR References [1] N. Cesa-Bianchi, Y. Freund, D.P. Helmbold, M. Warmuth, On-line prediction and conversion strategies, Mach. Learning [] N. Cesa-Bianchi, Y. Freund, D. Haussler, D.P. Helmbold, R.E. Schapire, How to use expert advice, J. Assoc. Comput. Mach

Balance properties of multi-dimensional words

Balance properties of multi-dimensional words Theoretical Computer Science 273 (2002) 197 224 www.elsevier.com/locate/tcs Balance properties of multi-dimensional words Valerie Berthe a;, Robert Tijdeman b a Institut de Mathematiques de Luminy, CNRS-UPR

More information

f(z)dz = 0. P dx + Qdy = D u dx v dy + i u dy + v dx. dxdy + i x = v

f(z)dz = 0. P dx + Qdy = D u dx v dy + i u dy + v dx. dxdy + i x = v MA525 ON CAUCHY'S THEOREM AND GREEN'S THEOREM DAVID DRASIN (EDITED BY JOSIAH YODER) 1. Introduction No doubt the most important result in this course is Cauchy's theorem. Every critical theorem in the

More information

Lecture 21. Hypothesis Testing II

Lecture 21. Hypothesis Testing II Lecture 21. Hypothesis Testing II December 7, 2011 In the previous lecture, we dened a few key concepts of hypothesis testing and introduced the framework for parametric hypothesis testing. In the parametric

More information

Department of Mathematics, University of California, Berkeley. GRADUATE PRELIMINARY EXAMINATION, Part A Fall Semester 2014

Department of Mathematics, University of California, Berkeley. GRADUATE PRELIMINARY EXAMINATION, Part A Fall Semester 2014 Department of Mathematics, University of California, Berkeley YOUR 1 OR 2 DIGIT EXAM NUMBER GRADUATE PRELIMINARY EXAMINATION, Part A Fall Semester 2014 1. Please write your 1- or 2-digit exam number on

More information

Induced subgraphs with many repeated degrees

Induced subgraphs with many repeated degrees Induced subgraphs with many repeated degrees Yair Caro Raphael Yuster arxiv:1811.071v1 [math.co] 17 Nov 018 Abstract Erdős, Fajtlowicz and Staton asked for the least integer f(k such that every graph with

More information

Online Bounds for Bayesian Algorithms

Online Bounds for Bayesian Algorithms Online Bounds for Bayesian Algorithms Sham M. Kakade Computer and Information Science Department University of Pennsylvania Andrew Y. Ng Computer Science Department Stanford University Abstract We present

More information

Exam 3, Math Fall 2016 October 19, 2016

Exam 3, Math Fall 2016 October 19, 2016 Exam 3, Math 500- Fall 06 October 9, 06 This is a 50-minute exam. You may use your textbook, as well as a calculator, but your work must be completely yours. The exam is made of 5 questions in 5 pages,

More information

4 Uniform convergence

4 Uniform convergence 4 Uniform convergence In the last few sections we have seen several functions which have been defined via series or integrals. We now want to develop tools that will allow us to show that these functions

More information

Principle of Mathematical Induction

Principle of Mathematical Induction Advanced Calculus I. Math 451, Fall 2016, Prof. Vershynin Principle of Mathematical Induction 1. Prove that 1 + 2 + + n = 1 n(n + 1) for all n N. 2 2. Prove that 1 2 + 2 2 + + n 2 = 1 n(n + 1)(2n + 1)

More information

Worst-Case Analysis of the Perceptron and Exponentiated Update Algorithms

Worst-Case Analysis of the Perceptron and Exponentiated Update Algorithms Worst-Case Analysis of the Perceptron and Exponentiated Update Algorithms Tom Bylander Division of Computer Science The University of Texas at San Antonio San Antonio, Texas 7849 bylander@cs.utsa.edu April

More information

8. Limit Laws. lim(f g)(x) = lim f(x) lim g(x), (x) = lim x a f(x) g lim x a g(x)

8. Limit Laws. lim(f g)(x) = lim f(x) lim g(x), (x) = lim x a f(x) g lim x a g(x) 8. Limit Laws 8.1. Basic Limit Laws. If f and g are two functions and we know the it of each of them at a given point a, then we can easily compute the it at a of their sum, difference, product, constant

More information

On reaching head-to-tail ratios for balanced and unbalanced coins

On reaching head-to-tail ratios for balanced and unbalanced coins Journal of Statistical Planning and Inference 0 (00) 0 0 www.elsevier.com/locate/jspi On reaching head-to-tail ratios for balanced and unbalanced coins Tamas Lengyel Department of Mathematics, Occidental

More information

Upper Bounds for Partitions into k-th Powers Elementary Methods

Upper Bounds for Partitions into k-th Powers Elementary Methods Int. J. Contemp. Math. Sciences, Vol. 4, 2009, no. 9, 433-438 Upper Bounds for Partitions into -th Powers Elementary Methods Rafael Jaimczu División Matemática, Universidad Nacional de Luján Buenos Aires,

More information

Fourier Series. 1. Review of Linear Algebra

Fourier Series. 1. Review of Linear Algebra Fourier Series In this section we give a short introduction to Fourier Analysis. If you are interested in Fourier analysis and would like to know more detail, I highly recommend the following book: Fourier

More information

16.4. Power Series. Introduction. Prerequisites. Learning Outcomes

16.4. Power Series. Introduction. Prerequisites. Learning Outcomes Power Series 6.4 Introduction In this Section we consider power series. These are examples of infinite series where each term contains a variable, x, raised to a positive integer power. We use the ratio

More information

Monte Carlo Methods for Statistical Inference: Variance Reduction Techniques

Monte Carlo Methods for Statistical Inference: Variance Reduction Techniques Monte Carlo Methods for Statistical Inference: Variance Reduction Techniques Hung Chen hchen@math.ntu.edu.tw Department of Mathematics National Taiwan University 3rd March 2004 Meet at NS 104 On Wednesday

More information

Learning Methods for Online Prediction Problems. Peter Bartlett Statistics and EECS UC Berkeley

Learning Methods for Online Prediction Problems. Peter Bartlett Statistics and EECS UC Berkeley Learning Methods for Online Prediction Problems Peter Bartlett Statistics and EECS UC Berkeley Course Synopsis A finite comparison class: A = {1,..., m}. Converting online to batch. Online convex optimization.

More information

On a kind of restricted edge connectivity of graphs

On a kind of restricted edge connectivity of graphs Discrete Applied Mathematics 117 (2002) 183 193 On a kind of restricted edge connectivity of graphs Jixiang Meng a; ;1, Youhu Ji b a Department of Mathematics, Xinjiang University, Urumqi 830046, Xinjiang,

More information

April 25 May 6, 2016, Verona, Italy. GAME THEORY and APPLICATIONS Mikhail Ivanov Krastanov

April 25 May 6, 2016, Verona, Italy. GAME THEORY and APPLICATIONS Mikhail Ivanov Krastanov April 25 May 6, 2016, Verona, Italy GAME THEORY and APPLICATIONS Mikhail Ivanov Krastanov Games in normal form There are given n-players. The set of all strategies (possible actions) of the i-th player

More information

Sequential prediction with coded side information under logarithmic loss

Sequential prediction with coded side information under logarithmic loss under logarithmic loss Yanina Shkel Department of Electrical Engineering Princeton University Princeton, NJ 08544, USA Maxim Raginsky Department of Electrical and Computer Engineering Coordinated Science

More information

1 Question related to polynomials

1 Question related to polynomials 07-08 MATH00J Lecture 6: Taylor Series Charles Li Warning: Skip the material involving the estimation of error term Reference: APEX Calculus This lecture introduced Taylor Polynomial and Taylor Series

More information

4.1 Eigenvalues, Eigenvectors, and The Characteristic Polynomial

4.1 Eigenvalues, Eigenvectors, and The Characteristic Polynomial Linear Algebra (part 4): Eigenvalues, Diagonalization, and the Jordan Form (by Evan Dummit, 27, v ) Contents 4 Eigenvalues, Diagonalization, and the Jordan Canonical Form 4 Eigenvalues, Eigenvectors, and

More information

Taylor and Maclaurin Series

Taylor and Maclaurin Series Taylor and Maclaurin Series MATH 211, Calculus II J. Robert Buchanan Department of Mathematics Spring 2018 Background We have seen that some power series converge. When they do, we can think of them as

More information

1 More concise proof of part (a) of the monotone convergence theorem.

1 More concise proof of part (a) of the monotone convergence theorem. Math 0450 Honors intro to analysis Spring, 009 More concise proof of part (a) of the monotone convergence theorem. Theorem If (x n ) is a monotone and bounded sequence, then lim (x n ) exists. Proof. (a)

More information

Agnostic Online learnability

Agnostic Online learnability Technical Report TTIC-TR-2008-2 October 2008 Agnostic Online learnability Shai Shalev-Shwartz Toyota Technological Institute Chicago shai@tti-c.org ABSTRACT We study a fundamental question. What classes

More information

International Competition in Mathematics for Universtiy Students in Plovdiv, Bulgaria 1994

International Competition in Mathematics for Universtiy Students in Plovdiv, Bulgaria 1994 International Competition in Mathematics for Universtiy Students in Plovdiv, Bulgaria 1994 1 PROBLEMS AND SOLUTIONS First day July 29, 1994 Problem 1. 13 points a Let A be a n n, n 2, symmetric, invertible

More information

Online prediction with expert advise

Online prediction with expert advise Online prediction with expert advise Jyrki Kivinen Australian National University http://axiom.anu.edu.au/~kivinen Contents 1. Online prediction: introductory example, basic setting 2. Classification with

More information

18.175: Lecture 17 Poisson random variables

18.175: Lecture 17 Poisson random variables 18.175: Lecture 17 Poisson random variables Scott Sheffield MIT 1 Outline More on random walks and local CLT Poisson random variable convergence Extend CLT idea to stable random variables 2 Outline More

More information

Constrained Leja points and the numerical solution of the constrained energy problem

Constrained Leja points and the numerical solution of the constrained energy problem Journal of Computational and Applied Mathematics 131 (2001) 427 444 www.elsevier.nl/locate/cam Constrained Leja points and the numerical solution of the constrained energy problem Dan I. Coroian, Peter

More information

1 Introduction It will be convenient to use the inx operators a b and a b to stand for maximum (least upper bound) and minimum (greatest lower bound)

1 Introduction It will be convenient to use the inx operators a b and a b to stand for maximum (least upper bound) and minimum (greatest lower bound) Cycle times and xed points of min-max functions Jeremy Gunawardena, Department of Computer Science, Stanford University, Stanford, CA 94305, USA. jeremy@cs.stanford.edu October 11, 1993 to appear in the

More information

Lecture 1: Overview of percolation and foundational results from probability theory 30th July, 2nd August and 6th August 2007

Lecture 1: Overview of percolation and foundational results from probability theory 30th July, 2nd August and 6th August 2007 CSL866: Percolation and Random Graphs IIT Delhi Arzad Kherani Scribe: Amitabha Bagchi Lecture 1: Overview of percolation and foundational results from probability theory 30th July, 2nd August and 6th August

More information

58 Appendix 1 fundamental inconsistent equation (1) can be obtained as a linear combination of the two equations in (2). This clearly implies that the

58 Appendix 1 fundamental inconsistent equation (1) can be obtained as a linear combination of the two equations in (2). This clearly implies that the Appendix PRELIMINARIES 1. THEOREMS OF ALTERNATIVES FOR SYSTEMS OF LINEAR CONSTRAINTS Here we consider systems of linear constraints, consisting of equations or inequalities or both. A feasible solution

More information

Exercises. Template for Proofs by Mathematical Induction

Exercises. Template for Proofs by Mathematical Induction 5. Mathematical Induction 329 Template for Proofs by Mathematical Induction. Express the statement that is to be proved in the form for all n b, P (n) forafixed integer b. 2. Write out the words Basis

More information

Solutions to Assignment-11

Solutions to Assignment-11 Solutions to Assignment-. (a) For any sequence { } of positive numbers, show that lim inf + lim inf an lim sup an lim sup +. Show from this, that the root test is stronger than the ratio test. That is,

More information

Part 2 Continuous functions and their properties

Part 2 Continuous functions and their properties Part 2 Continuous functions and their properties 2.1 Definition Definition A function f is continuous at a R if, and only if, that is lim f (x) = f (a), x a ε > 0, δ > 0, x, x a < δ f (x) f (a) < ε. Notice

More information

Universal Juggling Cycles

Universal Juggling Cycles Universal Juggling Cycles Fan Chung y Ron Graham z Abstract During the past several decades, it has become popular among jugglers (and juggling mathematicians) to represent certain periodic juggling patterns

More information

Sequences. Chapter 3. n + 1 3n + 2 sin n n. 3. lim (ln(n + 1) ln n) 1. lim. 2. lim. 4. lim (1 + n)1/n. Answers: 1. 1/3; 2. 0; 3. 0; 4. 1.

Sequences. Chapter 3. n + 1 3n + 2 sin n n. 3. lim (ln(n + 1) ln n) 1. lim. 2. lim. 4. lim (1 + n)1/n. Answers: 1. 1/3; 2. 0; 3. 0; 4. 1. Chapter 3 Sequences Both the main elements of calculus (differentiation and integration) require the notion of a limit. Sequences will play a central role when we work with limits. Definition 3.. A Sequence

More information

REVIEW FOR THIRD 3200 MIDTERM

REVIEW FOR THIRD 3200 MIDTERM REVIEW FOR THIRD 3200 MIDTERM PETE L. CLARK 1) Show that for all integers n 2 we have 1 3 +... + (n 1) 3 < 1 n < 1 3 +... + n 3. Solution: We go by induction on n. Base Case (n = 2): We have (2 1) 3 =

More information

On the number of ways of writing t as a product of factorials

On the number of ways of writing t as a product of factorials On the number of ways of writing t as a product of factorials Daniel M. Kane December 3, 005 Abstract Let N 0 denote the set of non-negative integers. In this paper we prove that lim sup n, m N 0 : n!m!

More information

Power series solutions for 2nd order linear ODE s (not necessarily with constant coefficients) a n z n. n=0

Power series solutions for 2nd order linear ODE s (not necessarily with constant coefficients) a n z n. n=0 Lecture 22 Power series solutions for 2nd order linear ODE s (not necessarily with constant coefficients) Recall a few facts about power series: a n z n This series in z is centered at z 0. Here z can

More information

ADJUSTED POWER ESTIMATES IN. Ji Zhang. Biostatistics and Research Data Systems. Merck Research Laboratories. Rahway, NJ

ADJUSTED POWER ESTIMATES IN. Ji Zhang. Biostatistics and Research Data Systems. Merck Research Laboratories. Rahway, NJ ADJUSTED POWER ESTIMATES IN MONTE CARLO EXPERIMENTS Ji Zhang Biostatistics and Research Data Systems Merck Research Laboratories Rahway, NJ 07065-0914 and Dennis D. Boos Department of Statistics, North

More information

ON STATISTICAL INFERENCE UNDER ASYMMETRIC LOSS. Abstract. We introduce a wide class of asymmetric loss functions and show how to obtain

ON STATISTICAL INFERENCE UNDER ASYMMETRIC LOSS. Abstract. We introduce a wide class of asymmetric loss functions and show how to obtain ON STATISTICAL INFERENCE UNDER ASYMMETRIC LOSS FUNCTIONS Michael Baron Received: Abstract We introduce a wide class of asymmetric loss functions and show how to obtain asymmetric-type optimal decision

More information

On the number of occurrences of all short factors in almost all words

On the number of occurrences of all short factors in almost all words Theoretical Computer Science 290 (2003) 2031 2035 www.elsevier.com/locate/tcs Note On the number of occurrences of all short factors in almost all words Ioan Tomescu Faculty of Mathematics, University

More information

Markov Sonin Gaussian rule for singular functions

Markov Sonin Gaussian rule for singular functions Journal of Computational and Applied Mathematics 169 (2004) 197 212 www.elsevier.com/locate/cam Markov Sonin Gaussian rule for singular functions G.Mastroianni, D.Occorsio Dipartimento di Matematica, Universita

More information

Contents. 2 Partial Derivatives. 2.1 Limits and Continuity. Calculus III (part 2): Partial Derivatives (by Evan Dummit, 2017, v. 2.

Contents. 2 Partial Derivatives. 2.1 Limits and Continuity. Calculus III (part 2): Partial Derivatives (by Evan Dummit, 2017, v. 2. Calculus III (part 2): Partial Derivatives (by Evan Dummit, 2017, v 260) Contents 2 Partial Derivatives 1 21 Limits and Continuity 1 22 Partial Derivatives 5 23 Directional Derivatives and the Gradient

More information

Discrete Applied Mathematics. Tighter bounds of the First Fit algorithm for the bin-packing problem

Discrete Applied Mathematics. Tighter bounds of the First Fit algorithm for the bin-packing problem Discrete Applied Mathematics 158 (010) 1668 1675 Contents lists available at ScienceDirect Discrete Applied Mathematics journal homepage: www.elsevier.com/locate/dam Tighter bounds of the First Fit algorithm

More information

Sharp Bounds for the Harmonic Numbers

Sharp Bounds for the Harmonic Numbers Sharp Bounds for the Harmonic Numbers arxiv:math/050585v3 [math.ca] 5 Nov 005 Mark B. Villarino Depto. de Matemática, Universidad de Costa Rica, 060 San José, Costa Rica March, 08 Abstract We obtain best

More information

AP Calculus Chapter 9: Infinite Series

AP Calculus Chapter 9: Infinite Series AP Calculus Chapter 9: Infinite Series 9. Sequences a, a 2, a 3, a 4, a 5,... Sequence: A function whose domain is the set of positive integers n = 2 3 4 a n = a a 2 a 3 a 4 terms of the sequence Begin

More information

FINAL EXAM Math 25 Temple-F06

FINAL EXAM Math 25 Temple-F06 FINAL EXAM Math 25 Temple-F06 Write solutions on the paper provided. Put your name on this exam sheet, and staple it to the front of your finished exam. Do Not Write On This Exam Sheet. Problem 1. (Short

More information

The average dimension of the hull of cyclic codes

The average dimension of the hull of cyclic codes Discrete Applied Mathematics 128 (2003) 275 292 www.elsevier.com/locate/dam The average dimension of the hull of cyclic codes Gintaras Skersys Matematikos ir Informatikos Fakultetas, Vilniaus Universitetas,

More information

On the Largest Integer that is not a Sum of Distinct Positive nth Powers

On the Largest Integer that is not a Sum of Distinct Positive nth Powers On the Largest Integer that is not a Sum of Distinct Positive nth Powers arxiv:1610.02439v4 [math.nt] 9 Jul 2017 Doyon Kim Department of Mathematics and Statistics Auburn University Auburn, AL 36849 USA

More information

DR.RUPNATHJI( DR.RUPAK NATH )

DR.RUPNATHJI( DR.RUPAK NATH ) Contents 1 Sets 1 2 The Real Numbers 9 3 Sequences 29 4 Series 59 5 Functions 81 6 Power Series 105 7 The elementary functions 111 Chapter 1 Sets It is very convenient to introduce some notation and terminology

More information

Circuit depth relative to a random oracle. Peter Bro Miltersen. Aarhus University, Computer Science Department

Circuit depth relative to a random oracle. Peter Bro Miltersen. Aarhus University, Computer Science Department Circuit depth relative to a random oracle Peter Bro Miltersen Aarhus University, Computer Science Department Ny Munkegade, DK 8000 Aarhus C, Denmark. pbmiltersen@daimi.aau.dk Keywords: Computational complexity,

More information

Complex Analysis Slide 9: Power Series

Complex Analysis Slide 9: Power Series Complex Analysis Slide 9: Power Series MA201 Mathematics III Department of Mathematics IIT Guwahati August 2015 Complex Analysis Slide 9: Power Series 1 / 37 Learning Outcome of this Lecture We learn Sequence

More information

Adaptive Online Gradient Descent

Adaptive Online Gradient Descent University of Pennsylvania ScholarlyCommons Statistics Papers Wharton Faculty Research 6-4-2007 Adaptive Online Gradient Descent Peter Bartlett Elad Hazan Alexander Rakhlin University of Pennsylvania Follow

More information

A characterization of consistency of model weights given partial information in normal linear models

A characterization of consistency of model weights given partial information in normal linear models Statistics & Probability Letters ( ) A characterization of consistency of model weights given partial information in normal linear models Hubert Wong a;, Bertrand Clare b;1 a Department of Health Care

More information

means is a subset of. So we say A B for sets A and B if x A we have x B holds. BY CONTRAST, a S means that a is a member of S.

means is a subset of. So we say A B for sets A and B if x A we have x B holds. BY CONTRAST, a S means that a is a member of S. 1 Notation For those unfamiliar, we have := means equal by definition, N := {0, 1,... } or {1, 2,... } depending on context. (i.e. N is the set or collection of counting numbers.) In addition, means for

More information

e x = 1 + x + x2 2! + x3 If the function f(x) can be written as a power series on an interval I, then the power series is of the form

e x = 1 + x + x2 2! + x3 If the function f(x) can be written as a power series on an interval I, then the power series is of the form Taylor Series Given a function f(x), we would like to be able to find a power series that represents the function. For example, in the last section we noted that we can represent e x by the power series

More information

Numerical Integration for Multivariable. October Abstract. We consider the numerical integration of functions with point singularities over

Numerical Integration for Multivariable. October Abstract. We consider the numerical integration of functions with point singularities over Numerical Integration for Multivariable Functions with Point Singularities Yaun Yang and Kendall E. Atkinson y October 199 Abstract We consider the numerical integration of functions with point singularities

More information

Lecture 32: Taylor Series and McLaurin series We saw last day that some functions are equal to a power series on part of their domain.

Lecture 32: Taylor Series and McLaurin series We saw last day that some functions are equal to a power series on part of their domain. Lecture 32: Taylor Series and McLaurin series We saw last day that some functions are equal to a power series on part of their domain. For example f(x) = 1 1 x = 1 + x + x2 + x 3 + = ln(1 + x) = x x2 2

More information

Estimates for probabilities of independent events and infinite series

Estimates for probabilities of independent events and infinite series Estimates for probabilities of independent events and infinite series Jürgen Grahl and Shahar evo September 9, 06 arxiv:609.0894v [math.pr] 8 Sep 06 Abstract This paper deals with finite or infinite sequences

More information

Comments on integral variants of ISS 1

Comments on integral variants of ISS 1 Systems & Control Letters 34 (1998) 93 1 Comments on integral variants of ISS 1 Eduardo D. Sontag Department of Mathematics, Rutgers University, Piscataway, NJ 8854-819, USA Received 2 June 1997; received

More information

Rademacher functions

Rademacher functions Rademacher functions Jordan Bell jordan.bell@gmail.com Department of Mathematics, University of Toronto July 6, 4 Binary expansions Define S : {, } N [, ] by Sσ σ k k, σ {, }N. For example, for σ and σ,

More information

Lecture 4: Probability, Proof Techniques, Method of Induction Lecturer: Lale Özkahya

Lecture 4: Probability, Proof Techniques, Method of Induction Lecturer: Lale Özkahya BBM 205 Discrete Mathematics Hacettepe University http://web.cs.hacettepe.edu.tr/ bbm205 Lecture 4: Probability, Proof Techniques, Method of Induction Lecturer: Lale Özkahya Resources: Kenneth Rosen, Discrete

More information

Essentials of Intermediate Algebra

Essentials of Intermediate Algebra Essentials of Intermediate Algebra BY Tom K. Kim, Ph.D. Peninsula College, WA Randy Anderson, M.S. Peninsula College, WA 9/24/2012 Contents 1 Review 1 2 Rules of Exponents 2 2.1 Multiplying Two Exponentials

More information

Lecture 6: The Pigeonhole Principle and Probability Spaces

Lecture 6: The Pigeonhole Principle and Probability Spaces Lecture 6: The Pigeonhole Principle and Probability Spaces Anup Rao January 17, 2018 We discuss the pigeonhole principle and probability spaces. Pigeonhole Principle The pigeonhole principle is an extremely

More information

MCS 256 Discrete Calculus and Probability Exam 5: Final Examination 22 May 2007

MCS 256 Discrete Calculus and Probability Exam 5: Final Examination 22 May 2007 MCS 256 Discrete Calculus and Probability SOLUTIONS Exam 5: Final Examination 22 May 2007 Instructions: This is a closed-book examination. You may, however, use one 8.5 -by-11 page of notes, your note

More information

On the convergence rates of genetic algorithms

On the convergence rates of genetic algorithms Theoretical Computer Science 229 (1999) 23 39 www.elsevier.com/locate/tcs On the convergence rates of genetic algorithms Jun He a;, Lishan Kang b a Department of Computer Science, Northern Jiaotong University,

More information

AP Calculus (BC) Chapter 9 Test No Calculator Section Name: Date: Period:

AP Calculus (BC) Chapter 9 Test No Calculator Section Name: Date: Period: WORKSHEET: Series, Taylor Series AP Calculus (BC) Chapter 9 Test No Calculator Section Name: Date: Period: 1 Part I. Multiple-Choice Questions (5 points each; please circle the correct answer.) 1. The

More information

The Pacic Institute for the Mathematical Sciences http://www.pims.math.ca pims@pims.math.ca Surprise Maximization D. Borwein Department of Mathematics University of Western Ontario London, Ontario, Canada

More information

Lecture 3: Lower Bounds for Bandit Algorithms

Lecture 3: Lower Bounds for Bandit Algorithms CMSC 858G: Bandits, Experts and Games 09/19/16 Lecture 3: Lower Bounds for Bandit Algorithms Instructor: Alex Slivkins Scribed by: Soham De & Karthik A Sankararaman 1 Lower Bounds In this lecture (and

More information

Lecture 3. Mathematical methods in communication I. REMINDER. A. Convex Set. A set R is a convex set iff, x 1,x 2 R, θ, 0 θ 1, θx 1 + θx 2 R, (1)

Lecture 3. Mathematical methods in communication I. REMINDER. A. Convex Set. A set R is a convex set iff, x 1,x 2 R, θ, 0 θ 1, θx 1 + θx 2 R, (1) 3- Mathematical methods in communication Lecture 3 Lecturer: Haim Permuter Scribe: Yuval Carmel, Dima Khaykin, Ziv Goldfeld I. REMINDER A. Convex Set A set R is a convex set iff, x,x 2 R, θ, θ, θx + θx

More information

Theorem 1 can be improved whenever b is not a prime power and a is a prime divisor of the base b. Theorem 2. Let b be a positive integer which is not

Theorem 1 can be improved whenever b is not a prime power and a is a prime divisor of the base b. Theorem 2. Let b be a positive integer which is not A RESULT ON THE DIGITS OF a n R. Blecksmith*, M. Filaseta**, and C. Nicol Dedicated to the memory of David R. Richman. 1. Introduction Let d r d r,1 d 1 d 0 be the base b representation of a positive integer

More information

2 2 + x =

2 2 + x = Lecture 30: Power series A Power Series is a series of the form c n = c 0 + c 1 x + c x + c 3 x 3 +... where x is a variable, the c n s are constants called the coefficients of the series. n = 1 + x +

More information

Mathematical Institute, University of Utrecht. The problem of estimating the mean of an observed Gaussian innite-dimensional vector

Mathematical Institute, University of Utrecht. The problem of estimating the mean of an observed Gaussian innite-dimensional vector On Minimax Filtering over Ellipsoids Eduard N. Belitser and Boris Y. Levit Mathematical Institute, University of Utrecht Budapestlaan 6, 3584 CD Utrecht, The Netherlands The problem of estimating the mean

More information

Lecture Notes 3 Convergence (Chapter 5)

Lecture Notes 3 Convergence (Chapter 5) Lecture Notes 3 Convergence (Chapter 5) 1 Convergence of Random Variables Let X 1, X 2,... be a sequence of random variables and let X be another random variable. Let F n denote the cdf of X n and let

More information

460 HOLGER DETTE AND WILLIAM J STUDDEN order to examine how a given design behaves in the model g` with respect to the D-optimality criterion one uses

460 HOLGER DETTE AND WILLIAM J STUDDEN order to examine how a given design behaves in the model g` with respect to the D-optimality criterion one uses Statistica Sinica 5(1995), 459-473 OPTIMAL DESIGNS FOR POLYNOMIAL REGRESSION WHEN THE DEGREE IS NOT KNOWN Holger Dette and William J Studden Technische Universitat Dresden and Purdue University Abstract:

More information

Big-oh stuff. You should know this definition by heart and be able to give it,

Big-oh stuff. You should know this definition by heart and be able to give it, Big-oh stuff Definition. if asked. You should know this definition by heart and be able to give it, Let f and g both be functions from R + to R +. Then f is O(g) (pronounced big-oh ) if and only if there

More information

Fat-shattering. and the learnability of real-valued functions. Peter L. Bartlett. Department of Systems Engineering

Fat-shattering. and the learnability of real-valued functions. Peter L. Bartlett. Department of Systems Engineering Fat-shattering and the learnability of real-valued functions Peter L. Bartlett Department of Systems Engineering Research School of Information Sciences and Engineering Australian National University Canberra,

More information

MARKOV CHAINS: STATIONARY DISTRIBUTIONS AND FUNCTIONS ON STATE SPACES. Contents

MARKOV CHAINS: STATIONARY DISTRIBUTIONS AND FUNCTIONS ON STATE SPACES. Contents MARKOV CHAINS: STATIONARY DISTRIBUTIONS AND FUNCTIONS ON STATE SPACES JAMES READY Abstract. In this paper, we rst introduce the concepts of Markov Chains and their stationary distributions. We then discuss

More information

COUNTING NUMERICAL SEMIGROUPS BY GENUS AND SOME CASES OF A QUESTION OF WILF

COUNTING NUMERICAL SEMIGROUPS BY GENUS AND SOME CASES OF A QUESTION OF WILF COUNTING NUMERICAL SEMIGROUPS BY GENUS AND SOME CASES OF A QUESTION OF WILF NATHAN KAPLAN Abstract. The genus of a numerical semigroup is the size of its complement. In this paper we will prove some results

More information

Homework 1 Due: Wednesday, September 28, 2016

Homework 1 Due: Wednesday, September 28, 2016 0-704 Information Processing and Learning Fall 06 Homework Due: Wednesday, September 8, 06 Notes: For positive integers k, [k] := {,..., k} denotes te set of te first k positive integers. Wen p and Y q

More information

IMC 2015, Blagoevgrad, Bulgaria

IMC 2015, Blagoevgrad, Bulgaria IMC 05, Blagoevgrad, Bulgaria Day, July 9, 05 Problem. For any integer n and two n n matrices with real entries, B that satisfy the equation + B ( + B prove that det( det(b. Does the same conclusion follow

More information

FIRST YEAR CALCULUS W W L CHEN

FIRST YEAR CALCULUS W W L CHEN FIRST YER CLCULUS W W L CHEN c W W L Chen, 994, 28. This chapter is available free to all individuals, on the understanding that it is not to be used for financial gain, and may be downloaded and/or photocopied,

More information

Classication of greedy subset-sum-distinct-sequences

Classication of greedy subset-sum-distinct-sequences Discrete Mathematics 271 (2003) 271 282 www.elsevier.com/locate/disc Classication of greedy subset-sum-distinct-sequences Joshua Von Kor Harvard University, Harvard, USA Received 6 November 2000; received

More information

Characterising FS domains by means of power domains

Characterising FS domains by means of power domains Theoretical Computer Science 264 (2001) 195 203 www.elsevier.com/locate/tcs Characterising FS domains by means of power domains Reinhold Heckmann FB 14 Informatik, Universitat des Saarlandes, Postfach

More information

Numerical Analysis: Interpolation Part 1

Numerical Analysis: Interpolation Part 1 Numerical Analysis: Interpolation Part 1 Computer Science, Ben-Gurion University (slides based mostly on Prof. Ben-Shahar s notes) 2018/2019, Fall Semester BGU CS Interpolation (ver. 1.00) AY 2018/2019,

More information

Statistical Learning Theory

Statistical Learning Theory Statistical Learning Theory Fundamentals Miguel A. Veganzones Grupo Inteligencia Computacional Universidad del País Vasco (Grupo Inteligencia Vapnik Computacional Universidad del País Vasco) UPV/EHU 1

More information

RANDOM WALKS AND THE PROBABILITY OF RETURNING HOME

RANDOM WALKS AND THE PROBABILITY OF RETURNING HOME RANDOM WALKS AND THE PROBABILITY OF RETURNING HOME ELIZABETH G. OMBRELLARO Abstract. This paper is expository in nature. It intuitively explains, using a geometrical and measure theory perspective, why

More information

Contents. 2.1 Vectors in R n. Linear Algebra (part 2) : Vector Spaces (by Evan Dummit, 2017, v. 2.50) 2 Vector Spaces

Contents. 2.1 Vectors in R n. Linear Algebra (part 2) : Vector Spaces (by Evan Dummit, 2017, v. 2.50) 2 Vector Spaces Linear Algebra (part 2) : Vector Spaces (by Evan Dummit, 2017, v 250) Contents 2 Vector Spaces 1 21 Vectors in R n 1 22 The Formal Denition of a Vector Space 4 23 Subspaces 6 24 Linear Combinations and

More information

Robust linear optimization under general norms

Robust linear optimization under general norms Operations Research Letters 3 (004) 50 56 Operations Research Letters www.elsevier.com/locate/dsw Robust linear optimization under general norms Dimitris Bertsimas a; ;, Dessislava Pachamanova b, Melvyn

More information

Section Taylor and Maclaurin Series

Section Taylor and Maclaurin Series Section.0 Taylor and Maclaurin Series Ruipeng Shen Feb 5 Taylor and Maclaurin Series Main Goal: How to find a power series representation for a smooth function us assume that a smooth function has a power

More information

TAYLOR AND MACLAURIN SERIES

TAYLOR AND MACLAURIN SERIES TAYLOR AND MACLAURIN SERIES. Introduction Last time, we were able to represent a certain restricted class of functions as power series. This leads us to the question: can we represent more general functions

More information

Taylor and Maclaurin Series. Copyright Cengage Learning. All rights reserved.

Taylor and Maclaurin Series. Copyright Cengage Learning. All rights reserved. 11.10 Taylor and Maclaurin Series Copyright Cengage Learning. All rights reserved. We start by supposing that f is any function that can be represented by a power series f(x)= c 0 +c 1 (x a)+c 2 (x a)

More information

What can you prove by induction?

What can you prove by induction? MEI CONFERENCE 013 What can you prove by induction? Martyn Parker M.J.Parker@keele.ac.uk Contents Contents iii 1 Splitting Coins.................................................. 1 Convex Polygons................................................

More information

MATH10040 Chapter 1: Integers and divisibility

MATH10040 Chapter 1: Integers and divisibility MATH10040 Chapter 1: Integers and divisibility Recall the basic definition: 1. Divisibilty Definition 1.1. If a, b Z, we say that b divides a, or that a is a multiple of b and we write b a if there is

More information

Lecture 6 Basic Probability

Lecture 6 Basic Probability Lecture 6: Basic Probability 1 of 17 Course: Theory of Probability I Term: Fall 2013 Instructor: Gordan Zitkovic Lecture 6 Basic Probability Probability spaces A mathematical setup behind a probabilistic

More information

Contribution of Problems

Contribution of Problems Exam topics 1. Basic structures: sets, lists, functions (a) Sets { }: write all elements, or define by condition (b) Set operations: A B, A B, A\B, A c (c) Lists ( ): Cartesian product A B (d) Functions

More information

The limitedness problem on distance automata: Hashiguchi s method revisited

The limitedness problem on distance automata: Hashiguchi s method revisited Theoretical Computer Science 310 (2004) 147 158 www.elsevier.com/locate/tcs The limitedness problem on distance automata: Hashiguchi s method revisited Hing Leung, Viktor Podolskiy Department of Computer

More information

CHAPTER 4. Series. 1. What is a Series?

CHAPTER 4. Series. 1. What is a Series? CHAPTER 4 Series Given a sequence, in many contexts it is natural to ask about the sum of all the numbers in the sequence. If only a finite number of the are nonzero, this is trivial and not very interesting.

More information