Blink Detection [DEMO #3]

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Erkannte Entitäten
we want to talk about things that interest me and all my all my theory
of hi I'm paul it's George both from simply don't interact strongly make these brain something EEG devices called news and then we can so we have the goal of the 2 identifier's links with an ECM as good as our own internal our the reuse the 5 points so the way energy device works is that it detects your brain signal as long as you're not moving in a few moves like like arises something that muscle movement overwhelms the brain signals and the can easily pick it out with your eyes so that was really a pretty easy test for an T restrained get up to speed I assume how it works and how they had a parameter that 1 so I used it as an EEG device also visible beta so it's like just a wave of data this 4 channels unless we decide to try to use a single channel which was the left ear and to try to build a detector so 1st of all you the idea of what the typical data looks like that the blue line here would be your E and then the red line is our own algorithm that we use to detect planes so you can see a link here goes down and goes up and goes back to normal and that's so you can visually of length and we're hoping to use the user existing algorithms to train the new edition of that so you see all these all these red spikes here our existing our them detecting a blank the 1st and then to much to show itself what's done very so we're trying the biggest problem we had was to try to get the data small enough so that we can actually process the the data in time to do something useful that's really kept shrinking down the input data because samples up to about 220 hertz per 2nd and so we we downsampled to average over 10 samples and this is the 1 on the on the left side is the in predicting whether you blanked and on the right side is existing algorithms photos from target resource said so this is this year and this is our existing so that seems to be doing pretty well it's only mispredicted occasionally services services so here we have some restrictions for most the time it seems to be working pretty well yeah so that's what we are human having and not be questions that belong the you know the predicted that wall and that's what that but you know when you have 3 predictions of Lincoln Rao help are part of separated time out you know that's that's a single point right so this is a very interesting in this case the cell loneliness a compilation a couple but there a link that didn't really this is actually the only time there's only 1 instance where a completely went wrong results of you know there is 1 instance where must pretend that how you making a prediction with the chance I mean he you build a temple molybdate about what what are you doing to what exactly are you going to make prediction clearly using our prediction what we have here this is a classification that is it's not really a prediction is 1 and they are that OK so here we talked about their internal and we talked about whether to use prediction anomaly detection or what sense is a binary volume like you pointed enabling the time can fit into prediction of this you based on the EEG data you predict this can have a way to classify the lights already don't have to send writing the letter I guess
you have labelled data for what exactly what is the file of like that you're feeding into the Asian of region prediction what's
like this so we voltages and then whether wake up and and and both of the simple both of those are being fed into the HCM and it's predicting the next time step user query OK so the right column is also being fed into the HTML the right column this was supposed to be using to predict the next sorry the left was used to predict the right column the evidence and like it's set up that way this is the literary reduce our output was assured to produce some of the word we think that it is you don't have existing code sample that does that the and curious how you set it up that way we load talk so what you would not have we cannot offline that will occur in the you know last to predict forward in time in the identity function so that you can use some of that desire
different interpretation of what this is showing us from minus and and so here we haven't can speak into the microphone here we have a kind value that the system is seen and here we have our potential predicted value so and this is the blind information that it's seeing and the right column it's what is predicting to see so right now sees 1 next 1 and produces the 1 as well and shares his 1 predicts that there will be no blank from that yes there is no thank you hang around for a
long time I need to
Quelle <Physik>
Physikalische Theorie
Demo <Programm>
Interaktives Fernsehen
Zellularer Automat
Digitale Photographie
Spezifisches Volumen
Güte der Anpassung
Einfache Genauigkeit
Gleitendes Mittel
Binder <Informatik>
Dienst <Informatik>
Rechter Winkel
Lineares Funktional
Wort <Informatik>
Funktion <Mathematik>
Quelle <Physik>
Gemeinsamer Speicher
Physikalisches System
Demo <Programm>


Formale Metadaten

Titel Blink Detection [DEMO #3]
Serientitel 2015 Spring NuPIC Hackathon
Anzahl der Teile 19
Autor Baronowski, Paul
Bit-Yunan, George
Lizenz CC-Namensnennung 3.0 Unported:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen.
DOI 10.5446/18048
Herausgeber Numenta Platform for Intelligent Computing (NuPIC)
Erscheinungsjahr 2015
Sprache Englisch

Technische Metadaten

Dauer 07:59

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract Using NuPIC to detect blinks in Muse data.

Ähnliche Filme