multiplicative weights method: a meta algorithm with applications to linear and semi-definite...
Post on 21-Dec-2015
220 views
TRANSCRIPT
![Page 1: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/1.jpg)
Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming
Sanjeev AroraPrinceton University
Based upon:Fast algorithms for Approximate SDP [FOCS ‘05]log(n) approximation to SPARSEST CUT in Õ(n2) time [FOCS ‘04]The multiplicative weights update method and it’s applications [’05]See also recent papers by Hazan and Kale.
![Page 2: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/2.jpg)
Multiplicative update rule (long history)
Applications: approximate solutions to LPs and SDPs, flow problems,
online learning (boosting), derandomization & chernoff bounds,
online convex optimization, computational geometry,
metricembeddongs, portfolio management… (see our survey)
w1
w2
.
.
.
wn
n agents weights
Update weights according toperformance:
wit+1 Ã wi
t (1 + ¢ performance of i)
![Page 3: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/3.jpg)
Simplest setting – predicting the market
N “experts” on TV
Can we perform as good as the best expert ?
1$ for correct prediction
0$ for incorrect
![Page 4: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/4.jpg)
Weighted majority algorithm [LW ‘94]
Multiplicative update (initially all wi =1):
If expert predicted correctly: wit+1 Ã wi
t
If incorrectly, wit+1 Ã wi
t(1 - )
Claim: #mistakes by algorithm ¼ 2(1+)(#mistakes by best expert)
Potential: t = Sum of weights= i wit (initially n)
If algorithm predicts incorrectly ) t+1 · t - t /2
T · (1-)m(A) n m(A) =# mistakes by algorithm
T ¸ (1-)mi
) m(A) · 2(1+)mi + O(log n/)
“Predict according to the weighted majority.”
![Page 5: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/5.jpg)
Generalized Weighted majority [A.,Hazan, Kale ‘05]
n agents Set of events (possibly infinite)
event j
expert ipayoffM(i,j)
![Page 6: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/6.jpg)
Generalized Weighted majority [AHK ‘05]
n agents Set of events (possibly infinite)
Claim: After T iterations,
Algorithm payoff ¸ (1-) best expert – O(log n / )
Algorithm: plays distribution on experts (p1,…,pn)
Payoff for event j: i pi M(i,j)
Update rule:pi
t+1 Ã pit (1 + ¢ M(i,j))
p1
p2
.
.
.
pn
![Page 7: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/7.jpg)
Game playing,Online optimization
Lagrangean relaxation
Gradient descent
Chernoff bounds
Games with Matrix Payoffs Fast soln to
LPs, SDPsBoosting
![Page 8: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/8.jpg)
Common features of MW algorithms
“competition” amongst n experts
Appearance of terms like
exp( - t (performance at time t) )
Time to get -approximate solutions is proportional to 1/2.
![Page 9: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/9.jpg)
Boosting and AdaBoost(Schapire’91, Freund-Schapire’97)
“Weak learning ) Strong learning”
Input: 0/1 vectors with a “Yes/No” label
(white, tall, vegetarian, smoker,…,) “Has disease”(nonwhite, short, vegetarian, nonsmoker,..) “No disease”
Desired: Short OR-of-AND (i.e., DNF) formula that describes fraction of data (if one exists)
(white Æ vegetarian) Ç (nonwhite Æ nonsmoker)
= 1- “Strong Learning” = ½ + “Weak Learning”
Main idea in boosting: data points are “experts”, weak learners are “events.”
![Page 10: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/10.jpg)
Approximate solutions to convexprogramming
e.g., Plotkin Shmoys Tardos ’91, Young’97, Garg Koenemann’99, Fleischer’99
MW Meta-Algorithm gives unified view
Note: Distribution $ Convex Combination
{ pi}
i pi =1
If P is a convex set andeach xi 2 P
then i pi xi 2 P
![Page 11: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/11.jpg)
Solving LPs (feasibility)a1¢ x ¸ b1
a2¢ x ¸ b2
am¢ x ¸ bm
x 2 P
k wk(ak¢ x – bk) ¸ 0x 2 P
P = convex domain
- -
-
w1
w2
wm
Event Oracle
![Page 12: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/12.jpg)
Solving LPs (feasibility)a1¢ x ¸ b1
a2¢ x ¸ b2
am¢ x ¸ bm
x 2 P
w1
w2
wm
[1 – (a1 ¢ x0 – b1)/] £[1 – (a2 ¢ x0 – b2)/] £
[1 – (am ¢ x0 – bm)/] £
x0
|ak ¢ x0 – bk| · = width
k wk(ak¢ x – bk) ¸ 0x 2 P
Oracle
Final solution =Average x vector
![Page 13: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/13.jpg)
Performance guarantees
In O(2 log(n)/2) iterations, average x is feasible.
Packing-Covering LPs: [Plotkin, Shmoys, Tardos
’91]
9? x 2 P:
j = 1, 2, … m: aj ¢ x ¸ 1
Want to find x 2 P s.t. aj ¢ x ¸ 1 -
Assume: 8 x 2 P: 0 · aj ¢ x ·
MW algorithm gets feasible x in O(log(n)/2)
iterations
Covering problem
![Page 14: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/14.jpg)
Connection to Chernoff bounds and derandomization
Deterministic approximation algorithms for 0/1 packing/covering problem a la Raghavan-Thompson
Solve LP relaxation of integer program.Obtain soln (yi)
Randomized Rounding O(log n) times)
xi à 1 w/ prob. yi
xi à 0 w/ prob.1 - yi
Derandomize using pessimistic estimators
exp(i t ¢ f(yi))Young [95] “Randomized rounding without solving the LP.” MW update rule mimics pessimistic estimator.
![Page 15: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/15.jpg)
Semidefinite programming (Klein-
Lu’97)
a1¢ x ¸ b1
a2¢ x ¸ b2
am¢ x ¸ bm
x 2 P
P =
aj and x : symmetric
matrices in Rn £ n
{x: x is psd; tr(x) =1}
Oracle: max j wj (aj ¢ x) over P
(eigenvalue computation!)
Application 2:
![Page 16: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/16.jpg)
The promise of SDPs…
0.878-approximation to MAX-CUT (GW’94),7/8-approximation to MAX-3SAT (KZ’00)plog n –approximation to SPARSEST CUT, BALANCED SEPARATOR (ARV’04).plog n-approximation to MIN-2CNF DELETION,MIN-UNCUT (ACMM’05), MIN-VERTEX SEPARATOR (FHL’06),…etc. etc….
The pitfall…
High running times ---as high as n4.5 in recent works
![Page 17: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/17.jpg)
Solving SDP relaxations more efficientlyusing MW [AHK’05]
Problem Using Interior
Point
Our result
MAXQP (e.g. MAX-
CUT)
Õ(n3.5) Õ(n1.5N/2.5) or
Õ(n3/*3.5)
HAPLOFREQ Õ(n4) Õ(n2.5/2.5)
SCP Õ(n4) Õ(n1.5N/4.5)
EMBEDDING Õ(n4) Õ(n3/d53.5)
SPARSEST CUT Õ(n4.5) Õ(n3/2)
MIN UNCUT etc Õ(n4.5) Õ(n3.5/2)
![Page 18: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/18.jpg)
Key difference between efficient and
not-so-efficient implementations of the MW idea: Width
management.
(e.g., the difference between PST’91 and GK’99)
![Page 19: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/19.jpg)
Recall: issue of width
Õ(2/2) iterations
to obtain
feasible x
= maxk |ak ¢ x –
bk|
is too large!!
MWa1¢ x ¸ b1
a2¢ x ¸ b2
am¢ x ¸ bm
Oracle
k wk(ak¢ x – bk) ¸ 0x 2 P
![Page 20: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/20.jpg)
Issue 1:Dealing with width
A few high width
constraints
Run a hybrid of
MW and
ellipsoid/Vaidya
poly(m, log(/))
iterations to
obtain feasible x
MWa1¢ x ¸ b1
a2¢ x ¸ b2
am¢ x ¸ bm
Oracle
k wk(ak¢ x – bk) ¸ 0x 2 P
![Page 21: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/21.jpg)
Hybrid of MW
and Vaidya
Õ(L2/2) iterations
to obtain
feasible x
L ¿
MWa1¢ x ¸ b1
a2¢ x ¸ b2
am¢ x ¸ bm
Oracle
k wk(ak¢ x – bk) ¸ 0x 2 P
Dual ellipsoid/Vaidya
Dealing with width (contd)
![Page 22: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/22.jpg)
Issue 2: Efficient implementation of Oracle: fast eigenvalues via matrix sparsification
Lanczos algorithm effectively uses sparsity of CSimilar to Achlioptas, McSherry [’01], but better in some situations (also easier analysis)
C
O(O(nnijij|C|Cijij|/|/)) non-zero entries
k k C – C’C – C’k ·k ·
Random sampling
C’
![Page 23: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/23.jpg)
O(n2)-time algorithm to compute O(plog n)-approximationto SPARSEST CUT [A., Hazan, Kale ’04]
(combinatorial algorithm; improves upon O(n4.5) algorithm of A, Rao,
Vazirani)
![Page 24: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/24.jpg)
Sparsest Cut
S
O(log n) approximation [Leighton Rao ’88]
O(plog n) approximation [A., Rao, Vazirani’04]
O(p log n) approximation in O(n2) time. (Actually, finds expander flows) [A., Hazan, Kale’05]
The sparsest cut:
SG) = 2/5
![Page 25: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/25.jpg)
MW algorithm to find expander flows
Events – { (s,w,z) | weights on vertices, edges, cuts}
Experts – pairs of vertices (i,j)
Payoff: (for weights di,j on experts)
Fact: If events are chosen optimally, the distribution on
experts di,j converges to a demand graph which is an
“expander flow”
[by results of Arora-Rao-Vazirani ’04 suffices to
produce approx. sparsest cut]
shortest path according to weights
we
Cuts separating i and j
![Page 26: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/26.jpg)
New: Online games with matrix payoffs
(joint w/ Satyen Kale’06)
Payoff is a matrix, and so is the “distribution” on experts!
Uses matrix analogues of usual inequalities
1+ x · ex I + A · eA
Can be used to give a new “primal-dual” view of SDP relaxations () easier, faster approximation algorithms)
![Page 27: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/27.jpg)
New: Faster algorithms for online learningand portfolio management (Agarwal-Hazan’06, Agarwal-Hazan-Kalai-Kale’06 )
Framework for online optimization inspired by Newton’s method (2nd order optimization). (Note: MW ¼ gradient descent)
Fast algorithms for Portfolio management and other online optimization problems
![Page 28: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/28.jpg)
Open problems
Better approaches to width management?
Faster run times?
Lower bounds?
THANK YOU
![Page 29: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/29.jpg)
Connection to Chernoff bounds and derandomization
Deterministic approximation algorithms a la Raghavan-Thompson
Packing/covering IP with variables xi = 0/1
9? x 2 P: 8 j 2 [m], fj(x) ¸ 0
Solve LP relaxation using variables yi 2 [0, 1]
Randomized rounding:
xi =
Chernoff: O(log m) sampling iterations suffice
1 w.p. yi
0 w.p. 1 – yi
![Page 30: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/30.jpg)
Derandomization [Young, ’95]
Can derandomize the rounding using exp(tj fj(x))
as a pessimistic estimator of failure probability
By minimizing the estimator in every iteration, we
mimic the random expt, so O(log m) iterations
suffice
The structure of the estimator obviates the need
to solve the LP: Randomized rounding without
solving the Linear Program
Punchline: resulting algorithm is the MW
algorithm!
![Page 31: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/31.jpg)
Weighted majority [LW ‘94]
If lost at t, t+1 · (1-½ ) t
At time T: T · (1-½ )#mistakes ¢ 0
Overall:
#mistakes · log(n)/ + (1+) mi
#mistakes ofexpert i
![Page 32: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/32.jpg)
Semidefinite programming
Vectors aj and x: symmetric matrices in Rn £ n
x º 0
Assume: Tr(x) · 1
Set P = {x: x º 0, Tr(x) · 1}
Oracle: max j wj(aj ¢ x) over P
Optimum: x = vvT where v is the largest
eigenvector of jw
j aj
![Page 33: Multiplicative weights method: A meta algorithm with applications to linear and semi-definite programming Sanjeev Arora Princeton University Based upon:](https://reader031.vdocuments.us/reader031/viewer/2022032704/56649d615503460f94a42962/html5/thumbnails/33.jpg)
Efficiently implementing the oracle
Optimum: x = vvT
v is the largest eigenvector of some matrix C
Suffices to find a vector v such that vTCv ¸ 0
Lanczos algorithm with a random starting vector
is ideal for this
Advantage: uses only matrix-vector products
Exploits sparsity (also: sparsification
procedure)
Use analysis of Kuczynski and Wozniakowski [’92]