[YPNG] YPNG, Friday 4 November 2016
sekhar.tatikonda at yale.edu
Mon Oct 31 20:30:51 EDT 2016
Donald Lee, from Yale SOM, will talk in the YPNG seminar this week:
Title: Sparse interpretable estimators for cyclic arrival rates
Abstract: Like many customer arrival processes, patient arrivals to the
Emergency Department (ED) exhibit strong cyclic behaviour. For example, ED
demand often peaks around midday and drops off at night. In this talk we
propose a simple procedure for estimating the intensity of a nonhomogeneous
Poisson process. The estimator is interpretable because it is a simple sum
of p sinusoids, although p and the frequency, amplitude, and phase of each
wave are not known and need to be estimated. This results in a flexible
specification that is suitable for use in modelling as well as in high
We first discuss some estimation pitfalls that can arise when estimating
the frequency components in the arrival rate using intuitive heuristics,
and then we show how our approach addresses these issues. Our procedure
sits in between classic periodogram methods and atomic/total variation norm
thresholding, and a novel aspect is the use of window functions with fast
decaying spectral tails, which we show improves frequency resolution as
well as prevent strong frequency signals from masking weaker ones nearby.
Interestingly the prolate spheriodal window that is usually considered
optimal in signal processing is suboptimal for frequency recovery.
Under suitable conditions, finite sample performance guarantees for our
procedure can be derived that resolve open questions and expand existing
results in spectral estimation literature, namely the infinite-dimensional
extensions of Lasso by Candes & Fernandez-Granda 2013 and Tang, Bhaskar &
Recht 2015. We apply the procedure to patient arrivals data from the Yale
ED to determine the extent to which the arrival rate can be represented by
simple harmonics. Joint work with Ningyuan Chen (HKUST) and Sahand
See you Friday at 11am in the Stat's classroom.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the YPNG