Continuous case: GMRF + Probability of error

Main points:

1. MSE comparisons for WLS (Network aided) vs LS  for different topologies.
2. Extension to the detection problem.
3. Some more performance comparisons ...

Staregic sensor placement for sentiment detection

With the star network, we can get reasonably nice analytical results... Q: How do we extend it?
Will one sensor cut it? Probably not!

Trivial vs MAP

How does probability of error vary with topology?
Do node weights matter?
Yep!

Star network + Noisy central node

Error exponent analysis for Star networks. Quite interesting results ...

Classification erro vs topology

An attempt at doing low SNR analysis of MSE which is UB of Probability of error via the Verdu-Estimation-marries-Info theory route.
Result: MSE in terms of the graph spectra. Or rather spectra of the covariance matrix!

Trick: Approximate the GMRF via a common Gaussian.