|1||Jun 29, 2009 7:31 PM||It should continue, but I don't think the freiburg dataset is good for this purpose|
|2||Jun 29, 2009 7:50 PM||Why aren't any of the "old" groups joining? After this year, nobody will want to do it anymore.|
|3||Jun 29, 2009 7:51 PM||It is very valuable because it reflects the current inability of the people in the field to predict seizures. This is important given all the flawed and over-optimistic publications.|
|4||Jun 30, 2009 12:53 AM||The "detection" component of the contest should be emphasized.|
|5||Jun 30, 2009 6:19 AM||It was interesting to realize, for one more time, that we are still quite far away from the goal of seizure prediction. congratulations on the people who actually submitted their results.|
|6||Jun 30, 2009 8:44 AM||At the moment the results are far from encouraging, but the reason for the meeting is seizure prediction so I think we have to keep trying. When someone hits the jackpot we will know we are getting somewhere.|
|7||Jun 30, 2009 9:21 AM||Would be very good to have more types of signals, i.e. very clean signal with no artifacts and clear generalized seizure on all channels. Then with some artifacts, seizure on subset of channels, etc.
Would be also nice to know why the algorithm failed. Is that because signals are too complicated, limited training datasets, etc.|
|8||Jun 30, 2009 9:29 AM||There should be a way to find out how more people could attend that competition.
It might be interesting to talk to all the groups out there to see what they like to be changed before they join in on the competition.|
|9||Jun 30, 2009 2:07 PM||We must know why the number of entries is so small, given that the contest was open to detection algorithms as well. Are there questions about the quality or value of the data set?|
|10||Jun 30, 2009 5:28 PM||No harm to continue, although the seizure prediction is not matured yet.|
|11||Jun 30, 2009 5:37 PM||No harm to continue, although the seizure prediction is not matured yet.|
|12||Jul 1, 2009 5:44 PM||No.|
|13||Jul 2, 2009 1:43 AM||I was disappointed that the major labs and industrial representatives do not participate. This seems counterproductive to the meeting's guiding principle. Current state-of-the-art benchmarking would be useful (knowing what is possible is useful even if I don't know how it is done). If labs won't compete in the competition because of connections to industry and/or intellectual property issues then perhaps the contest should be changed such that results can be posted anonymously without revealing code and origin of the results.|
|14||Jul 2, 2009 6:35 PM||I think patient 2's data set may have had the power spectrum changed at the end. I have contacted the competition's organizers about this. We may not use it in the future.
I know there is consideration that we charge people to participate in the competition so that we ensure submissions after people download the data. I think the charge should be nominal. I think $1000 would be prohibitive except for labs that are very serious about seizure prediction, which would exclude new labs who may want to try new ideas but are not seizure prediction labs, per-se.|
|15||Jul 2, 2009 8:09 PM||I think that it'd be better to have the BBQ party on Saturday. I saw many of people in the conference tired on Saturday.|
|16||Jul 4, 2009 10:36 PM||The format of presenting results is not clear from the website.|
|17||Jul 6, 2009 9:47 AM||people who download EEG data and agree to participate in the contest should present their findings even if algorithm(s) doesnt work. even negative findings are helpful to the community; could help avoiding making the same mistakes again and again.|
|18||Jul 6, 2009 8:34 PM||No comment.|
|19||Jul 7, 2009 1:04 AM||no|
|20||Jul 8, 2009 12:28 PM||I think it should be required. It is frustrating to hear talk after talk from people claiming they've solved the problem (and training their algorithms on their own in-house data to make the point). We can all do this. The trick is to have an algorithm that works on new data. That's what the contest is for. 50+ folks downloaded the data, and only 3 entered the contest (of which only 2 showed up). This tells me most people don't want to be independently evaluated. One of the best things about this workshop is how rigorous it is and how critically it shines the light of truth on everyone's work. As part of this I'd make submitting an algorithm to the contest a requirement for presenting on that algorithm.|
|21||Jul 8, 2009 7:50 PM||No|
|22||Jul 9, 2009 8:50 AM||Of course a pity that only few partisipated. More emphasis could have been put on the results (I didn't hear the result as I was discussing my poster with other partisipants). E.g. the results/price could have been presented during the day sessions about "State of seizure prediction". A lot of people may also have been missing as it was during the evening.
In principle I think the contest is an easy way to get students into the subject of seizure prediction.|
|23||Jul 10, 2009 5:36 AM||None at the moment. I am glad to be a part of this field as a general electrical engineering consultant.|
|24||Jul 10, 2009 9:08 PM||It helps to focus, define, and weed out practical problems.|
|25||Jul 12, 2009 5:38 PM||I believe that the data should be update, because the current technology allows us to collect data with more quality. This challenge should be notified to people with experience in other fields of data series prediction. Moreover, perhaps the existence of a prize will increase the number of interested people as already happens in another machine learning contests (e.g. http://www.incf.org/community/competitions/spike-time-prediction/2009).|
|26||Jul 22, 2009 2:05 PM||Great initiative|
|27||Jul 24, 2009 4:55 PM||I feel the organizers should have led the way in submitting their algorithms for evaluation as well. Personally, i think any algorithm is as good as the data is. The prediction/detection contest would be more interesting if more types of data were available. (Eg, depth electrode data sampled at higher frequencies) With noisy, artifact filled data, any algorithm would fail. The bottleneck in prediction/detection is just as much data acquisition as it is in building these complicated mathematical models.|
|28||Jul 24, 2009 6:04 PM||The members of Advisory Committee representing groups which supposedly are working on prediction and published in this area should submit their algorithms for the contest.
Any talk or poster submission claiming detection or prediction should be required to submit it to contest.|
|29||Jul 24, 2009 7:00 PM||NO|
|30||Jul 25, 2009 8:35 AM||Loosen the frame so that more can contribute. Let people suggest evaluation criteria that the organizers can apply on the test data|
|31||Jul 27, 2009 6:24 AM||besides providing open dataset, use unknown dataset (not published) to test the seizure detection algorithm.|
|32||Jul 28, 2009 11:10 PM||I think a person or a group's choice to NOT participate in the group is almost as revealing as a group's actual participation. Thus, even if participation is low, the act of not participating is still an immensely valuable data point.|
|33||Aug 5, 2009 3:32 PM||Would like to see more participation incentivised - perhaps with a prize.|