IBM Research - Ireland Internship Project: Event detection in networked systems - overview
What sampling rate is needed for a system to detect events with a certain delay and privacy guarantees? Or, alternatively, what is the expected delay, given a budget on the number of samples per second and a constraint on privacy? We would like to explore on-line, real time, algorithms for problems, where there is a network structure, discrete states change abruptly, and the exchange of information comes at a price, and prove:
upper-bounds on the expected difference between the estimate of a state and the actual state (expected regret), given a budget for communication (sampling rate), and constraint on privacy, or equivalently
expected delay until detection, given a budget for communication (sampling rate) and a constraint on privacy.
- background in applied probability (with experience on e.g. multi-armed bandit problems or Bayesian networks);
- familiarity with dynamical systems and stochastic different and differential equations (in particular stability);
- proficiency with Python.
The proposed research is interdisciplinary and combines in a novel way concepts from control theory and applied probability.