we wanted to ask a few questions about this project 2 redition :
First , we derived an Expectation-Maxmimization algorithm based on an unpublished method from the professor we are working with. As it is unpublished, we cannot reference it. But it is too long to be included in the report. Is it possible to put an appendix with a pdf of this unpublished method ? (It is basically a mathematical formulation of a maximum likelihood estimation).
Second, in the first part of the project we are dealing with images and we trained a deep neural network with a published library. Do we need to provide the whole training set, the trained models , etc.. (which are multiple Gb of size?)
The training set had to be built manually to some extent so there is no script linked with it, we cannot post anything else than the actual images themselves. What do you suggest us to give you back ?
We decided to choose a project that was a bit ambitious, there is multiple big parts, and we fear not to have finished everything for next week. What do you suggest to do if some part is missing ? Could we run a simulation to replace the missing data ?
Thanks a lot for the support ! Enjoy the week-end !
EM new variant: yes you can explain it in an appendix, and you can also cite it in addition (no matter if it's published or not)
the whole pipeline should be as reproducible as possible. at least provide scripts so that the training data could potentially be obtained if needed. you can also provide intermediate models if you have several steps, if people want to see if they can reproduce some part, if the entire pipeline is too costly to reproduce
provide the training set (images and labels), you can use a private link such as on gdrive/epfl, in case your hosting lab does not want the whole training set to become public
yes you could mention the work in each part and explain additional work needed, and as you say try to bridge parts if they are not ready yet
Project 2 - few questions
Hi !
we wanted to ask a few questions about this project 2 redition :
First , we derived an Expectation-Maxmimization algorithm based on an unpublished method from the professor we are working with. As it is unpublished, we cannot reference it. But it is too long to be included in the report. Is it possible to put an appendix with a pdf of this unpublished method ? (It is basically a mathematical formulation of a maximum likelihood estimation).
Second, in the first part of the project we are dealing with images and we trained a deep neural network with a published library. Do we need to provide the whole training set, the trained models , etc.. (which are multiple Gb of size?)
The training set had to be built manually to some extent so there is no script linked with it, we cannot post anything else than the actual images themselves. What do you suggest us to give you back ?
We decided to choose a project that was a bit ambitious, there is multiple big parts, and we fear not to have finished everything for next week. What do you suggest to do if some part is missing ? Could we run a simulation to replace the missing data ?
Thanks a lot for the support ! Enjoy the week-end !
EM new variant: yes you can explain it in an appendix, and you can also cite it in addition (no matter if it's published or not)
the whole pipeline should be as reproducible as possible. at least provide scripts so that the training data could potentially be obtained if needed. you can also provide intermediate models if you have several steps, if people want to see if they can reproduce some part, if the entire pipeline is too costly to reproduce
provide the training set (images and labels), you can use a private link such as on gdrive/epfl, in case your hosting lab does not want the whole training set to become public
yes you could mention the work in each part and explain additional work needed, and as you say try to bridge parts if they are not ready yet
hope this helps a bit
1
Add comment