Upload
argus
View
40
Download
0
Embed Size (px)
DESCRIPTION
Grand Challenges in Methodologies for Complex Networks. Ness Shroff Dept. of ECE and CSE The Ohio State University E-mail: [email protected]. September 20, 2012. Complex Networks. Heterogeneous Mobile Dynamic System Rule-based or Selfish “agents” interact Multi-time scale - PowerPoint PPT Presentation
Citation preview
Ness ShroffDept. of ECE and CSEThe Ohio State UniversityE-mail: [email protected]
Grand Challenges in Methodologies for Complex Networks
September 20, 2012
Complex Networks
Heterogeneous Mobile Dynamic System Rule-based or Selfish “agents” interact
Multi-time scale Varied Aggregation Limited feedback Uncertainty (stochasticity) Local and Global (Resource) Constraints
Examples of Complex NetworksCommunication Networks
InternetWireless & Sensor Networks
Online Social NetworksProfessional (LinkedIn…)Personal (Facebook, Twitter…)
Cyber-physical Smart-gridActuator based sensor networks
CloudData-center networks…
Methodological SuccessesStochastic optimization and control unified with
combinatorial techniquesMathematical Decomposition Framework Distributed and robust low-complexity protocols
Opportunistic scheduling (MAC)Congestion controlRoutingEnergy/Power control…
Glauber Dynamics (statistical physics)Global optima can be achieved through purely local interactions
Focus: Long-term metrics (stability, throughput, lifetime, energy…)Less so on short-term metrics (delay, convergence speeds…)
Grand ChallengesAnalytical framework to design solutions that
simultaneously achieve: low complexity, high-throughput, and low delayDeep connections between calculus of variations, probabilistic
methods, limit theorems, and combinatorial techniquesControl “meta-dynamics” taking into account user preferences,
social interactions, cyber-physical interplay to achieve global behavior (optimality, consensus, equilibria…)New methodologies involving dynamic game theory, but now with
underlying social/cyberphysical graph structures and user behavior (rational vs myopic behavior)
Manage uncertainty and sensitivities to imperfections (e.g., feedback delays, errors, non-observability…)Breakthroughs in partially observable decision processes (POMDP)New learning techniques to infer system and user behavior in this
highly dynamic setting
Thank you
● 6