Dear Organizers, I am wondering, have you already tried to average the scores predicted by different teams? It is very easy and fast to try (if output files of all models were saved) and would give a very first baseline for the Community method. Also, by averaging teams by pairs one could get a feeling about which methods are "orthogonal" and improve the results and which don't - useful information for a future fused model. Next we could discuss it during the Webinar. Best, Yaroslav

Created by Yaroslav Nikulin (Therapixel) ynikulin
Hi Yuanfang, The teams will be invited to briefly describe their approach and ideas for improvement early in the collaborative phase. In next week meeting, we will describe the structure and goals of the competitive phase: timeline, meetings, target performance, incentives, datasets, insights into the results from the competitive phase, IT infrastructure, etc. We will also give suggestions on different aspects that may help to further improve the inference methods. > will there be a discussion of transferability and further improvement work of the most versatile/flexible models (e.g. the two or three of models that do not rely on training/color mapping in DREAM data)? Yes, and if there is no time left, this will be discussed in the second meeting. > is it allowed/encouraged to use more instant communication avenue among people before and after the meeting. Everyone will express their preferences but we definitively encourage participants to communicate among themselves and with organizers as much as possible. Thomas
hi, thomas,   I am confused about the agenda about the meeting. Will the top team (therapixel) give a presentation of their approach + Q and A (that could be very useful for everyone), or it is only you presenting for 1.5 hours and give out some assignments?   will there be a discussion of transferability and further improvement work of the most versatile/flexible models (e.g. the two or three of models that do not rely on training/color mapping in DREAM data)? it would be a killer app to deliver a model that can work across imaging technology (film/digital) and cohorts.   is it allowed/encouraged to use more instant communication avenue among people before and after the meeting: e.g. Skype beyond asking (and waiting/searching for answers) every tiny technical question on the forum? related to this i guess is the nature of the community phase that it is still some how competitive or completely collaborative in nature?   thank you   yuanfang
Perfect, thank you for fast reply. Looking forward to discussing.
Hi Yaroslav, > I am wondering, have you already tried to average the scores predicted by different teams? It is very easy and fast to try (if output files of all models were saved) and would give a very first baseline for the Community method. Yes, I have already done that. I'll show the results in next week's meeting. Thanks for suggesting! Thomas

Community Phase Baseline - averaging the predictions ? page is loading…