network analytics across scales
martin is down; hasn't done it before
what heather is doing now
dynamic functional connectivity
standard HMM: implicitly the sojourn distribution is geometrically distributed
dont want to make that assumption; might estimate more state transitions than might actually be happening
semi HMM; can estimate the sojourn distribution
estimating networks/driving changes in the networks
martin: this could be useful as a service to the community
variance shrinkage for t statistics: LIMMA
Caffo: variance shrinkage in neuroscience
variance smoothing
lots of papers on fully bayesian approaches to t-like statistics
smythe writes limma, which has t-test tricks built in to it
take full human connectome project data at different sample sizes
one idea: see how much doing something like limma helps
dimishing returns as your sample size increases
voxelwise GLM multi-subject group analyses
havent thought about it in the context of connectomes
combine w HRF estimation
want ot estimate HRF AND estimate whether there is significance at a voxel
just take whatever is peoples' t statistic soln and perform some shrinkage; does it help?
taylor: cone statistic: max t-value?
VC: derivative boosting?
ideas about conditional independence breaking down as you summarize things
network analytics across scales martin is down; hasn't done it before what heather is doing now dynamic functional connectivity standard HMM: implicitly the sojourn distribution is geometrically distributed dont want to make that assumption; might estimate more state transitions than might actually be happening semi HMM; can estimate the sojourn distribution estimating networks/driving changes in the networks martin: this could be useful as a service to the community variance shrinkage for t statistics: LIMMA Caffo: variance shrinkage in neuroscience variance smoothing lots of papers on fully bayesian approaches to t-like statistics smythe writes limma, which has t-test tricks built in to it take full human connectome project data at different sample sizes one idea: see how much doing something like limma helps dimishing returns as your sample size increases voxelwise GLM multi-subject group analyses havent thought about it in the context of connectomes combine w HRF estimation want ot estimate HRF AND estimate whether there is significance at a voxel just take whatever is peoples' t statistic soln and perform some shrinkage; does it help? taylor: cone statistic: max t-value? VC: derivative boosting?
ideas about conditional independence breaking down as you summarize things