grant issues - Githubissues

jovo commented 8 years ago

super important

[ ] each subtask should have a very clear and bolded goal. THIS ONE IS PARTICULARLY IMPORTANT.
[ ] subtasks and tasks should all have "action names", ie, a verb describing what we will do/achieve. THIS ONE IS ALSO VERY IMPORTANT.

other stuff

[x] define A & B clearly before use.
[x] eq 3 has an "x"?
[x] before defining \Omega, add the word "respectively"
[x] shouldn't \Omega have a * too?
[x] i don't think MGC is a correlation. "generalized" is not an adjective, it changes the meaning.
[x] i think fig 1 is weird. is d=1 for all? what is n? i don't think it shows how awesome mgc is.
[x] you still need about another page of content, this is no joke, they will care, don't screw around, do it right. alternately, you can use the layout from greg's paper (it is more spacious): https://www.overleaf.com/read/tdwkxxckpdhm
[x] task2 summary should mention node dependence stuff.
[x] task 4 be more specific in summary.
[x] i'd make definition of mantel and dcov look much more similar if possible
[x] "Equation 4 is the same when we instead set A = HA ̃ and B = B ̃H" is not true. do you mean the p-value is invariant to that change?
[x] the nomenclature "MGC by X" eg "MGC by Mantel" does not make sense in english. use something like MGC(Mantel), or MGC applied to Mantel.
[x] "that although distance cor- relation under single centering is the same as distance correlation under double centering" also not true, they are not the same. do you mean same performance?
[x] "quests" not the right word here
[x] "&" should not be used in formal writing.
[x] i don't understand what you are proposing to do in task I.2
[x] same with task I.3. be much more concrete and specific about what we will prove/demonstrate. clearly defined goals is super important. state explicit conjectures, or forms thereof is appropriate.
[x] "In particular, we would like to prove the testing consistency of distance-based correlations under network models, compare their performance with other graph-specific testing methods, and understand how can Mgc fare better in the new domain." is a pretty good example, though replace "would like to" with "will"
[x] define indicator function before use
[x] why is indicator function inside summand in eq 5? i would think it would go before the "Bernoulli"?
[x] "This sub-task is dedicated to the distance metric of the graph topology, in an effort to find a suitable distance such that distance correlation and Mgc can be consistent for testing independence." not sure what you mean here, independence between what and what?
[x] "[12] characterizes " use citet here, not cite. references are like footnotes, they can't be referred to explicitly without accompanying text.

cshen6 commented 8 years ago

I am currently working on all the issues locally; pls post here if you spot more. I should be able to solve all of them tonight; and enough time for going over again Monday.

jovo commented 8 years ago

Agreed. All super minor

On Sunday, November 6, 2016, cshen6 notifications@github.com wrote:

I am currently working on all the issues locally; pls post here if you spot more. I should be able to solve all of them tonight; and enough time for going over again Monday.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/neurodata/MGC/issues/146#issuecomment-258703196, or mute the thread https://github.com/notifications/unsubscribe-auth/AACjcgFlhLB9dpEuA8nnKD6NiyRNFybvks5q7iefgaJpZM4KqcEI .

the glass is all full: half water, half air. neurodata.io, jovo calendar https://calendar.google.com/calendar/embed?src=joshuav%40gmail.com&ctz=America/New_York

jovo commented 8 years ago

[x] "Unfortunately, the i.i.d. assumption does not hold for an undirected graph" i don't agree with that. you might be thinking of the adjacency matrix, but just think of the upper triangle.
[x] "Intuitively, at moderate t, the diffusion distance better reflects the connectivity since it takes into account every possible path between two nodes." i don't see why. is that always true? i doubt it.
[x] i think worth making the connection to the non-parametric anova test explicitly is worthwhile in III.1
[x] don't say "clearly", it might not be clear to a reviewer
[x] in task III.3, you refer to preliminary analysis, but don't show it?
[x] don't say "the most", say "one of the most", and remove "arguably". we don't want to argue with reviewers :)
[x] "gola" --> "goal"

jovo commented 8 years ago

[x] " every scientist having data. " not just scientists. also analysts. maybe others?
[x] do you know we have a "proper" website: http://docs.neurodata.io/MGC/#1. we could link to that, instead of github.com/neurodata/mgc/ and just update it a bit (even after submission).
[x] "Third, we plan to incorporate Mgc to our neurodata website (neurodata.io), providing web-service of Mgc toolbox to brain data and graphs, for processed data on our website and also user-uploaded data." make this more general. anybody will be able to upload (xi,yi) for i \in [n], and we will provide a url that provides plots such as the ones provided here. this further lowers the barrier to entry.
[x] "We will leverage exist- ing cloud resources to achieve this step: through the high-performance computing infrastructure of XSEDE (www.xsede.org), we deliver about 30,000 mapped brains per year; and our partner- ships with commercial cloud service providers allows a spatial database to store up to 10 petabytes in Amazon, as part of the MICrONS project from Intelligence Advanced Research Projects Activity." doesn't really make sense. we haven't talked about computations at all in this grant, not even computational complexity. nor how we could deal with 10 PB of data. i'd remove this.
[x] what are the "(V1)" things? "requested for Neurons SfN special issue (V1), on a Global Brain Initiative (V2), and the first to demonstrate science in the cloud (SIC) (V3)."
[x] i'd make the course more exciting. check out mauro's language on his course. specify high-dimensional noisy complex data, testing various settings, theory and algorithms, understanding geometry of data, things like that, manifold learning, connections to harmonic analysis, blah blah.

jovo commented 8 years ago

"action names" for subtasks is not done, so i unchecked it. it is important to provide a structure to ease the reader along.

cshen6 commented 8 years ago

I pushed current local changes to overleaf. The action names are there now.

Also, I copied the prior support from your version. So yes, what are the V1, V2, V3 thing? Cannot find it elsewhere on your doc.

jovo commented 8 years ago

i must have screwed that up, not sure, maybe they were references. all those papers are on my cv and/or in arxiv.

On Sun, Nov 6, 2016 at 9:16 PM, cshen6 notifications@github.com wrote:

I pushed current local changes to overleaf. The action names are there now.

Also, I copied the prior support from your version. So yes, what are the V1, V2, V3 thing? Cannot find it elsewhere on your doc.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/neurodata/MGC/issues/146#issuecomment-258732677, or mute the thread https://github.com/notifications/unsubscribe-auth/AACjcqEcvKvNWQ-BW8I8a1QHKQB4seoBks5q7on8gaJpZM4KqcEI .

the glass is all full: half water, half air. neurodata.io, jovo calendar https://calendar.google.com/calendar/embed?src=joshuav%40gmail.com&ctz=America/New_York

jovo commented 8 years ago

ok, action titles/sentences are better. but let's be much more concrete. NSF doesn't want to pay us to "investigate" or "explain/understand". this is a stats grant.

we will prove, develop, build, analyze, etc.

eg, Task I.1: Prove finite-sample MGC consistency under various transformations.

and

We will prove that multiple variants of MGC are consistent, and determine whether the conditions for consistency different upon differently centering or ranking the distance measurements.

Task I.2 and I.3 don't have a bold sentence about what you will prove/do, neither do any of the others?

On Sun, Nov 6, 2016 at 9:17 PM, joshua vogelstein jovo@jhu.edu wrote:

i must have screwed that up, not sure, maybe they were references. all those papers are on my cv and/or in arxiv.

On Sun, Nov 6, 2016 at 9:16 PM, cshen6 notifications@github.com wrote:

I pushed current local changes to overleaf. The action names are there now.

Also, I copied the prior support from your version. So yes, what are the V1, V2, V3 thing? Cannot find it elsewhere on your doc.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/neurodata/MGC/issues/146#issuecomment-258732677, or mute the thread https://github.com/notifications/unsubscribe-auth/AACjcqEcvKvNWQ-BW8I8a1QHKQB4seoBks5q7on8gaJpZM4KqcEI .

the glass is all full: half water, half air. neurodata.io, jovo calendar https://calendar.google.com/calendar/embed?src=joshuav%40gmail.com&ctz=America/New_York

the glass is all full: half water, half air. neurodata.io, jovo calendar https://calendar.google.com/calendar/embed?src=joshuav%40gmail.com&ctz=America/New_York

neurodata / MGC-paper

grant issues #146

super important

other stuff