Hello everyone, I am happy to share my work, I have to put it back online and I will post to the kaggle forum today. I was hoping to use nupic to preprocess raw eeg data to extract an anomaly signal and then apply classical approaches down stream. The computational complexity makes it impractical so I will have to preprocess the raw data in some meaningful way before feeding it to the HTM. Not quite sure what this would look like and I am happily taking suggestions.
Nicolas. On Thu, Oct 23, 2014 at 7:07 AM, Fergal Byrne <[email protected]> wrote: > Hi Guys, > > Thanks for posting that Matt, it was really informative, and thanks to Dr > Pantera for his kindness and patience with us! > > I've set up a NuPIC team for the second Kaggle competition, and invited > some of the people from the weekend. Anubhav has joined and is transferring > over his scripts today. > > If you'd like to join, please email me at this address with your Kaggle > email and github handle (I have a private repo set up again). If Nicolas is > agreeable, we could start submitting based on his work in the next day or > two.. > > Regards, > > Fergal Byrne > > On Thu, Oct 23, 2014 at 6:42 AM, Anubhav Chaturvedi < > [email protected]> wrote: > >> Hi >> >> Its great to see that NuPIC community is now involved actively in >> pursuing seizure detection and there is sufficient interest. Thank you Matt >> for organizing this talk in the Hackathon and hope you all found it >> intriguing too. >> >> I have been studying more on the subject and have also made submission >> trying out different techniques. So far I have reached a score of 0.7 but >> traditional approaches don't seem to be helping me anymore. Let me know if >> someone would like to discus more on this. >> >> I loved the talk and here are a few things that I found particularly >> interesting. >> >> It is known that artifacts like muscle or eye movement have been known to >> mess up the EEG recording and has been a problem with BCI as well. The fact >> that 80% of EEG recordings of epilepsy patients is normal was new. I did >> not know that and it is interesting to think how we can overcome this >> constraint. >> >> Another thing was the mention of Deep Brain Simulators(DBS). These are >> electrodes that are implanted to overcome seizure by providing electrical >> impulses. Now when I read about them earlier, there was a skepticism >> surrounding their efficiency. Sure they have been shown to be helpful but >> to what extent, it cannot be determined. This has mainly been because of >> poor seizure detectors. The problem is if you provide the impulse and you >> claim that the seizure has stopped because of it, you need to be 100% sure >> in the first step itself that a seizure was about to occur. And since the >> seizure did not occur, you cannot be sure if it was a false positive or if >> it was because of DBS. >> >> One more thing that I have noticed is that neurologists mainly observer >> scalp EEG signals unlike the intracranial signals provided in the >> competition. Physicians and surgeon, I have observed, have developed an >> intuitive understanding of when seizure has occurred or maybe is about to >> occur. This is what they have learned from their years of experience and >> the surrounding data like ECG, physical condition, etc. are also helpful. >> Because of this they find it very difficult to explain if you ask them just >> on the basis of EEG hoe to detect seizures. >> Also intracranial is only recorded when the patient is undergoing >> surgery and they already have spit open his brain. This has been a bottle >> neck because there is only a small duration of recording you can make in >> the first place and even fewer duration of seizure clips occur. This is >> evident in the competition data. >> >> *Regards,* >> *Anubhav Chaturvedi* >> >> *Birla Institute of Technology & Science, Pilani* >> KK Birla Goa Campus >> >> On Wed, Oct 22, 2014 at 9:44 AM, Matthew Taylor <[email protected]> wrote: >> >>> Hideaki, >>> >>> You'll probably really enjoy one of the hack demos I'll be publishing >>> soon. I won't spoil the surprise, but it involves EEG mind control. ;) >>> --------- >>> Matt Taylor >>> OS Community Flag-Bearer >>> Numenta >>> >>> >>> On Tue, Oct 21, 2014 at 8:39 PM, Hideaki Suzuki <[email protected]> >>> wrote: >>> > Hi Matt, >>> > >>> > Thank you for sharing the interesting video!! >>> > >>> > I also watched the hackathon raw, and I was very impressed that many >>> people >>> > were now interested in reading EEG by HTM. (in addition to those fun >>> staff >>> > like MindCraft ;-) >>> > >>> > I recently read the below article and would like to share. This is >>> also a >>> > good short article about EEG(actually ECG) and remote-controlling a >>> robot >>> > limb (IEEE spectrum), using the data collected from seizure patients. >>> > >>> > >>> http://online.qmags.com/IEEESM12819043?sessionID=BC9E010ABEF488AB56FC61EE2&cid=1010491&eid=19043#pg39&mode2 >>> > >>> > Regards, >>> > Hideaki Suzuki. >>> >>> >> > > > -- > > Fergal Byrne, Brenter IT > > http://inbits.com - Better Living through Thoughtful Technology > http://ie.linkedin.com/in/fergbyrne/ - https://github.com/fergalbyrne > > Founder of Clortex: HTM in Clojure - > https://github.com/nupic-community/clortex > > Author, Real Machine Intelligence with Clortex and NuPIC > Read for free or buy the book at https://leanpub.com/realsmartmachines > > Speaking on Clortex and HTM/CLA at euroClojure Krakow, June 2014: > http://euroclojure.com/2014/ > and at LambdaJam Chicago, July 2014: http://www.lambdajam.com > > e:[email protected] t:+353 83 4214179 > Join the quest for Machine Intelligence at http://numenta.org > Formerly of Adnet [email protected] http://www.adnet.ie >
