EECS Divisional Presentation Computing, Algorithms and Applications

EECS Divisional Presentation
Computing, Algorithms and Applications
May 25, 2006
Current CAA Faculty
Primary Members:
• Ming-Yang Kao: theoretical computer science
• Jorge Nocedal: continuous optimization
Secondary Members:
• Yan Chen: networking and security
• Peter Scheuermann: databases
• Hai Zhou: CAD algorithms and formal methods
Tertiary Members:
• Alan Toflove: computational electrodynamics
A Framework to Understand CAA Research
Algorithms
Externals
Models of
Computation
(applications of
computation to other fields,
and vice versa)
Complexity
(resources used
by computation)
Strategic Bidding
J. Nocedal and R. Waltz
• Your company sells electric power (internet resources, wireless
bandwidth).
• You and other producers submit competitive bids to generate
power.
• An Independent Operator purchases at a single “spot price.”
• Your strategic guidance:
– submit low bids  spot price
– submit high bids to drive up the spot price
– Demands, etc, uncertain
110 000
100 000
Powernext Day-Ahead™: daily volume and baseload price
250
MWh
80 000
200
70 000
60 000
150
50 000
40 000
En €/MWh
90 000
100
30 000
20 000
50
10 000
0
-
Daily volume
Baseload price
Independent operator solves an (easy) optimization problem -given the bids, determines amount gj to buy from you.
min g J
b g
J
j
j
s.t. 0  g J  capacity J
g
j
J  1,...
 demand J
J  1,...
bj = bid of company j
cj = gener cost for company
gj = gener sold by plant j
j
Spot price is Lagrange multiplier.
Your problem (j=1)
max b J ( LD  c J ) g J
subject to bJ  c J
IndepOp solution
Optimization
Problem!!
IndepOper : bJ  g J , LD
Bi-level Optimization Problem:
• What about bids from competitors? Use stochastic
optimization.
• Very large and nonlinear problem
• Mathematically deficient --- need new theory
Northwestern Lab for Internet and
Security Technology (LIST)
Yan Chen
High-performance Network Anomaly/Intrusion Detection
and Mitigation (HPNAIDM) Systems
• Data streaming computation: 10s Gigabit-link network
traffic recording and analysis (with P. Dinda and G. Memik)
• Combinatorial statistics: first online network-based
polymorphic worm signature generation with provable
attack resilience (with M. Kao)
• Formal verification: vulnerability analysis of 802.16
protocols using formal methods (with H. Zhou, J. Fu
(Motorola) )
• Information theory: network anomaly & intrusion detection
(with D. Guo)
The Spread of Sapphire/Slammer
Worms
Northwestern Lab for Internet and
Security Technology (LIST)
Yan Chen
Internet Measurement, Diagnosis & Inference
• Linear Algebra: Scalable
and deterministic network
monitoring, diagnosis, and
link-level properties (e.g.,
loss rate) inference
• Statistics: Network router
configuration (e.g., QoS)
inference (with F.
Bustamante and G. Lu
(Tsinghua))
Why is it
so slow?
C&W
AT&T
UUNet
Sprint
Qwest
Earthlink
AOL
It’s so
slow!
Applied Computational Geometry
Peter Scheuermann
SENSOR RELOCATION
Critical Region
R
Problem: How to optimize the guidance of
mobile sensors which need to be
brought into a critical region, to
ensure a desired level of coverage for
that region?
Variants use convex hull of critical region
r
Publication: “Mission-Critical Management of Mobile
Sensors (or, How to Guide a Flock of Sensors)
in DMSN 2004
1. fastest arrival time for the desired
number of sensors
2. largest number of sensors to ensure
desired quantity inside the region
3. optimal time to ensure “fair” coverage
under the constraint that a minimum
number of sensors are inside the region
DYNAMIC TOPOLOGICAL PREDICATES FOR
MOVING OBJECTS
Problem: Notify me when an object
is continuously_moving_towards
the landmark LM, for more than
5 min., based on periodic
(location,time) updates
(primitive events)
A
F
E
B
C
To Send
Solution:
Use Voronoi diagram (for the LM)
and monitoring of only two
consecutive updates;
- Issue: consumption of primitive
events?
Publication: “Dynamic Topological Predicates and Notifications in
Moving Object Databases”
in MDM 2005
D
To Send or
Not To Send?
(have the previous
simple events been
“consumed”)
Send update!
Optimal and Efficient Algorithms for Circuit Retiming
Hai Zhou
• Retiming is an effective technique for circuit optimization by
relocating registers without changing functionality
• We developed the most efficient algorithm for clock period
minimization considering both long and short paths (in
O(n2m) time)
• Our algorithm is correct no matter what order is used for
selecting nodes
Gate Sizing for Coupling Noise Control as
Distributed Optimization
Hai Zhou
• Noise on a signal is proportional
to attacker gate sizes and
inversely proportional to its own
gate size
• Given the coupling relations and
the noise upper bound for each
signal
• Need to find minimal gate sizes
such that all noises are under
constraints
Our algorithm:
Each gate starts at lower bound
Repeat:
Each signal with violation
up-size its gate to the
smallest with tolerable noise
• Correct no matter what order is
taken
• Will converge to the optimal
solution if there is one
• Very efficient practically
• May be used in wireless networks
DNA Algorithmic Self-Assembly
TILE
GCATCG
CGTAGC
DNA Algorithmic Self-Assembly
Program = Tiles + Lab Steps
Output
DNA Algorithmic Self-Assembly
Input: the description of a shape
Output: a set of tiles and a sequence of lab steps
to produce the shape
Computational Objectives:
• minimize the # of tile types
• minimize the range of temperatures
• minimize the # of lab steps
• minimize errors
Sequencing Bio-molecules
Input: information about small pieces of a target molecule
Output: the character sequence of the target molecule
Examples:
• Peptide Sequencing: linear structure (with a group at
Harvard Medical School)
• Glycan Sequencing: tree structure (with a group at
Kyoto University)
Sequencing Bio-molecules
Given: a target bio-molecule B
Steps:
1. Make many copies of B.
2.
Cut each copy of B into pieces.
3.
Sequence each piece (recursively).
4.
Assemble the character sequences of the pieces into
the character sequence of B.
Protein Analysis: HPLC-MS-MS
HPLC
Proteins
Mass
Spectrometer
Peptides
Mass
Spectrometer
Fragmentation
& ionization
B-ions / Y-ions
One Peptide
Mass/Charge
De Novo Peptide Sequencing
Protein Database Searching
Mass/Charge
Tandem Mass Spectrum
Synergies with Other Divisions
Cognitive Systems
+
Graphics & Interactive Media
Signals & Systems
Computational Economics
Network Optimization
DNA Computing
Musical Retrieval
CAA
Quantum Computing
Cryptography
Solid State & Photonics
Bioinformatics
Computer Worm Detection
Design Optimization
DNA Computing
Computer Engineering
& Systems
CAA’s Mission:
To Understand the Nature, Power, Limit of Computation; and
to Apply Such Understanding to Benefit the Society.
Basic Understanding about Computation:
Computation is an intellectual tool as powerful and universal as
mathematics.
Computation can be used not only to solve mathematical problems, but
also to understand and design complex systems.
Examples of Computation:
•
How many bits of information does a black hole compute?
•
How do we make web search efficiently provide the information that we want?
•
How do we create a biological “computer” that uses DNA/RNA-like materials to
produce medicines?
The End
Thank You!